Apple filed a federal lawsuit against OpenAI on Friday, accusing the AI company and its chief hardware officer of stealing its trade secrets.

"OpenAI and its cohorts, led at least in part by former Apple employees, have recruited candidates from Apple, extracted their knowledge of Apple’s sensitive and confidential information, and then continued to exploit that knowledge once they arrived," the complaint reads. "As a result, OpenAI has misappropriated Apple’s trade secrets and confidential information in a variety of ways."

The suit, filed in the Northern District of California, names OpenAI technical staff member Chang Liu, chief hardware officer Tang Tan, OpenAI, and io Products as defendants. The last of that group is notable because it was founded by Tan in collaboration with former Apple design head Jony Ive, Evans Hankey (Ive's successor at Apple), and former Apple designer Scott Cannon. Notably, the complaint seems to attempt to avoid naming the founders, though Ive's name is cited in a URL.

Tan previously served as a vice president of product design at Apple, working on the iPhone, AirPods, and Apple Watch. Liu served at Apple as a senior electrical engineer.

In the complaint, Apple alleges that it reached out to OpenAI in February with concerns, but that OpenAI did not respond. Apple claims that Tan attempted to gain secrets from Apple employees, including asking prospective job candidates to bring components for "show and tell" sessions and used his knowledge of the company to squeeze more information out of candidates. The suit claims that Liu never returned a company laptop, and used an authentication bug to access Apple files.

Apple also claims that OpenAI told incoming employees how to leave their former job, suggesting they stay as long as possible and not disclose their former employer in order to continue to access confidential information.

"At every level, from members of its Technical Staff to its Chief Hardware Officer, and in coordination with business partners, OpenAI has been stealing Apple’s trade secrets and confidential information," the suit reads. "As a natural result, OpenAI’s nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misappropriated trade secrets."

OpenAI did not immediately respond to a request for comment from Tom's Hardware. Apple's lawsuit claims that over 400 former Apple employees currently work at OpenAI.

Apple is rumored to be working on a number of AI-powered hardware projects, including AirPods with cameras, a pendant, and home robots. It's less clear what hardware OpenAI may be working on, though The Information suggested the company has a HomePod-style smart speaker in the works.

Apple is requesting a jury trial, damages, attorney fees, and orders that the OpenAI may not use Apple's trade secrets, among other injunctions.

In May, Bloomberg reported that OpenAI was considering legal action against Apple because it expected deeper integration and more users from ChatGPT features built into iOS.

If the trial does go to court, it's sure to be a dramatic one, potentially dragging several former high-level Apple employees into testimony through discovery and testimony.The trial, Apple Inc. v. Liu et al, is case 5:26-cv-07078 in the United States District Court in Southern California.

OpenAI has discussed handing the U.S. government a 5% ownership stake, the Financial Times has reported, citing two people familiar with the talks, with CEO Sam Altman proposing that every leading U.S. AI developer contribute the same share of its equity to a vehicle modeled on the Alaska Permanent Fund, which pays annual dividends to state residents from Alaska’s oil wealth. At the $852 billion valuation OpenAI set in its March funding round, a 5% stake is worth roughly $42.6 billion. The FT characterized the discussions as conceptual and early-stage, and reported that implementing any deal might require an act of Congress.

Altman is understood to have raised the idea with President Donald Trump, Commerce Secretary Howard Lutnick, and Treasury Secretary Scott Bessent, and has spoken with Senator Bernie Sanders (I-Vt.) in recent weeks. The all-labs structure would pull equity from companies, including Google, Meta, and Anthropic, none of which have indicated they would participate. OpenAI declined to comment to the FT, and the White House didn’t immediately respond.

Altman’s 5% is the smallest figure we’ve seen to date attached to public ownership of the AI sector. Sanders filed the American AI Sovereign Wealth Fund Act in June, seeking 50% of the voting shares of U.S. AI companies through a fund his office valued at $7 trillion, enough to pay every American a $1,000 annual dividend. Trump said last month that he was exploring options to give the public a stake in leading AI firms, and Vice President JD Vance said the president prefers equity over cash payouts.

The administration has already run this playbook on chipmakers, with the federal government having taken a 9.9% stake in Intel last August by converting CHIPS Act grants into equity at $20.47 per share, and AMD and Nvidia agreed to hand over 15% of their China chip revenue in exchange for export licenses. OpenAI itself proposed a “public wealth fund” in an April policy paper, and Altman first pitched a government stake to the administration in early 2025, CNBC reported last month.

News of the talks comes just six days after OpenAI delayed the full public launch of GPT-5.6 at the government’s request, with Lutnick reportedly warning Altman against releasing the model without prior approval. Anthropic spent most of June with its Claude Fable 5 and Mythos 5 models disabled worldwide under the first U.S. export controls ever applied to an AI model rather than to hardware; access was restored yesterday.

Both OpenAI and Anthropic have confidentially filed for initial public offerings, and OpenAI faces a probe from a coalition of 42 state attorneys general. A government shareholding negotiated before a listing would lock in Washington’s position ahead of the ownership expansion that a full float brings, however.

OpenAI CEO Sam Altman said during a staff Q&A meeting that its latest model, GPT-5.6, is available in limited preview to only a small group of customers handpicked by the U.S. government. According to The Information, the federal government, specifically the Office of the National Cyber Director and the Office of Science and Technology Policy, asked the AI tech company to stagger the release of its latest model. While Altman did not mention how long the delay for the general release of GPT-5.6 will be, he said in a memo that he hoped it would happen in a couple of weeks. In the meantime, the U.S. government is granting access to the latest model on a case-by-case basis only.

Despite OpenAI’s agreement to the delay, sources say that Commerce Secretary Howard Lutnick called Altman to warn him against releasing GPT-5.6 to the public without prior approval from government agencies. “We’ve made clear to the U.S. government that this is not our preferred long-term model and will work with them and others in industry to achieve a more sustainable approach for future releases,” the OpenAI chief said in the Thursday memo.

This wasn’t the first time that an American AI lab has delayed the release of its frontier model due to security concerns. Back in early April, Anthropic released Claude Mythos Preview to select key institutions first, allowing them to prepare for the general release of the powerful AI model. It eventually built Fable 5, a watered-down version of Mythos with built-in safeguards to prevent misuse, and released it in June 2026. However, the U.S. government disagreed with the company’s belief that it was a safer model and put both Fable 5 and Mythos on an export control list just three days after it dropped. This meant that foreign nationals, even those who work for Anthropic, are banned from accessing the model. Since the company cannot enforce compliance, it just decided to pull the model completely from the market.

The increasing advancement of AI models has the White House scrambling to prevent them from falling into the wrong hands. This is especially true as it continues to compete with rival China for supremacy. Although the U.S. has taken steps like export controls to slow Beijing’s progress, many industry leaders believe that it’s only a matter of time before the East Asian country catches up. So, even though the Trump administration initially promised that it would reduce regulations to help AI advance much more quickly in the country, President Donald Trump has changed his tune and signed an executive order earlier this month that asks U.S. AI labs to give the government access to their latest models 30 days before it gets a general release.

However, this move has got some industry experts concerned. “…this escalation of government intervention is nothing to celebrate. It is horrible for the broader AI ecosystem,” Head of AI Policy and think tank Abundance Institute and former FTC Chief Technologist Neil Chilson said in their blog. “Continued arbitrary, unexplained deployment of export control authority will make companies slow-walk new models, depriving the public of powerful new tools. Every AI model, like all software before it, will have vulnerabilities that require patching. The US government should not hang a Sword of Damocles over every lab’s head, with no indication when it might drop or why.”

OpenAI and Broadcom have introduced Jalapeño, a custom-built inference processor designed specifically for modern large language models and future agentic AI workloads, which is designed to deliver performance per watt they claim is higher than today's leading-edge hardware. OpenAI considers its hardware project a strategic one and envisions Jalapeño to be the first generation of its inference hardware.

Not another AI accelerator

OpenAI stresses that Jalapeño is a purpose-built inference ASIC and not a repurposed training accelerator or a general-purpose AI processor. OpenAI says the architecture of Jalapeño was designed based on its understanding of LLM behavior and is meant to address practical bottlenecks that matter for inference at scale, including costly data movement, balance between compute and memory resources, networking efficiency, and overall behavior. OpenAI also states that the design of the processor is meant to wed high throughput with low latency (which is why it uses a huge compute chiplet and HBM memory and not cheaper types of DRAM like many other inference accelerators), which will be particularly handy for reasoning and agentic workloads.

In addition, OpenAI and Broadcom claim the processor is built to deliver higher effective utilization than conventional AI accelerators and deliver performance that is close to the theoretical maximum, which means very high efficiency both in terms of costs and in terms of power. Meanwhile, the companies did not disclose performance targets for their Jalapeño ASIC, so these claims should be taken with a grain of salt.

Engineering samples are already operating in the lab at target clock speed and power (though Broadcom and OpenAI do not disclose details about this, either), and OpenAI says it is running machine learning workloads, such as GPT-5.3-Codex-Spark.

The two companies also claim that early internal testing indicates that Jalapeño's performance-per-watt is substantially better than 'current state-of-the-art hardware,' although no hard numbers, benchmarks, memory configuration, or other details are disclosed, so again, we will have to take the claims with a grain of salt. In addition, one must bear in mind that while Jalapeño can purportedly beat existing AMD's Instinct MI350-series and Nvidia's Blackwell-based accelerators, it remains to be seen how competitive it will be against AMD's Instinct MI400-series and Nvidia's Rubin-based offerings.

"Jalapeño was designed from the ground up for LLM inference using detailed insights from our close collaboration with OpenAI researchers," said Richard Ho, who leads OpenAI's hardware program. "We optimized the architecture around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models. Based on early testing, Jalapeño will efficiently execute our most important workloads close to the hardware’s theoretical limits."

A massive chip with six HBM modules

While Broadcom and OpenAI did not disclose specifications of Jalapeño, they did show its wafer and packaging, so we can do a brief analysis. The package appears to contain one large compute chiplet surrounded by six HBM modules and another chiplet that likely packs input/output interfaces and is surrounded by two structural dummy dies.

The wafer image does look like a Broadcom-style systolic-array-heavy accelerator, in the sense that it shows a very regular, repeated, columnar floorplan with what looks like replicated compute regions and fixed infrastructure macros. Yet, keep in mind that we are speculating, and the image is not clean enough to say that this is definitely Broadcom's standard TPU-like systolic array template with some perks from OpenAI,

From the image alone, it is impossible to tell whether Jalapeño uses a true 2D systolic array, a set of 1D/2D matrix engines, a collection of vector or tensor tiles, or some other inference datapath. All we can say is that the die has a highly repetitive floorplan consistent with several kinds of tiled AI accelerator architectures.

What we can tell from the image is the approximate die size of Jalapeño's compute chiplet based on the size of HBM3/4 packages (10.975 mm × 10.975 mm) that surround it. From what we can tell, the chiplet measures 25.46 mm (width) × 33 mm (height), which means that its die size is around 840 mm², which is very close to the reticle size of EUV lithography systems (858 mm²). Given that the quality of the shot is poor, the die size we estimate cannot be 100% accurate, but we suspect it is close enough.

The die size of Jalapeño's compute chiplet implies that it packs quite a lot of compute oomph, though, of course, we cannot make performance estimates based on this metric. Yet, it is safe to say that Jalapeño's compute die is considerably bigger than compute dies of other inference accelerators on the market and more resembles processors for AI training. Speaking of processors for AI training, we increasingly see multi-chiplet designs for these workloads as companies like AMD and Nvidia want to pack as much performance as possible. Meanwhile, the fact that OpenAI and Broadcom chose to go with a large compute chiplet possibly indicates that they wanted to reduce latencies by as much as possible.

Designed in nine months

The companies say the chip reached tape-out in just nine months and is slated for deployment beginning in late 2026, which represents an extremely fast turnaround time in ASIC design. It is unclear whether Broadcom and OpenAI extensively used artificial intelligence to define and then develop Jalapeño, though the companies admitted that they used OpenAI's models to speed up parts of the chip's design and optimization work. Typically, it takes 1.5 – 2 years to design an ASIC from scratch, so AI can shrink the development cycle. Another means to accelerate the design cycle is Broadcom's extensive reuse of its logic across different custom designs to deliver new chips faster than other companies.

It is noteworthy that, according to the announcement, Jalapeño is designed to support not only OpenAI's own workloads but also present and future LLMs across the industry, which potentially lets OpenAI sell its hardware to third parties, assuming that it can get enough supply from Broadcom and TSMC. Meanwhile, the chief executive of Broadcom indicates that Jalapeño will be deployed at gigawatt-scale data centers with Microsoft and other partners starting this year, though it is unclear whether the processor will be used exclusively for OpenAI workloads or will be available for other tenants as well.

"Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI," said Hock Tan, President and CEO, Broadcom. "This is just the beginning of a multi-generation roadmap. By co-developing our industry-leading silicon directly with OpenAI, we are enabling the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026."

A coalition of US state attorneys general has launched a sweeping investigation into OpenAI. According to a Wall Street Journal report, OpenAI was served on June 12 with a broad subpoena spearheaded by New York Attorney General Letitia James. The subpoena seeks documents related to a wide range of the company’s activities and their potential impact on users, including OpenAI’s advertising practices, user engagement and retention strategies, handling of consumer and health data, activities involving minors and seniors, use of deep learning models, model sycophancy, and internal company policies.

In a statement following the subpoena, an OpenAI spokesperson said, “AI is a new and powerful technology, and we work every day to safely bring its benefits to people in a responsible way. We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices.”

The investigation comes just five days after OpenAI revealed it had confidentially filed paperwork with the Securities and Exchange Commission, seeking to go public via an IPO that will reportedly value the company at up to $1 trillion. While the subpoena appears to be an information-gathering step rather than a formal accusation of wrongdoing, its breadth suggests state regulators are examining both OpenAI’s business practices and the safety risks associated with increasingly human-like AI systems.

The company is already facing real legal troubles elsewhere. Earlier this month, Florida officially sued OpenAI and CEO Sam Altman, following a criminal inquiry launched in April 2026. The civil lawsuit, filed by Florida Attorney General James Uthmeier on June 1, accuses OpenAI of knowingly releasing and aggressively marketing ChatGPT to the public, including children, while allegedly concealing serious risks, suppressing internal safety warnings, and misleading users about the product’s dangers. Florida’s complaint claims the chatbot can facilitate harm, including self-harm and violence, while also alleging that OpenAI collects data from minors without meaningful parental oversight and has downplayed the risk of dangerous errors.

In addition to these concerns, the recent subpoena focuses on OpenAI’s handling of consumer and health-related data, a key issue given that users often share sensitive personal information with AI chatbots. Unlike traditional search engines, conversational AI systems can invite users to disclose medical concerns, emotional distress, financial details, family problems, or other private information during ordinary use.

The subpoena reflects a broader reckoning over a technology that has scaled faster than the legal frameworks meant to govern it. For now, the investigation is an information-gathering exercise rather than a finding of wrongdoing, and OpenAI has said it takes the attorneys general's concerns seriously and will cooperate with the investigation.

The cost of serving AI via a subscription model has steadily increased for AI firms, especially as the decrease in cost per token has not kept pace with the spike in token usage. According to SemiAnalysis, the subscriptions that both Anthropic and OpenAI offer are much cheaper than the actual cost you have to pay if you maximize their usage. The research firm purchased every subscription from the two AI providers and discovered that the approximate maximum possible spend (assuming API pricing) is far larger than what users pay every month. For example, Claude Max 20x costs $200 a month, but maximizing it would cost $8,000 a month in token spend, while ChatGPT Pro 20x, which is also $200 monthly, has a maximum possible spend of around $14,000.

Anthropic breaks even on its two lower plans (Claude Pro and Claude Max 5x) at 20% utilization, while OpenAI starts losing money if utilization on its base plans (ChatGPT Plus and ChatGPT Pro 5x) exceed 11.4%. Things are much worse for the two companies’ top-end offerings, with Anthropic hitting 0% gross margin if utilization reaches 10%, while OpenAI is in the red if usage exceeds 5.7%. This is certainly unsustainable, but cutting features or raising subscription prices is likely off the table for these companies as well.

It’s not all bad news, though — as new models arrive and more data centers go online, the cost of serving existing models is bound to decrease, with SemiAnalysis predicting that serving Opus 4.8-level models at $20 a month could become profitable soon. On the other hand, frontier models, like Mythos, will still be much more expensive to run, so it’s likely that the latest, most advanced features could be reserved for API access only, meaning you’ll need to pay for it on a per-token basis.

Expensive frontier models have firms looking elsewhere

As SemiAnalysis showed, subscription tiers are more affordable than API access. However, you’d still need the latter if you want to access the full capabilities of these AI models, and this is where budgets start breaking. Powerful agentic AI uses up to a thousand times more tokens than the average model, and big firms like Microsoft, Meta, and Amazon are backing off “tokenmaxxing” as costs spiral out of control. One unnamed company even blew through $500 million in one month after failing to impose a usage limit on its employee licenses.

Because of this, some firms have started using tools that switch these expensive frontier models for cheaper, more affordable ones, including Chinese open-source models like DeepSeek. A Wall Street Journal report says costs could be reduced by up to 95% by allowing agents to switch between AI models as needed. “You don’t need a model that knows quantum gravity,” Columbia University vice dean Vishal Misra told the publication. “These open-source models are very capable, and the ability to charge a big premium for AI is going to diminish.”

Flo Crivello, the founder of Lindy, a startup providing AI executive assistant services, also told WSJ that the company has moved towards DeepSeek V4, as it proved to be as capable as Sonnet while costing ten times less. Although it still reserves Anthropic’s models for advanced work like coding, Crivello said that using the cheaper model has “saved the company millions of dollars.”

Other firms have begun building their own AI using open-source models, which are tailored to their specific needs and trained on in-house data. While this might seem complicated and expensive at first, it could save the company in the long run, as it would not have to rely on third-party providers for its AI needs. Some even claim it could outperform frontier models, as they’re built for the firm's specific needs and applications.

The availability of cheaper models and AI agents that optimize operational costs by using the more expensive options only as needed is putting pressure on OpenAI and Anthropic to lower their prices. OpenAI CEO Sam Altman has talked about the issue of ballooning AI token costs and said the company is looking for ways to help users “get more value for less spend” when using ChatGPT.

OpenAI says it has banned two clusters of ChatGPT accounts it believes are operating from China, which used its models for covert influence campaigns targeting U.S. tech and policy debates, including one called “Data Center Bandwagon,” that produced social media comments and comic strips blaming AI data centers for rising household electricity bills. Whoever was operating the accounts prompted ChatGPT in Simplified Chinese via VPNs, and posed on X as Americans from a range of backgrounds. Yet OpenAI’s full threat report found that the activity generated virtually no authentic engagement.

A second cluster, "Tech and Tariffs," generated anti-tariff cartoons under instructions to depict President Trump but never Chinese leader Xi Jinping, and was linked to a network of fake X accounts spreading false claims that ChatGPT user data had been stolen.

OpenAI assessed that the Data Center Bandwagon operators were likely a social media team at a private Chinese tech company working for provincial-level government clients. Among other requests, they asked ChatGPT for comic strips about a grid operator's capacity auction prices, drawing on a regional newspaper's reporting, then posted the output on X under hashtags such as #capacityauction, alongside links to legitimate news coverage.

While no grid operator was named, capacity auction pricing links back to a real, well-documented dispute: PJM Interconnection's independent market monitor has blamed data centers for an "irreversible" 75.5% increase in power costs across the largest U.S. grid region, with wholesale prices near some data center clusters having climbed as much as 267% in five years. Three U.S. senators have demanded answers from Amazon, Google, and Meta over costs passed to residential customers.

"This was not a case of an influence operation creating a debate," Ben Nimmo, principal investigator at OpenAI, said. OpenAI rated the activity Category One on the Breakout Scale, meaning it stayed on one platform, with no evidence that it reached genuine audiences.

The Tech and Tariffs cluster poked fun at U.S.-China competition around tariffs, rare earths, AI, 5G, and industrial resilience, and generated bulk comment batches in English, Italian, Japanese, and Traditional Chinese, the latter aimed at audiences in Taiwan. One operator described the accounts as a "water army," a Chinese term for coordinated troll networks, and asked ChatGPT to design a system for scraping and analyzing social media posts from individuals flagged as risks. OpenAI said its model returned generic data storage advice and declined to help with collection. Fake accounts in the same X network repeatedly posted fabricated claims that ChatGPT user data had been compromised, which OpenAI reckons is an attempt to damage its own reputation.

In its full threat report, OpenAI compared the campaigns to the 2022 Spamouflage operation, which researchers at ASPI and Mandiant found targeting Lynas Rare Earths, Appia, and USA Rare Earth after Beijing's 14th Five-Year Plan prioritized rare earths. The new activity followed the adoption of the 15th Five-Year Plan recommendations, elevating AI as a strategic industry for China.

OpenAI CEO Sam Altman has said in an interview that companies are now concerned about the growing costs of AI use. Speaking during the Intelligence at Work event, he said this is the first time that OpenAI’s clients raised the issue and that the startup is now looking for ways to make its models more efficient.

“People are really saying, you know, it’s kind of a meme now, but ‘My company spent my entire 2026 budget in Q1. Can you make this more efficient?’” Altman said on stage. “We are continuing to push on that more with models. I think we’ll have a lot of ways we can help people get more value for less spend. But that went from, at the beginning of this year, an issue that never came up (people were totally happy with the amount they were spending) to, all of a sudden, a huge issue.”

There have recently been a lot of stories of companies getting massive AI bills as they experiment with “tokenmaxxing.” A few company leaders believed that AI use would increase the productivity of their workers, thus increasing revenue. Nvidia CEO Jensen Huang famously said that his engineers should use AI tokens that are worth at least half their annual salary, or else he'd be “deeply alarmed.” We also saw another example with OpenClaw creator Peter Steinberger, whose team spent $1.3 million on OpenAI API tokens in a month, totaling 603 billion tokens.

However, it seems that this move is starting to backfire on some companies. Amazon employees admitted that they were using AI agents for unnecessary tasks just to stay on the internal AI leaderboard, while Microsoft has reportedly cut back on Claude Code licenses due to increased costs. Even the Uber CEO admitted that there is currently no link yet between going all-out on AI and delivering successful products.

Despite that, Altman projects that AI token usage will continue to increase. He said that six-and-a-half years ago, the top token spender at the startup used 100,000 tokens a month — today, that is the global per capita average token usage, and that OpenAI’s token leader uses about 100 billion a month. The OpenAI chief also admitted, to his own embarrassment, that someone else uses even more. So, if token usage were to grow linearly, then he would expect the global per capita token usage to hit 100 billion monthly.

However, this is likely under the assumption that token prices will decrease faster compared to the increase in the number of tokens used across the globe. Because, at the moment, some are finding that it’s now more expensive to run AI models compared with hiring people.

Jevons paradox says that the cheaper a particular resource becomes, the more people will use it, and we’re seeing this with AI. But as agentic AI becomes more popular and sophisticated, the number of tokens these systems use has been increasing exponentially, seemingly outpacing the efficiency gains that AI labs have been making on training and inference.

A federal jury in Oakland, California, on Monday unanimously rejected every claim in Elon Musk's lawsuit against OpenAI, CEO Sam Altman, co-founder Greg Brockman, and Microsoft. The nine-member jury found that Musk filed too late, with all claims barred by the statute of limitations. Deliberations began at 8:30 a.m. Pacific and ended at 10:23 a.m. Judge Yvonne Gonzalez Rogers immediately accepted the verdict as her own.

Musk had sought around $130 billion in damages paid to OpenAI's nonprofit arm, the removal of Altman and Brockman from their leadership roles, and the dismantling of the for-profit entity that has turned OpenAI into an $852 billion company. The jury's finding on timeliness meant it never reached the underlying question of whether Altman and Brockman breached their duty to OpenAI's original nonprofit mission.

The case ultimately hinged on when Musk became aware of the alleged breach: California law imposes a three-year window for charitable trust claims and a two-year window for unjust enrichment. Musk testified that he waited to sue because he believed Altman's reassurances over the years, and that Microsoft's $10 billion investment in OpenAI's for-profit arm in 2023 was the moment he concluded the charity had been "stolen".

OpenAI's attorneys countered that Musk had known about the for-profit transition since at least 2017 and had even pushed for it himself, registering a company through his family office intended to serve as a for-profit version of OpenAI. "There's a substantial amount of evidence to support the jury's finding, which is why I was prepared to dismiss on the spot," Judge Gonzalez Rogers said after the verdict.

While Musk's legal team reserves the right to appeal, the judge suggested that it would be difficult because the statute of limitations question was a factual determination, not a legal ruling. The verdict ultimately removes the most prominent legal threat to OpenAI's ongoing restructuring from a nonprofit into a for-profit public benefit corporation.

The company closed a $122 billion funding round in March at an $852 billion valuation, with $30 billion from Nvidia, $50 billion from Amazon, and $30 billion from SoftBank. OpenAI has been preparing for a potential Q4 2026 IPO, though analysts at PitchBook recently suggested the timeline could slip into 2027 given the company's cost structure and $1.15 trillion in long-term infrastructure commitments.

Microsoft, which invested $13 billion in OpenAI between 2019 and 2023 and was named as a co-defendant for allegedly aiding the breach, was also cleared by the jury on the same statute of limitations grounds. "The finding of the jury confirms that what this lawsuit was was a hypocritical attempt to sabotage a competitor," OpenAI attorney William Savitt told reporters outside the courthouse, according to CNN.

Neither Musk, Altman, nor Brockman was present in court to hear the verdict. The three-week trial featured testimony from six tech billionaires, including OpenAI co-founder Ilya Sutskever, and hundreds of pages of private emails, text messages, and internal meeting notes entered into evidence.

Peter Steinberger, the Austrian developer behind OpenClaw who joined OpenAI in February, posted a screenshot of his API usage dashboard on Friday showing $1,305,088.81 in OpenAI spending over 30 days.

The bill covered 603 billion tokens across 7.6 million requests, all generated by roughly 100 Codex instances operated by a team of three people working on the open-source OpenClaw project. OpenAI, which employs Steinberger, covers the cost. The top model on the dashboard was GPT-5.5, dated April 23, 2026. On the day Steinberger posted the screenshot, his account logged $19,985.84 in spend and 206,000 requests.

Steinberger's fleet of Codex agents autonomously reviews pull requests, scans commits for security vulnerabilities, deduplicates GitHub issues, and writes fixes. Some agents open PRs based on the project's broader roadmap, while others monitor performance benchmarks and flag regressions to the team's Discord server. According to The Decoder, certain agents even attend meetings and generate PRs for features that come up in conversation.

OpenClaw itself has had a turbulent few months in the public eye, from wiping Meta's AI Alignment director's inbox to prompting Nvidia to develop its own competitor. But Steinberger has consistently called the project a laboratory for stress-testing what AI-assisted development looks like without budget constraints.

Steinberger clarified in a follow-up post that the $1.3 million figure reflects Codex's "Fast Mode" pricing, which consumes credits at a significantly higher rate than standard execution. Disabling Fast Mode alone would reduce the raw API cost to around $300,000, he said. That itself, however, is revealing, given that a single $200-per-month Codex Pro subscription provides roughly $5,000 to $6,000 in API-equivalent value per billing cycle. By that math, Steinberger’s non-fast-mode usage would equate to approximately 60 Codex Pro subscriptions.

OpenAI estimates that Codex costs between $100 and $200 per developer per month on average, though it warns of high variance depending on factors like model choice and automation intensity. Steinberger's usage sits at the extreme end of that variance, but it puts a number on the gap between what developers pay and the underlying compute costs.

AI coding tools are currently facing growing scrutiny over their cost economics. Codex, Claude Code, and Cursor are all competing aggressively for developer adoption, and all three subsidize inference costs well below API rates to attract and retain users. OpenAI shifted Codex to token-based billing in April, a move that made such subsidies more transparent but also more variable for power users.

Steinberger appeared unconcerned about the bill — easy enough when you’re not paying for it — describing the spending as research into how software development would change if token costs weren’t a constraint. Everything his team builds, he noted, remains open source.

The U.S. Department of War has announced deals with "seven of the world’s leading frontier artificial intelligence companies" for operational use. According to the Classified Networks AI Agreements press release, SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, and Amazon Web Services will deploy their LLMs across the Pentagon’s classified networks “for lawful operational use.” The government said that this move will help turn the United States military into “an AI-first fighting force” and will help with “decision superiority across all domains of warfare.”

It seems that the AI tools that these companies offer will, for now, be limited to data analysis and help make decision-making faster and easier as the U.S. faces complex situations. These tools are accessible via GenAi.mil, the Pentagon’s official AI platform, through the Department of War’s network and are widely available for its personnel.

“Over 1.3 million Department personnel have used the platform, generating tens of millions of prompts and deploying hundreds of thousands of agents in only five months,” the Pentagon said. “Warfighters, civilians and contractors are putting these capabilities to practical use right now, cutting many tasks from months to days.”

Nevertheless, there have been concerns about the use of AI in military applications. Anthropic has famously refused to budge on the Department of War’s demand to lower its safeguards, saying that doing so could mean that its AI products could be used for mass surveillance or to create autonomous weapons. This move resulted in President Donald Trump banning the company from federal agencies, even going as far as designating it a supply chain risk for refusing to bow to the federal government’s demands.

While AI is certainly useful for distilling massive amounts of information and spotting patterns that humans can miss, it’s still not a 100% reliable tool for making decisions that could have a global impact. A researcher discovered this when they pitted GPT-5.2, Claude Sonnet 4, and Gemini 3 against each other in a wargame, with 95% of the outcome ending in a tactical nuclear strike. Three scenarios even ended in a strategic nuclear strike that would have ended the world.

But even though these AI tools are limited to analysis and support, with a human operator at the helm still responsible for every decision, there’s also the risk of automation bias. This is a person’s tendency to follow a computer’s suggestion despite contradictory information, especially as AI systems can process a ton of data so much more quickly than any human could. However, the data the AI is relying on could be false, erroneous, or misinterpreted, so it’s crucial that humans apply their intuition and experience before accepting AI suggestions at face value.

The U.S. military isn’t the only one experimenting with and deploying AI technologies in operational use. China, for example, has been showing off a 200-strong AI drone swarm that can be controlled by a single soldier, as well as ground-based drone wolfpacks armed with machine guns and grenade launchers for urban combat. While we cannot stop these armed institutions from deploying AI tools for intelligence-gathering, reconnaissance, and decision-making on the battlefield, we can only hope that they do not ignore safeguards and never give AI the triggers to any weapon.

In early 2025, OpenAI announced Stargate, a joint venture with Oracle and SoftBank, which aimed to invest $500 billion in AI data centers in the United States. But after more than a year of challenges and disagreements, it seems that the startup has abandoned the original idea of directly owning infrastructure alongside its two partners. According to the Financial Times, OpenAI now prefers to rely on third-party providers and lease capacity in the long term.

This is a sensible idea for the startup, which is burning through cash and has reportedly missed internal revenue targets in recent months. But it has also caused chaos among its partners and put its reliability into question. According to the report, OpenAI has "in practice... abandoned the joint venture," choosing instead large bilateral deals with Oracle and more. One person involved with Stargate reportedly said the company had "sidelined first-party data centres," while OpenAI itself admitted that Stargate is merely an "umbrella for our compute strategy."

Stargate’s initial goal was to build 20 data centers, with the first project at Abilene, Texas, already operational. However, the three partners reportedly squabbled among themselves for months as they could not agree on who would have ultimate control of the planned data centers. In the end, SoftBank agreed to own and develop the Texas data center, while OpenAI would design and operate it on a long-term lease.

Other Stargate projects located in other areas have also been hit by uncertainties. The UK government signed a deal with OpenAI, among other partners, to build a data center in the UK, but the startup has put it on hold earlier this month. It cited “restrictive regulations” and “high energy costs” as the reason behind the move, but UK AI Minister Kanishka Narayan told the Financial Times that the “only thing that has changed [since] the moment of those commitments…has been the financing environment for OpenAI.”

It has also done the same for another Stargate project in Narvik, Norway, with Microsoft stepping up to take over the lease for the site. OpenAI will then lease compute capacity from Redmond, instead of getting it directly from Nscale, the British company that developed the site and also worked on the canceled UK project.

All these changes have got some partners “feeling let down and misled by OpenAI,” a person familiar with Microsoft’s decision said. Thankfully, the software giant has stepped in on some of the projects that the startup has supposedly abandoned. One source told the publication that money is not unlimited, no matter what Sam Altman might say, while another said that they prefer Microsoft over OpenAI as a tenant, as “they are more creditworthy.”

Even though OpenAI has made a name for itself in AI, the startup has not turned a profit since it was founded in 2015. Many institutions believe in its potential, though, with the firm securing $110 billion in its latest funding round — the biggest amount secured in Silicon Valley history and $10 billion more than what the company initially targeted. Still, some analysts estimate that it could run out of cash by mid-2027 with the massive amounts of money it’s been throwing around to secure more compute.

Anthropic CEO Dario Amodei criticized moves like this, saying that some of his company’s rivals are pushing infrastructure investments too far. However, OpenAI says that it’s ahead of the exponential compute curve, allowing it to have an advantage over everyone else. For example, Anthropic has had to limit access to some features on its various products due to limited resources, and Amodei has had to spend more on securing capacity to satisfy the increasing demand

The biggest difference between startups, like OpenAI and Anthropic, and their more established rivals, like Microsoft, Google, Meta, and Amazon, is cash flow. The startups still rely on external funding to fuel their growth, while the big tech companies have billion-dollar revenue that they can rely on to pour into expensive hardware and infrastructure projects.

OpenAI has reportedly missed its internal targets for the number of active ChatGPT users, as well as multiple revenue goals. Because of this, The Wall Street Journal reports that CFO Sarah Friar has expressed worries about whether the firm can afford the billions of dollars of future compute contracts it has taken on.

The company is banking on explosive growth to fund all these expenses, with Altman and Friar saying in a joint statement toWSJ, “We are totally aligned on buying as much compute as we can and working hard on it together every day.” Many corporate investors felt that it was on the right track, with the startup raising $122 billion in its latest funding round, exceeding it $100 billion target. However, one analyst said that OpenAI could run out of cash by mid-2027 unless it continues bringing in massive amounts of investments, like what we saw recently.

Even though executives from Nvidia, Oracle, SoftBank, and many other companies invested in OpenAI see potential, the market is beginning to doubt them. As soon as the news of OpenAI’s shortfall started circulating, MarketWatch reported that firms with heavy involvement in the AI firm saw a drop in their stock prices during pre-market trading. This included Nvidia (-1%), AMD (-4%), Oracle (-5%), and CoreWeave (-5%). SoftBank closed at 9.9% lower in the Tokyo Stock Exchange, making it one of the worst performers in the Nikkei 225.

Microsoft seems to be the only company that’s closely intertwined with OpenAI and hasn’t been affected by a drop in stock prices. That’s because although it has ended its exclusivity agreement with OpenAI recently, it still holds 27% of OpenAI’s for-profit business and has invested billions into the startup.

While OpenAI kickstarted the LLM race when it publicly released ChatGPT in December 2022, many other competitors have since gained ground. The company has lost market share to Anthropic and its Claude family of models, especially among programmers and corporate users. Google's Gemini family has also started to outpace ChatGPT in several benchmarks, leading OpenAI CEO Sam Altman to declare a “Code Red” late last year.

Despite this, Altman has signed deals worth billions of dollars to secure future computing power, including a 4.5-gigawatt contract with Oracle worth $300 billion and a $100 billion alliance that will deliver 10 gigawatts’ worth of Nvidia hardware to data centers.

The market has seemingly panicked about the news of OpenAI’s missed targets, renewing fears that the AI fever among investors could break. Nevertheless, it seems that Altman is pushing to acquire more computing power, arguing that shortages in capacity is what’s limiting the startup’s growth. Anthropic CEO Dario Amodei once said in a conference that some companies are pushing for infrastructure investments too far, but OpenAI disagreed. In a memo addressed to its investors, OpenAI said, “In hindsight, that caution looks less like discipline and more like underestimating how fast demand would arrive" — a statement that might now seem overly optimistic.

Microsoft and OpenAI have once again renegotiated the terms of their deal with one another, but it might be what's best for both of them. OpenAI and Microsoft have announced an end to their exclusive arrangement, and a re-jigging of how they handle model oversight, revenue sharing, and cloud deployments. Microsoft will no longer pay OpenAI for what it makes from Copilot, but OpenAI no longer has to exclusively use Azure servers for ChatGPT, opening it up for further deals with other cloud service providers.

What this means for the ever-nebulous AGI clause that both companies were so keen to retain access to and control over, if and when it materializes, remains to be seen. It's an intriguing move that leaves the immediate future of both companies' AI efforts uncertain, but perhaps it's better than Microsoft's legal department firing all barrels at OpenAI over its recent deal with Amazon.

Where's the ROI?

One of the biggest questions of the AI industry over the past year and a half has been the source of profit. Not the infrastructure investment, or the circular deals and token IOUs, but the real profit. For the investors who pumped tens of billions of dollars into OpenAI, Anthropic, and xAI, and for the shareholders who ballooned Microsoft, Google, and Meta's stock prices off the back of these mega deals and unprecedented investment plans.

Microsoft CEO Satya Nadella hinted at this in January, when he said at the World Economic Forum that AI companies needed to find a clear use for the technology or risk losing the "social permission" to continue the work.

That seems to be more of a pressing issue for Microsoft by April, when it announced that Copilot use on GitHub would move to token-based billing — that is, charging users for the amount of tokens they use, rather than on a per-request basis. No longer would shorter requests with shorter responses cost as much as longer, more in-depth queries. From June, this will result in users paying more when Copilot is verbose in its responses, or when it has to analyze more data before making its suggestions.

Microsoft is already doing that with Azure agents, and it's also set to raise the price of Microsoft 365 with its Copilot integration by several dollars a month for most tiers.

According to internal documents reportedly shared with journalist Ed Zitron, this move came because Microsoft had faced a more-than-doubling of its Copilot-related costs from January this year. He also claims Microsoft will take further steps to tighten controls and increase earnings from individual AI users, including reducing rate limits and forcing users onto different models, which could more than double costs.

Things aren't much better at OpenAI, either. It was projected in January to be on track to run out of money entirely by the end of 2027, and despite announcements of enormous investments in the company, it's projected to burn through tens of billions over the coming years. All while somehow planning to turn a profit by the end of the decade, but to manage that, it would need to earn hundreds of billions of dollars a year. OpenAI's annualized revenue run rate is reportedly sitting at roughly $2 billion per month, or $24 billion a year.

OpenAI also performed several major pivots and navigational shifts in recent months. We learned about its chip manufacturing ambitions in February, it announced it was building a GitHub competitor in March, the company warned that it would shutter the Sora text-to-video generation tool in April, and it bought a podcast for over $100 million that same month.

Even OpenAI's own financial officer has said she doesn't see how OpenAI can afford its own promised infrastructure spending, as it misses key revenue targets in 2026, according to a new WSJ report.

It's very hard to see how any of this takes OpenAI from a heavy-loss-making company to one that's incredibly profitable in just a few years.

Don't drop the bag

OpenAI was under pressure in 2025. To secure the promised investment of billions from Japanese investment firm Softbank, it needed to convert to a for-profit company and settle its disagreements with Microsoft. It managed that just in time, finally securing a long-term partnership agreement with Microsoft in the Fall. The Softbank money came rolling in, and just a few months later, the deal was renegotiated again.

But rejigging the deal may be OpenAI's way of securing the next round of funding — the $50 billion promised investment from Amazon in February, which Microsoft was none-too-pleased about. But in doing so, it's lost one of its limited revenue streams from Microsoft's Copilot earnings, and will still have to pay Microsoft 20% of its own limited earnings.

That Amazon investment could come alongside another $60 billion from Nvidia and SoftBank (though not the $100 billion Jensen originally promised), if all goes to plan. That would also value the company at around $730 billion, making a potential IPO incredibly profitable for Altman and anyone else holding OpenAI shares at the time of a public offering.

But even with OpenAI more than halving its compute ambitions from $1.4 trillion in expenditure to $600 billion by 2030, that's still contingent on increasing its own revenue to $280 billion a year by that same date. As of the time of writing, OpenAI hasn't even managed to earn 10% of that, while having close to a billion active users (though crucially, it also missed that milestone by the end of 2025), and it is losing mindshare to competitors like Anthropic.

Regardless, OpenAI seems keen to push forward with its IPO plans. At this stage, that may be the only real avenue left for it to get anywhere close to its ambitious goals. Even with shifting goalposts, the timeline for its profitability is shrinking rapidly, and it still hasn't made a clear path toward it.

When Nathan Calvin, vice president and general counsel at AI advocacy group Encode, received a press inquiry last week from a reporter named Michael Chen, the email looked slightly off, featuring loaded questions with the only format offered being a written Q&A

Alarm bells began to ring, and Calvin forwarded it to Tyler Johnston, executive director of the AI safety nonprofit The Midas Project, who ran it through an AI detection tool. The email, the reporter, and nearly every article on the publication that sent it turned out to be machine-generated, according to an investigation Johnston published on Friday in Model Republic.

The site, called The Wire by Acutus, has published 94 articles since late December using a fully automated pipeline that drafts stories, reviews them, and deploys bots to solicit quotes from real people under fake bylines. An “AI detection” scan of the full archive found 69% of the articles were entirely machine-generated, with another 28% partially so. And in true AI vibe-coder fashion, the site’s publicly accessible JavaScript and API endpoints laid bare the entire content production system.

Johnston found that the Acutus website is built as a React application, and its client-side JavaScript contains elements of an internal editorial dashboard that absolutely weren’t intended to be public-facing. Fields in the dashboard include "AI Background Context," described as background material for the AI to draw on when producing questions and writing stories, and a large "Generate Story Draft" button that automates article creation. A separate "Regenerate" function allows operators to re-run the process if the output is unsatisfactory.

The site's API, accessible at a standard URL in any browser, returned not just finished articles but the full internal record of how each piece was produced. That record includes an automated multi-pass editorial review scored across categories like AP style compliance, quote accuracy, and source verification. Johnston reported that the median time between the first review issue being resolved and the last was 44 seconds, with publication typically following 10 seconds later. Of the 94 stories in the database, 42 carried an automated status of "needs_revision" from the site's own AI reviewer, but all 42 were published regardless.

The investigation began when Calvin, having received the press inquiry, couldn’t find any record of a “Michael Chen” as being associated with the publication. The email itself, which came from a generic reporters@acutuswire.com email address, was flagged as AI-generated by the detection tool Pangram. The site's source code also contained an interview infrastructure designed to conduct outreach and gather quotes through automated written Q&A exchanges.

In addition, Johnston traced a connection between Acutus and OpenAI's political operation. The site had almost no public profile, and its articles had been shared on X only four times, but roughly half of that engagement came from a single person: Patrick Hynes, president of the PR firm Novus Public Affairs.

Novus lists Targeted Victory among its clients. Targeted Victory is the Republican consulting firm whose CEO, Zac Moffatt, co-founded Leading The Future, a $125 million super PAC backed by OpenAI president Greg Brockman and venture capital firm Andreessen Horowitz. The super PAC launched in August 2025 with the stated goal of opposing state-level AI regulation and supporting pro-AI candidates.

According to Johnston, Hynes appeared as a quoted source in an Acutus article, praising a New Hampshire governor's housing policy on behalf of Novus, with no disclosure that his firm appeared to be operating the publication that was quoting him.

The article's angle matched the deregulatory position of the New Hampshire Home Builders Association, a Novus client. The site's content followed no coherent editorial identity but instead closely mirrored what a PR firm's client roster might produce, Johnston wrote, with articles favorable to the pharmaceutical industry, the cryptocurrency lobby, and multiple 2026 Republican Senate campaigns appearing alongside pro-AI policy coverage.

Several U.S. data centers slated for completion in 2026 are at risk of being delayed as strict schedules encounter regulatory friction, supply chain bottlenecks, and the lack of available utility. According to a report by the Financial Times, major data center projects involving Microsoft, OpenAI, and other tech companies will miss projected deadlines by more than three months. The estimate is based on data from SynMax, a geospatial data analytics company that uses satellite imaging and AI to deliver real-time insights and predictive analytics on the maritime and energy sectors.

Satellite imagery is used to estimate progress on various construction projects, looking for various milestones like land clearing and foundation work. It’s then cross-checked against industry intelligence, including public statements, regulatory and permit documents, and on-the-ground interviews.

For example, a 1,200-acre, 10-building campus is under construction in Shackelford County, Texas, for Oracle, which it will then equip for OpenAI. The entire project is expected to have a 1.4-GW capacity and a delivery date in the latter half of 2026, but imagery from early April 2026 shows that only six plots of land have been cleared for construction, with only one of them showing signs of development. SynMax estimates that one building could possibly be delivered by the end of the year, but a more realistic timeline sees this pushed to 2027. Another OpenAI-linked project, a 1.2-GW site in Milam County, Texas, is showing signs of slow progress, with only one building under construction as seen from space.

The companies involved in these data center projects denied the reported delays. “Our historic data center build-out is on schedule and we will accelerate from here,” OpenAI told the publication. “In partnership with Oracle, SB Energy and a broader ecosystem of partners, we are delivering rapid progress in Abilene, Shackelford County and Milam County in Texas.” Oracle also said to FT, “Each data center we’re developing for OpenAI is moving forward on time, and construction is proceeding according to plan,” while SB Energy noted that “The Milam County Data Center is on schedule and on pace to be one of the fastest data centers of its kind ever delivered.”

People on the ground report otherwise, though. Construction executives report that the building sites are lacking in specialist workers like electricians and pipe fitters, an issue that has been reported since late 2025. Note that OpenAI’s data center projects aren’t the only ones apparently suffering from a delay, with another recent report claiming that half of planned U.S. data centers are reportedly being canceled or delayed because of shortages.

Building the structures to house AI GPUs isn’t the only bottleneck in building the data centers — local utility providers are straining to catch up with the increased demand for electricity. Even though AI companies are supposedly paying for the needed infrastructure upgrades required to supply the massive amounts of electricity these data centers will consume, it will take time to order, deploy, and build the necessary systems that will deliver the power that AI GPUs demand.

Some AI hyperscalers are turning to on-site generators like turbines as an alternative power source, but these require EPA permits, adding regulatory friction. Furthermore, jet engines are suffering from their own supply chain shortages, with orders from 2025 being slated for a 2028 to 2030 delivery.

These delays do not mean that the projects will not be built, only that they will take longer than expected. This might get investors, who are pouring in trillions of dollars on these projects, feeling antsy, though, as they still cannot see the massive returns that they’re expecting to materialize in the next few years, if these AI data centers can even turn a profit at all.

Iran's Islamic Revolutionary Guard Corps (IRGC) has issued a clear public warning to the U.S. that any damage inflicted on Iran’s power infrastructure will be met with decisive retaliation. Specifically, IRGC spokesperson Brigadier General Ebrahim Zolfaghari threatened the “complete and utter annihilation” of U.S. and Israeli facilities. The ‘hidden’ $30 billion Stargate AI datacenter in Abu Dhabi was singled out as a juicy target for Iranian destruction later in the video. The threats come on the heels of Iran reportedly delivering enough damage via rocket strikes to some Amazon AWS data centers that they have shut down.

Above, you can see and hear Zolfaghari make the headlining threats against U.S. action in Iran. “Should the USA proceed with its threats concerning Iran’s power plant facilities the following retaliatory measures shall be promptly enacted,” declares the military spokesperson. “All power plants, energy infrastructure, and information and communications technology of the Zionist regime, and all similar companies within the region that have American shareholders shall face complete and utter annihilation.”

After the remarks from Zolfaghari, the video switches to a shot of the Earth from space, which zooms into Abu Dhabi on Google Maps. A zone not far from the coast is then centered on, showing an apparently ‘empty’ area of desert. However, a message is overlaid on this bleak view, stating “Nothing stays hidden to our sight, though hidden by Google.” The video then switches to a ‘night vision’ view of the same area of the map with the full extent of the Stargate AI datacenter in Abu Dhabi clear to see.

Escalation potential or a bluff?

One might query why Iran hasn’t already struck targets like the Stargate AI data center during the last month of hostilities if it were at all possible. It has already inflicted enough damage to disrupt operations at some Amazon AWS data centers. Perhaps these were just ‘lucky’ shots that managed to get through defenses deployed in the Gulf States. Moreover, Iran has made similar threats against Nvidia, Microsoft, Apple, Google, and 14 other U.S. tech companies over recent weeks.

With neither side looking like it will cool down the rhetoric and throttle back on the use of force right now, we might actually find out whether Iran can mount a devastating attack on U.S. business-related data centers or not. There will be lives at stake, of course, people are working at these facilities, so it isn’t all about the $30 billion that has been sunk into these huge projects in the Middle East.

OpenAI keeps cutting deals left and right, one of the latest ones being a massive partnership with Amazon that ought to see a total of $188 billion circle between both companies. Sam Altman may have cut that one a little too close, according to Microsoft, which is reportedly considering releasing the lawyers over an API exclusivity clause.

Sources from the Financial Times (FT) say the key item in this discussion is OpenAI's Frontier multi-agent platform targeted at large enterprises. Broadly speaking, Frontier offers to make it easy for large enterprises to effectively use AI by wiring up multiple agents ("workers") with shared memory and business content.

Microsoft is apparently taking umbrage with the situation, despite its position in the partnership having been repeatedly revised. Redmond was originally OpenAI's sole cloud services provider, but eventually changed to having the right of first refusal over said services, and was further weakened in October 2025.

The PR about that latest agreement states that "API products developed with third parties will be exclusive to Azure. Non-API products may be served on any cloud provider." Under that logic, OpenAI has the freedom to develop and implement new products, but if they offer them as APIs, they have to go through Azure.

Redmond believes that OpenAI's offering access to Frontier via Amazon Web Services (AWS)'s Bedrock platform would be in breach of the agreement. Getting even more technical, the dispute may well come down to the definition of a "stateless" versus "stateful" when applied to AI models.

Even though it appears to remember your information, a standard chatbot is actually stateless — adding a new question requires the bot to re-process the entire conversation again. A storage and orchestration layer to facilitate something like Frontier is arguably a "stateful" implementation, more specifically a "Stateful Runtime Environment."

According to FT's sources, Microsoft thinks that running Frontier on AWS instead of Azure would breach either the spirit or the letter of the contract. This is illustrated by a report that Amazon is pointedly instructing its staff to never say that SRE "enables access" or "calls on" ChatGPT as a backend, instead preferring vaguer terms like "powered by," "enabled by," or "integrates with."

The whole stateless/stateful discussion has reportedly been a hot topic among lawyers from both camps, though FT states that a Microsoft employee referring to the situation isn't mincing words, saying that "we know our contract," and that "we will sue them if they breach it. If Amazon and OpenAI want to take a bet on the creativity of their contractual lawyers, I would back us, not them."

Predictably, OpenAI's position is the opposite, as the firm seemingly believes the Amazon deal is compatible with the Microsoft agreement. FT further reports that Amazon and OpenAI are building an unspecified system meant to work around the contract.

Nevertheless, FT points out that this latest development may cast a pall over OpenAI's upcoming IPO by placing doubts in prospective stockholders' hearts. That would be a nightmare scenario for the company, as the magnitude of its ongoing investments means that the flow of money cannot stop.

As AI clusters grow larger, they begin to use optical interconnects for scale-out connectivity. However, the day when they require optical interconnects for scale-up connectivity may be approaching soon. To prep for that day, hyperscalers Microsoft, Meta, and OpenAI teamed up with hardware designers AMD, Broadcom, and Nvidia to develop a protocol-agnostic scale-up interconnection technology for AI clusters.

To do so, the group of companies this week established the Optical Compute Interconnect (OCI) Multi-Source Agreement (MSA) group to define an open optical connectivity specification for scale-up interconnections used inside large AI systems and racks to enable hyperscalers to use optical cables instead of copper to connect more accelerators at high speed and predictable power. In practice, this means the consortium will develop a common optical physical layer (PHY) and unified components to support various protocols, such as UALink for AMD and Broadcom, and NVLink for Nvidia.

The OCI connectivity technology for short-reach optical links used in AI racks and scale-up clusters will define a common PHY based on NRZ signaling and wavelength-division multiplexing (WDM), starting at four wavelengths × 50 Gb/s (200 Gb/s per direction) and scaling all the way to 800 Gb/s per fiber. Over time, the roadmap is expected to expand both wavelength counts and signaling rates, targeting 3.2 Tb/s per fiber and beyond as the ecosystem evolves. The technology will support pluggable optical modules, on-board optics, and co-packaged optics (CPO) integrated directly with compute silicon.

"The growing need for optical scale-up interconnect to support large AI systems later this decade is clear," said Brian Amick, Senior Vice President, Technology & Engineering at AMD. "AMD is a founding member and strong supporter of the OCI MSA as it establishes an open specification for the industry to foster a robust, multi-vendor optical scale-up interconnect ecosystem."

The common optical layer will enable different processors and interconnect protocols to operate over the same fiber infrastructure and switches from different suppliers, ensuring flexibility for hyperscalers while retaining the competitive advantages of the protocols used by developers of AI accelerators, AI GPUs, XPUs, and other processors. In addition, the standardized OCI roadmap is meant to simplify system integration, reduce development risk, and shorten deployment cycles for new generations of AI hardware.

"Broadcom is proud to draw upon our multi-generational CPO platform and industry partnerships to drive the OCI specification forward," said Near Margalit, Vice President & General Manager, Optical Systems Division at Broadcom. "The OCI-MSA allows for seamless integration with existing electrical SerDes-based ASICs while providing a clear path to direct ASIC integration, ensuring the ecosystem remains flexible and high-performing."

While the OCI MSA group is headed by AMD, Broadcom, and Microsoft, which are known supporters of open industry standards, this is clearly not a traditional standard body like the Ultra Ethernet Consortium or UALink Consortium, which will have an impact on how the technology is developed.

Firstly, the OCI MSA is hyperscaler-driven rather than vendor-driven. The arrangement is unlike most industry consortia, which are organized and led by independent hardware vendors (IHVs), IP companies, and networking suppliers.

Secondly, OCI targets a very specific architectural layer of AI systems — short-reach links that connect accelerators and switches within a scale-up domain. By contrast, traditional hardware development groups tend to standardize on a vertically integrated set of technologies to be widely adopted across a market or markets.

Thirdly, as the organization is a Multi-Source Agreement (MSA) group, it will, by definition, be faster than a typical industry standard setting body. MSAs are meant to enable select companies to align on electrical/optical interfaces and build interoperable products quickly, without the lengthy consensus processes typical of classic organizations like JEDEC or the Ultra Ethernet Consortium (which are designed to unite tens or hundreds of companies and support an entire industry). The OCI MSA — at least for now — will enable AMD, Broadcom, and Nvidia to build interoperable short-reach interconnections for Microsoft, Meta, and OpenAI.

"Nvidia is a founding member of the OCI MSA to establish a common optical standard across global AI infrastructures," said Gilad Shainer, Senior Vice President of Networking at Nvidia. "By equipping best-in-class compute with state-of-the-art optics, the OCI MSA can deliver the scale and performance required by the next era of super-intelligence."

Oracle and OpenAI dropped plans to expand their flagship Stargate campus in Abilene, Texas, beyond its committed 1.2-gigawatt build earlier this month, after financing negotiations failed and winter weather disrupted parts of the liquid-cooling infrastructure.

Bloomberg initially reported on the decision on March 6, Reuters followed, and within hours, both outlets and the social accounts aggregating them had reduced a multi-factor infrastructure decision to a single word: "canceled." Oracle posted a rebuttal on X.com early Monday morning, calling all coverage "false and incorrect,” and by March 9, a screenshot purportedly from an OpenAI employee was circulating on social media, appearing to confirm the original reporting. It has since been confirmed that the LinkedIn post was fake, however. And digging deeper into the story yields even more intriguing wrinkles.

Abilene's expansion lease

What needs to be made clear is that the existing Abilene campus is unaffected. Oracle has confirmed that its 1,000-acre, eight-building facility (developed by Crusoe on Lancium's Clean Campus) continues to operate. Two buildings are already running training and inference workloads on Nvidia GB200 Blackwell racks, and the remaining six are scheduled for completion by mid-2026, bringing total capacity to roughly 1.2 gigawatts.

What did in fact collapse was a separate, never-finalized expansion lease that would have pushed the site toward 2 gigawatts. OpenAI's compute scaling executive, Sachin Katti, said publicly the company "considered expanding it further" but chose to direct that capacity to other locations instead.

Financing difficulties and a multi-day winter weather outage that took several buildings offline by disrupting liquid cooling equipment drew most of the coverage. The Information also reported that power at the expansion site won't be ready for roughly a year — calling it the “real reason” OpenAI walked away from the expansion — and by the time it is, Nvidia's Vera Rubin architecture will be available.

Rubin delivers approximately ten times lower cost per token and five times better inference performance over Blackwell while requiring four times fewer GPUs for mixture-of-experts training; it is due to ship in the second half of 2026. OpenAI and Nvidia signed a letter of intent in September 2025 to deploy at least 10 gigawatts of Rubin-class systems. For OpenAI, building 600 megawatts of additional Blackwell capacity at a site where the grid won't be ready until after Rubin ships means paying for hardware that is a generation behind before the buildings are live — that just doesn’t make sense commercially.

Nvidia's GPU generations are turning over faster than power infrastructure can follow — Rubin is set to be superseded by an 'Ultra' style variant in 2027, with the very much under wraps Feynman coming in 2028. More details surrounding Nvidia's roadmap are expected at the upcoming GTC 2026 conference.

Oracle’s position is showing us how badly these provisioning timelines can diverge, with the company reportedly carrying more than $100 billion in debt to fund its Stargate commitments, with free cash flow now negative.

Oracle's rebuttal

Oracle's X post on March 9, which called reports about Abilene "false and incorrect," said Crusoe and Oracle are "operating in lockstep." It confirmed two buildings are fully operational, and stated that leasing for the broader 4.5-gigawatt commitment had been "completed." Unfortunately, none of these statements addresses whether the Abilene expansion lease was dropped, as Bloomberg and Reuters reported.

Bloomberg's Ed Ludlow clarified on X shortly after Oracle's post that the two companies "are not moving ahead with the planned expansion lease," while the broader 4.5-gigawatt agreement "remains on track, with additional projects announced, including a site near Detroit." Oracle opted to defend the bigger picture without engaging the specifics, leaving many unable to determine what was actually going on.

The company then came out again early Tuesday morning with a new post to X, stating that recent media reporting about its data centers “[reflects] a fundamental misunderstanding of how AI data centers are built and operated,” going on to add that the flagship Abilene site “remains on schedule…. any claim that the planned capacity at this site is delayed is inaccurate.”

Shareholders have separately sued Oracle over disclosure claims related to the buildout, which helps explain why Oracle's communications team came out so hard against reporting it could have simply declined to address. The broader 4.5-gigawatt Oracle-OpenAI capacity agreement announced in July 2025 remains on track across multiple U.S. sites, including campuses near Detroit and in Wisconsin.

Nvidia steps in

With the Abilene expansion space unoccupied, Crusoe began looking for a new tenant for the 600 megawatts of capacity OpenAI had walked away from. Bloomberg reported that Nvidia paid a $150 million deposit to Crusoe to secure that space and then approached Meta about taking the lease, specifically to prevent AMD Instinct hardware from landing where its own silicon was already deployed. But no deal between Meta and Crusoe has been confirmed.

If all this wasn’t already confusing enough, a screenshot then began circulating on social media, purportedly showing a LinkedIn post from a "Michael Su," described as an OpenAI or Crusoe affiliate with an insider account of the Abilene situation.

Tech writer Ed Zitron flagged it on Bluesky as "more than likely fake," citing a deepfake scanner result, a LinkedIn URL reading "michaelsermon" that did not match the name on the post, and no verifiable employment history for the person at any of the companies listed. No official channel authenticated the post, and the LinkedIn profile has since been removed.

The Abilene expansion, for now, appears to be cancelled, but the deal still reportedly remains on-track. With the AI data center buildout still in its early stages, expect more turbulence ahead. Nvidia's Jensen Huang believes that it could take up to 50 years to complete global 'AI factory' rollout, and if true, stories like Abilene will become ever more commonplace.

Oracle has pushed back against recent media coverage of its Stargate AI data center project in Abilene, Texas, saying reports about problems at the site are “false and incorrect” while confirming the broader buildout tied to OpenAI remains on track.

In a statement posted to X today, Oracle said it and developer Crusoe are “operating in lockstep” to deliver one of the world’s largest AI data centers at the Abilene campus, adding that two buildings are already operational and the remainder of the site is progressing as planned. The company also said it has completed leasing arrangements for an additional 4.5 gigawatts of capacity to support its commitments to OpenAI.

This rebuttal follows recent reporting from Bloomberg and Reuters that said Oracle and OpenAI had abandoned plans to expand the Abilene site with an additional 600 megawatts of capacity after financing negotiations dragged on and OpenAI’s infrastructure needs shifted.

The Abilene campus, developed by Crusoe, already includes eight data center buildings spanning roughly 1,000 acres, with Oracle operating the facilities as part of its Oracle Cloud Infrastructure footprint. Two buildings are currently running workloads while the remaining structures are still under construction.

The reported change relates specifically to a planned expansion near the existing campus rather than the main Stargate site itself. Sources cited by Reuters have reportedly said that the extra 600 MW originally discussed for Abilene would instead be fulfilled at other data center campuses tied to the broader Stargate rollout. Oracle and OpenAI announced an agreement in July to develop up to 4.5 gigawatts of additional AI data center capacity across multiple U.S. locations as demand for training and inference continues to surge.

In addition to Reuters reporting, Bloomberg claimed that Meta is evaluating the additional space originally earmarked for OpenAI, with Nvidia helping facilitate discussions to ensure its AI accelerators are used at the site rather than competing hardware from AMD.

Bloomberg’s Ed Ludlow, who co-authored the piece with Brody Ford and Dina Bass, also took to X.com to issue a correction shortly after Oracle’s rebuttal, stating that while Oracle and OpenAI “are not moving ahead with the planned expansion lease,” the separate agreement between the two to develop 4.5 gigawatts of data center capacity for OpenAI “remains on track, with additional projects announced, including a site near Detroit.”

Oracle and OpenAI have cancelled plans to expand their flagship AI data center campus in Abilene, Texas, after lengthy negotiations broke down over financing arrangements and OpenAI's changing capacity projections, reports Bloomberg. The Abilene campus is part of the Stargate project announced at the White House last year, and yet, it seems like it faces slowdowns. Just like other Oracle's data centers meant for OpenAI. Yet, Meta could take the yet-to-be-expanded space.

Since mid-2025, Oracle, Crusoe, and OpenAI have discussed increasing data center power capacity from about 1.2 GW to roughly 2.0 GW, amid reluctance from locals. Negotiations got complicated due to difficult financing terms and OpenAI’s shifting capacity forecasts, which led to their collapse, according to Bloomberg. Nonetheless, development of the 1,000-acre campus remains underway, and multiple facilities are already in service, though preliminary agreements to rent a substantial expansion were ultimately dropped. Could it be a signal that the Stargate project fails while the whole AI industry is on the rise?

The Abilene campus remains one of the biggest AI data center projects announced so far, yet to date, it has been known primarily as a part of the widely publicized Stargate project. Oracle has been rapidly installing Nvidia-based servers used by OpenAI to train and deploy AI models and systems. However, relations between Oracle and Crusoe have been strained by reliability issues. Earlier this year, winter weather disrupted parts of the liquid-cooling infrastructure, forcing several buildings offline for multiple days. Both companies say cooperation remains strong and development continues swiftly, yet the source report clearly notes hiccups.

Given the rising tensions between the Stargate partners, Crusoe began searching for another tenant, according to Bloomberg. At that point, Nvidia reportedly stepped in to help ensure the site would continue deploying its hardware rather than systems powered by AMD. Furthermore, Nvidia provided Crusoe with a $150 million deposit and assisted efforts to attract Meta — which is not a part of the Stargate project — as a prospective tenant for the additional capacity, the report says. Meanwhile, Meta has yet to confirm its expansion at the Abilene campus.

Despite shelving the expansion of one Stargate project, Oracle's general partnership with OpenAI remains unchanged. In July last year, Oracle agreed to develop 4.5 GW of data center capacity for OpenAI, and that program continues. The companies have also announced projects in other locations, including a site near Detroit owned by Related Digital.

*One gigawatt is comparable to the output of a nuclear reactor and can supply electricity to hundreds of thousands of homes at peak usage. That being said, a nuclear power plant was not reported to be a part of negotiations, which perhaps explains why locals were against increasing power capacity using things like coal or gas generators, yet we are speculating here.

Since late 2025, thanks to the AI boom, we’ve seen prices skyrocket for RAM, video cards, and now storage, making building or buying a PC today much more expensive than it was. Where you might, in the past, spend more on mid-range or even a premium-class motherboard, now that build budget is probably going to some other piece of high-priced hardware.

So, where can you save money without losing performance? One of those areas is the motherboard. For the latest-generation Intel and AMD boards, prices range from wallet-emptying $900 to $1,200 for Asus, Gigabyte, and MSI flagships, down to around $90 across the three primary chipsets (Intel Z890/B860/H810 and AMD X870E/B850/B840). There are lesser desktop chipsets from both camps – H810 for Intel and B840 for AMD. And although these motherboards are entry-level and cost less, they’re mainly meant for office use, everyday computing, or budget-oriented PCs. They’re essentially for those who really don’t need a ton of connectivity but still want the benefits of the latest platform, whereas others are more feature-rich and capable.

As you can see, there’s a huge price and feature gap between the cheapest and most expensive motherboards. The priciest boards offer the best hardware available for the platform, including 10 GbE, Wi-Fi 7, flagship-class audio, robust power delivery, fast memory support, fast (and more) storage, and more. The cheap boards use slower networking, a lower-quality audio codec, fewer power-delivery phases, and slower memory support, with fewer storage ports. But some boards punch above their weight class, while others may be overpriced relative to their features. So you should understand your wants and needs before making a purchase.

But the real question you should ask yourself when buying a motherboard today is, do you need most of these high-end features, or can you work with a much less expensive options with fewer, or perhaps, slower features? How bad, really, are the cheapest (or at least cheaper) motherboards?

What do you give up, how much can you save, and of course, how cheap is too cheap? The answers will vary depending on your needs, but we’ll dig into all of these questions below. Hopefully, you walk away with more knowledge to make a better-informed decision about saving money where you can, and perhaps spend the savings on RAM, video cards, or storage, where prices have really gone up.

AMD and Intel Chipsets: What you get on paper

Let’s start with what each chipset includes, so you can get a high-level view of what each offers. Note that what you see listed in our table is a minimum. Sometimes boards will add controllers for additional USB or SATA ports, a second Ethernet port, or additional functionality, such as bifurcating the PCIe slots. But here’s a table showing what each of the chipsets offers as a base:

As you can glean from the chart, AMD’s dual PROM21 chips that make up the X870E chipset offer the most native connectivity. The single-PROM21 chip X870 drops some things, and B850 and B840, which also use the same single PROM21 chip, drops more. When you get down to B850, and especially B840, you can lose PCIe 5.0 on the slot (B840 only supports PCIe 4.0), and some M.2 storage, and you generally won’t see USB4 ports either. In short, the further down you go in chipset families, the fewer of everything will be available.

Similar to AMD, Intel’s current-gen flagship chipset, Z890, offers all the bells and whistles (from a single chip, note), whereas B860 and especially H810 offer less of almost everything. Fewer USB ports, PCIe 5.0 slots, and M.2 storage. The further down you go, the less there is to start, and more becomes optional.

The key is knowing what you need today and what you’re likely to want in the future, so you can decide whether a cheap motherboard without some high-end features will be sufficient for your needs.

The main features you’re likely to lose by going cheap, and what matters most

One thing that definitely declines when going from flagship to more mainstream boards is the quality of Voltage Regulation Modules (VRMs). Budget-class motherboards list support for all compatible CPUs, but the MOSFETs and Chokes used on extreme-budget boards (in particular, the business-class chipsets from Intel and AMD) may not allow a high-power CPU to maintain its performance, as they can get too hot and throttle, lowering the voltage and clock speed. So one thing you definitely don’t want to do is pair a cheap H810 or B840 motherboard with a flagship-class processor and expect 100% performance all the time. Unless you plan to use an APU or low-power desktop CPU, I’d avoid any board without heatsinks on the VRMs.

Memory support is another specification that looks great on paper but doesn’t matter in the grand scheme of things, at least when talking about performance. The sweet spot for AMD machines today is around 6000-6400 MT/s, with the lowest CL rating. Intel supports higher memory speeds than AMD, thanks to CU DIMMs (with a built in clock driver to support the higher speeds). But the price and benefits of going that high (9000 MT/s or more) are rarely worth the cost of admission unless you're trying to break records. So Intel’s price-to-performance sweetspot, regardless of the higher supported speeds, is still a lot lower than the ceiling for most motherboards, and similar to AMD in the 6400 MT/s range, or even a bit higher.

In short, most motherboards will happily run faster RAM, but the performance difference between a DDR5-6000 CL36 kit and a DDR5-7200 CL34 kit (as we run in our testing) isn’t much. And the price for the same 32GB capacity at the higher speed is almost 25% more ($379.99 versus $474.99 for the same Patriot kit). The performance difference between the two memory speeds is only a couple of percent at best across real-world applications (in part due to the memory fabric dropping from 1:1 to 1:2). Unless you’re trying to break records or need extreme memory bandwidth for your work, you don’t need to worry about memory support on cheap motherboards, as most will run past what the platform is rated for and outside of the lowest Intel chipset (H810), capable of reaching these sweet spot speeds.

PCI-Express excess?

Another consideration is PCIe support. The fastest available on current platforms is PCIe 5.0, and you can use that bandwidth in both the PCIe slot(s) and the M.2 socket(s). PCI scaling, even on the best video card today, the RTX 5090, doesn’t matter much when you’re talking PCIe 5.0 x16/x8, or 4.0 x16. The difference is a margin of error for gaming, but can be more for other activities (like video rendering and game development - according to Puget Sound), with lower bandwidth. Even on the extreme budget side of things, it shouldn’t matter, as there’s at least one full bandwidth PCIe slot. Keep it above PCIe 4.0 x8, and you’d only notice any difference in benchmarks. Just be careful: On some boards, there is lane sharing between the PCIe slots and M.2 sockets, so installing a drive will cut the PCIe slot bandwidth in half; check the specs closely before buying.

Storage speeds and quantity

Storage is another important element. Unless you’re only ever going to install one drive, M.2 socket count, speed, and SATA port count are all important considerations when choosing a motherboard. On the most expensive motherboards, you get up to seven M.2 sockets (using included add-in-cards), with four PCIe 5.0 (128 Gbps) capable. And at the bottom end, it’s PCIe 4.0 x4 (64 Gbps) or half that with PCIe 3.0. That sounds like a big difference, and on paper it is, but you won’t notice a difference between PCIe 5.0 and 4.0-based M.2 storage unless you’re often transferring huge files between the fastest storage devices. And given the current price of SSDs, many more people will likely be living with PCIe 3.0 speeds, which is still generally fine for mainstream computing and gaming.

Rear expansion options

USB count on the rear IO (and front panel) is also a critical are of consideration. Too few, and you don’t have enough ports for your peripherals without adding a hub. Too many, or paying for speed you won’t use, can also be a waste, but more is generally better in this case. Most boards come with at least one Type-C and several Type-A with varying speeds. The higher up the chipset, the more speed and ports you’ll see, but the lower you go, the fewer. Case in point: Many of the really inexpensive motherboards don’t include a front-panel Type-C port of any kind, rendering that useful port on your case useless (at least without spending more money on adapters or an add-in card).

Wired and wireless networking

Networking on the cheapest of boards will still be fast enough for most users. Even if the board comes with a single 1 GbE and integrated Wi-fi 6/6E, that’s still plenty fast for most users. And many don’t have a 6E or above router to take advantage of the increased Wi-Fi speeds/specs. Obviously, as you climb the product stack, you see faster speeds (2.5/5/10 GbE) and the same with Wi-Fi (up to Wi-Fi 7). But most of us are using Gigabit internet and Ethernet, or less in the case of internet, so the only way to take advantage of the extra bandwidth is through a LAN (say, a NAS) with the same speed or faster ports. That said, some boards don’t ship with Wi-Fi at all, which is fine if you’re using Ethernet. Adding even the fastest M.2-based Wi-Fi 7 card is relatively cheap ($33.99) if you end up needing it.

Audio is another item that tends to fall by the wayside for most users. Right now, there are five prevalent codecs on the market. The older, basic Realtek ALC897, the last-generation Realtek ALC1200/1220, and the latest, ALC4080/4082. Even the ALC897 is sufficient for most people, but if you’re a gamer or a discerning listener with a decent set of speakers or cans, you’ll want to see the 1200 or 4000 series codecs in use. Or, if you have a pair of AudioEngine A2+ speakers (see our best PC speakers page) or another set of speakers with a built-in DAC, it’s irrelevant. As you move towards the top of the stack, Boards often include third-party DACs and amplifiers, which further improves things with the right equipment.

Build quality and good looks

Last but not least is build quality and aesthetics. Build quality is one of those things that sounds more important than it usually is. I’m not discounting the importance so much as saying it’s not often we see a spate of failures plague motherboards (though ASRock and Asus’ woes recently with AMD processors could constitute such a situation). More often, it's random, one-off issues. So, the build quality from the factory is generally good (or at least good enough), regardless of board class.

While the properties of motherboard components do differ through the product stack (like layers of the PCB or amount of copper used in the traces), for the most part, it doesn’t matter. More is generally still better, especially for those using high-end processors and planning to overclock (PBO or manual), but it also adds complexity and potential failure points. In other words, any board can be faulty or fail in several ways, regardless of price. So keep your receipts for at least a while after your system is up and running without issues.

Aesthetics is a polarizing subject. If you go cheap, budget-class boards tend to have fewer heatsinks, exposing more of the PCB, and they lack RGB (though you can add lighting through any onboard ARGB/RGB headers); budget boards generally do not look as good as the more expensive offerings. As you step up in price, you’ll see larger heatsinks, RGB lighting, and more ornate designs and features such as LCD screens on a few high-end/flagship models, or even faux-wood accents like on the Gigabyte X870E Aero Wood (pictured above). But if your board is going into a case without a window, or it's a function-over-form machine, looks don’t really matter. Still, the further down the stack you go, the ‘worse’ a motherboard generally looks.

Which Features Matter Most

When you buy a system, the primary objective is to generally maximize performance while minimizing cost. And you can find our expert selections on the best motherboard and best motherboard deals pages. But the most critical features are those that align with the system's use case, although certain essential features remain important regardless of the planned use. One of the things we’ve learned from years of motherboard testing is that there isn’t a significant performance difference between flagship and inexpensive motherboards, so long as cooling doesn’t put a glass ceiling on your processor.

Below, you can see several benchmarks, including games, highlighting the small performance difference between a $189.99 motherboard and a $1,099.99 motherboard. Most results are extremely close together, sometimes falling within the margin of error.

General Advice/Bottom Line

Generally, my advice is not to buy the cheapest board you can, as the savings you may find now can (and often will) cost you in the long term if you need expansion or faster storage. Think about your use case, not just the price tag. A bottom-of-the-barrel board is OK for ultra-budget builds, office environments where performance isn’t a factor, and secondary or temporary systems, perhaps an HTPC or a NAS. In 2026, you can expect to find well-equipped motherboards from both Intel and AMD for $190-$300. In that price bracket, power delivery generally isn’t a concern unless you’re pushing high-end chips and overclocks to extremes; there are typically ample USB ports (though often not the fastest around); and you tend to get generous storage options, fast networking, and a decent appearance.

So when is it worth paying up and when isn’t it? That’s a complicated question that varies by situation. But in general, I would pay the piper if:

• Performance is notably worse (it’s generally not)
• Extra features are actually useful to you
• There are bottlenecks in your use case (think USB, storage, networking, CPU)

Also, consider how much you will have to pay in the future to get features back if you need them.

I would not pay more for a board if:

• It’s for an ultra-budget build or office / HTPC use
• Higher-end features won’t be used
• It’s a secondary/temporary system you aren’t going to use all the time

When choosing a budget motherboard, you want to consider:

• CPU support (current and next gen)
• VRM quality and cooling (particularly when using a high-end processor)
• Do you overclock (you’ll need to select a chipset that supports that)
• Storage and expansion needs (how many SATA ports, M.2, PCIe slots do you need?)
• Network and USB requirements (how fast and how many?)

In the end, most of the cheapest boards aren’t inherently ‘bad’. They’re just stripped down and lack some extra features that higher-end models offer. However, cheap boards can become a problem if you pair them with hardware that's high-end; they’re not really designed to handle it.

If you choose wisely, you can cut costs on the motherboard without hurting performance and use the savings for upgrades that make a bigger difference. The goal isn’t to spend as little as possible (although that can work in specific situations) but to spend wisely and get the most out of your investment, no matter what the cost. If you need some help, we’ve picked out the best motherboards we’ve tested and are keeping an eye out for the best motherboard deals, too.

OpenAI is developing its own code repository platform as an alternative to Microsoft's GitHub, according to a report from The Information. The project was prompted by a rise in GitHub outages that left OpenAI engineers unable to commit or collaborate on code for stretches of up to several hours, two people working at large GitHub customers told the publication.

The project is still in early stages and probably won't be completed for months, a person with knowledge of it told The Information. Employees working on it have discussed the possibility of selling access to the platform to OpenAI customers — though the company could ultimately keep it exclusively for internal use.

GitHub's reliability has degraded noticeably over the past year, following an overhaul of its infrastructure. GitHub CTO Vladimir Fedorov told employees in an October memo that the platform would migrate all of its software to Microsoft Azure within two years, calling the move "existential" to meet the demands of AI-powered tools such as GitHub Copilot.

Platform migration began in October 2025 and is still in progress, which means GitHub is running in a split-traffic state across its legacy Virginia data center and Azure. Multiple recent outages have been attributed either directly to Azure or to configuration issues introduced during the migration. In early February, a four-hour outage was traced to an underlying Azure problem. A separate outage a week later, which took down many GitHub services for around three hours, was attributed to a configuration change — with GitHub acknowledging in a public incident report that its availability was "not yet meeting our expectations."

GitHub reported a 58% year-over-year increase in incidents during the first half of 2025, rising from 69 cases to 109 — with 17 classified as "major" — totaling over 100 hours of disruption, according to a mid-year report from GitProtect.

If OpenAI does sell the platform commercially, particularly bundled with its Codex coding agents, it will mark a direct competitive move against Microsoft. Microsoft currently holds roughly 27% of OpenAI, and acquired GitHub in 2018 for $7.5 billion. OpenAI has already encroached on other Microsoft territory: It's reportedly developing ChatGPT features that overlap with Office applications for document collaboration and presentation editing.

Building internal code repositories is not unusual for large tech companies. Google runs Piper, and Meta runs Sapling — though neither has been released as a commercial product. An OpenAI commercial offering would be a different proposition — though losing OpenAI as a customer would be mostly symbolic for GitHub, given its tens of millions of paying users, according to the source from The Information.

With billions being invested into AI and the infrastructure around it. The industry has picked up a breakneck pace ever since the popularization of ChatGPT several years ago. Now, the entire semiconductor industry is seemingly revolving around skyrocketing demand for AI data centers. The question on everyone's lips: Are the models good enough to make a material impact, and what risks come with using AI?

Machine learning technology has certainly helped make strides in many areas of industry and research. Voice recognition is far more reliable, medical analysis is faster and more accurate, materials science is quickly evolving, and even weather prediction and climate tracking are seeing massive strides, thanks to the ability of bots to vastly speed up or add precision to processes performed by humans.

Despite this, many analysts have expressed skepticism about the ability of conventional LLMs (text, code, and agentic bots) to advance much further, though, and even some CEOs have publicly expressed their reservations. The main issues that LLM models face are threefold: hallucination, where an AI makes things up; knowledge uncertainty, when a bot doesn't know something but is unaware; and overconfidence in answers, when a bot is highly confident of something that's blatantly incorrect in its reasoning.

An image is worth a thousand words; the limitations on image and video generators are quite obvious: signs with garbled text, hands with a variable number of fingers, and impossible architecture. Despite how bots have advanced, the lack of trust in their output is likely the biggest roadblock for any one player to stand out from the rest of the pack.

Is AI really getting better?

And yet, anyone who's lived through the past few years has witnessed the almost-monthly improvement across every front: ChatGPT keeps getting smarter and doesn't forget context as easily, Perplexity digs information ever more effectively, Midjourney no longer creates six-fingered humans, and video generators like Sora don't defy basic physics so often. Gigantic disasters can and do happen due to over-eager, agentic bots, but the error rate is being reduced by the day, and the number of guardrails continues to grow.

Anthropic's CEO said that AI could cause up to 20% of unemployment in the next five-years, and Microsoft's ongoing ceaseless charge to integrate Copilot into every facet of its OS means that AI is inescapable for the average user. So, if AI is going to be everywhere, what makes it tick, and what factors could improve a given model?

To understand that, we must break down what makes AI function, and what could make any given model better. After all, the models' outputs need to become more trustworthy and/or of higher quality than a common bowl of digital slop.

How LLM's work

To that end, LLM-based models (both text and agentic) are expanding their reasoning capabilities and reducing the hallucination rate. This is achieved in several ways, but one common theme among all the latest versions of popular models is extra-large context windows and hundreds of billions, sometimes trillions, of parameters.

Context windows for LLMs are measured in tokens (words, fragments, or symbols) and grew from around 512 tokens in 2018 to over 1 million in the current-generation models, an improvement of over 2,000x over just 7 years. Larger windows give the model a bigger workspace to formulate its response, enabling much more detailed "thinking," better conversation memory, contextual awareness, and the ability to consult additional data like webpages, documents, and even entire code repositories.

A larger window doesn't imply a model is smarter, but it is necessary to support more advanced reasoning, particularly multi-step reasoning and multi-modal reasoning (more on those below). Image and video generators don't use context windows per se, and their tokens are instead pixels and movement vectors, but the respective analogs to context windows enable the much-improved final rendering quality we see these days, as they're able to consult more images/videos as source material.

Parameters are values in the model that lend more or less weight to certain connections between their training information, like relationships between words and facts. Having more parameters generally allows models to capture more complex, interconnected information, though increasing the number also increases the cost of running queries. A high number of parameters is essential for research-grade models, while simple search/classification engines will be fine with "only" a few billion.

Multi-modality is also one of the lynchpins of contemporary models of various types. The advancement means that models consider not just text (or pixels for images, or vectors for video) when generating their output. For example, chatbots now know to read images, charts, code, and even videos, and use them as references in their replies when formulating and answering your queries. Retrieval-Augmented Generation (RAG) is becoming commonplace, where a bot refers to and/or verifies its information using external information it looked up.

Conversely, visual generators can rely on textual information to better understand prompts (prompt adhesion), provide captions, and cross-reference information. One particularly neat trick is "zero-shot learning," in which the model infers what a certain animal (say, a lion) is and generates a picture of it, having obtained information from textual context and description rather than being specifically trained on images of lions.

Multi-step reasoning is another feature you might have noticed about some bots, but it is quickly becoming commonplace. It's probably the closest analog to human reasoning: a bot breaks down a task or question into separate parts, effectively using most of its brainpower for each step and evaluating the results before moving on. You might even have noticed some bots backtracking on their footsteps when hitting a dead end, just like humans would.

This type of reasoning is powerful, but since it takes a long time to compute, it's generally reserved for premium usage plans. Models like Anthropic's Claude are particularly adept at multi-step reasoning, having been designed with development tasks in mind, even going as far as saving its "state" to files for better handling long-term tasks. Most, if not all, contemporary models have "fast" and "thinking" modes of operation.

Tool use is quickly becoming critical. Almost by definition, a repetitive task should be automated by a computer, and to that end, a model needs to integrate with and use APIs for commonly available tools. As examples, Google's Gemini can interact with most of the Google Workspace ecosystem, while Anthropic's Claude made a living from day one as a coding assistant, integrating with many developer tools. Anthropic is also testing how LLMs run entire businesses, with mixed results. ChatGPT also has a plug-in system of its own. In effect, these models can now interact with these services just as well (or much better) as any human.

Training set sizes. Any bot of whichever type is only as good as the data it's trained on. This characteristic's evolution is fairly predictable, given that it's mainly limited by the capabilities of the underlying hardware, and that too has seen massive leaps in under a decade.

For an LLM, the average training set size was around 13 billion tokens in 2018, and the amount is now estimated to be well over 20 trillion. Image generators were initially trained on less than 10 million images, a stark contrast to the multiple billions of today. Videos take up a lot of space and RAM, and early generators made do with under 1 million videos evaluated, while today they analyze billions of clips.

All combined, the techniques detailed above help lower hallucination rates, make for "smarter" bots overall, that are capable of executing more tasks than before. Answer accuracy is improving all the time, and the agentic bots are also much less prone to making boneheaded decisions when manipulating their respective tools.

Trust in a bot's output or operations includes a concept of safety — not just in the politico-social sense of defining what information is safe for a bot to provide, but also the relative safety of its operations when using tools. After all, it's not ideal for your bot to suddenly email everyone in your contact list because it misinterpreted an exclamation, executed irreversible operations on a batch of images you want touched up, or cleaned up your thesis's formatting by removing all the content.

Safety is a fairly hot topic right now, given the growth of agentic and tool-based AI. Grok has been under the microscope for safety in particular, as legislation begins to surface as a result of AI's ease-of-use.

Each vendor has its own mixed set of approaches to this topic, called "guardrails." Safety is, however, a trade-off, as some models will be far more conservative than others when answering questions or executing tasks and can err too much on the side of caution, refusing to answer innocuous questions. Generally speaking, the more capable they are, the more careful they tend to be. After all, with great power comes great responsibility.

Highlights of popular models

The characteristics and improvements described above generally apply to most any contemporary, full-sized model, but here are a few key highlights from each vendor:

GPT 5.2 (OpenAI): The newer version of OpenAI's flagship model claims to have a much lower hallucination rate (37%, down from 62%) and should be up to 10x more computationally efficient, as well as have much-improved response quality, whether on text or code. It's now fully multi-modal and can interpret images, video, and audio to formulate responses. It's also capable of using real-time information.

Although it's a generalistic model at its core, its plugin architecture allows it to be integrated almost anywhere, serving as easily as a browser search or a coding assistant. ChatGPT is also customizable with custom instructions and has multiple personalities available, letting the user select the desired style and tone for responses. However, when GPT-5 was initially released, some users were not happy with its outputs.

Gemini 3 (Google): Released in late 2025, although Gemini 3 is a generalist model, equipped with Deep Think architectures that allow it to plan, pause, and self-correct before responding. Google claims the multi-step reasoning improvements let it top benchmarks in coding and reasoning tasks. It's natively multi-modal, taking in common types of digital media and code repositories as inputs. Users of the Google ecosystem (Gmail, Chrome, Workspace, etc) can benefit from Gemini's tight integration with those services.

There are also Gemini Gems, shareable chatbots that you can tailor for specific tasks. Google's AI Studio ought to make it easy for developers to integrate Gemini into their applications, too. Google's Antigravity platform also allows users to expand on its abilities for bigger tasks, but it doesn't quite stick the landing. In one infamous example, one of Gemini's agents wiped a user's entire HDD.

Claude 4.5 (Anthropic): Claude has been designed as a model for programmers from the get-go, so it's no wonder that it claims to be optimized for multi-hour tasks and scores particularly well in coding and reasoning benchmarks. It excels at complex operations and uses hybrid reasoning (a mix of fast and accurate reasoning modes), and is naturally well integrated with GitHub and other development tools, being capable of using several in parallel.

All Claude 4.5-based models are multimodal and multilingual. Anthropic prides itself on designing Claude with a safety-first approach and particularly strong guardrails, with the model reportedly scoring quite high on safety tests. That's a particularly welcome feature in a bot whose main output is code, which intrinsically needs to be scientifically correct. Interestingly, Claude can write its "state" to files if given access to, letting it improve its continuity on long-term tasks.

Grok 4.1 (xAI): Grok 4.1 is one of the most powerful AI models on the planet, and that's due to its multi-modality, high two-million-token context window, and reasoning capabilities, built on a MoE (Mixture of Experts) architecture, in which the model activates specialist parts of itself to answer a question rather than its entirety, making for faster answers and more efficient computing while retaining answer quality. This has led to the Elon Musk-led company's flagship thinking model excelling in various benchmarks, including text generation and search in particular.

Unlike other models, like GPT-5 and Claude, Grok 4.1-thinking exists on a real-time data set, which may give it an advantage, as it has a later knowledge cutoff. While safety is an issue on Grok imodels in particular, it excels in thinking and reasoning.

Mistral Large and variants (Mistral AI): Mistral has the Mistral Large model as its flagship offering (released in 2024), but the company focuses on offering multiple variants for integration into products and services, each optimized for a particular type of task and/or desired computing efficiency. As examples, Mixtral uses a mixture-of-experts, Codestral and Devstral are targeted at development services, Pixtral and Voxtral handle visual and audio recognition, and Magistral excels at reasoning.

Many of Mistral's models are published as open-weight models under the Apache 2.0 license, while generally the higher-end variants require a commercial license. They're generally better thought of as models-as-service; Mistral doesn't have many end-user applications compared to other models, like ChatGPT.

Where AI is headed next

At this point, you may be asking yourself what's beyond "models keep getting smarter". In the short term, that's definitely where all the low-hanging fruit is, enabled by Nvidia and AMD's technological advancements with their respective accelerators, plus all the investment in AI data centers. Though TSMC itself is reportedly 'very nervous' over an AI bubble.

In AI, optimization is also paramount, as Total Cost of Ownership (TCO) is king for an AI data center, due to the power-guzzling nature of the tasks at hand. Any optimization is welcome, and for example, a few years ago, it would have been difficult to predict that a data format like FP4 (4-bit floating point) would ever become useful. Now, Nvidia is spinning off its own standard, NVFP4.

The first endgame goal is for AI to become deeply integrated into software ecosystems. From web- or device-based applications, to operating systems. A good portion of the internet and devices are already dependent on cloud services like Amazon Web Services (AWS), Azure, et al.

AI services will soon be no different — as their APIs and models get integrated into every single bit of software, in the medium-term, a good portion of the computing world will cease to function without them.

For example, almost every application has a search function of some sort, something that AI bots are particularly adept at. Yes, on-device AI is widespread, but much like it happened with cloud service providers, the convenience and ease of development of using an external API will trump almost everything else, implicitly sending out lots of your data for processing.

Agents and integrations

AI Agents set most of the scene for the future of AI. Theoretically, you can ask an agent to perform a task, and it will do it for you, feeding into a larger LLM, which is working on a larger task. However, the main issue for Agentic AI is trusting their actions, just ask the person who had their application's production environment wiped by Replit for no apparent reason. At least the bot was honest; not every employee is that forthcoming.

Getting developers hooked into using AI APIs in apps is one thing, but you can cut out the middleman if you are the app. OpenAI's ChatGPT's Atlas, Perplexity's Comet, and Atlassian's Arc are all browsers that put their respective services front and center, conveniently bypassing Chrome, Firefox, Safari, and other points of entry into the internet.

Being the internet's gatekeeper is an absolute position, as you have control over the user's eyeballs, can collect advertising money, and suggest, cajole, plead, and strong-arm users into using your services. Last year, Perplexity and Search.com put in offers to buy Chrome from Google to the tune of $35 billion, a deal that ultimately didn't go through.

A revenue stream selling the abilities of your bots is all well and good, but trading in user data is the business gift that keeps on giving. The amount of data that conventional services already know about people is already staggering, but with heavy AI usage, it may elevate itself to another level.

AI's privacy problem

The issue is twofold: firstly, people have long, in-depth conversations with LLMs, where they provide lots of personal details, rather than just a handful of Google searches. Secondly, once you grant a bot access to your data or services, there's little more than a Terms of Service statement stopping it from siphoning it all away. Many developers might not even be aware of just how much of the user's data is traveling through their app and being sent elsewhere.

Chatbot logs have already been used in court multiple times, and their much longer and detailed nature makes them far better proof of conditions or intent than simple search terms. At one point, an AI bot (or all of them) may well have a better insight into your life and patterns than you do yourself. Such detailed information is worth a lot of money to the right bidder, and the amount, accuracy, and price of said information are all likely to rise.

AI companies like OpenAI are planning to go one step further and make their own devices. It's not that difficult to imagine that at some point, OpenAI or Meta might release their own smartphones where everything is AI-centric, and intimately know each byte of your documents. The Ray-Ban Meta Glasses may have interesting utilities, but it's a chilling awareness knowing that one day, AI might be watching and parsing every inch of it.

All told, there might not be one grand unifying vision on AI companies, but one thing is fairly certain: they're all looking, and will likely become fully entrenched in your professional and personal lives.

OpenAI CEO Sam Altman announced late Friday night that the company had reached an agreement with the U.S. Department of Defense (“rebranded” as the Department of War under the current administration) to deploy its AI models on the Pentagon's classified network, with the same two safety conditions Anthropic was effectively blacklisted for insisting on: no domestic mass surveillance, and human oversight of decisions involving lethal force and autonomous weapons.

"Two of our most important safety principles are prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems," Altman wrote in a post on X. "The DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement."

Altman’s announcement came not long after President Trump “ordered” every federal agency to immediately stop using Anthropic's technology, following weeks of tense negotiations between Anthropic and Pentagon officials that ultimately collapsed. The DoD had labeled Anthropic a supply chain risk and demanded that it drop restrictions on its Claude model, requiring the model to be available for "all lawful purposes." Anthropic refused. Hours later, the Pentagon accepted functionally identical conditions from OpenAI.

It’s understood that no formal contract between OpenAI and the Pentagon has been signed yet, and that the agreement also limits OpenAI's deployment to cloud environments, not edge systems such as aircraft or drones.

Anthropic argued that the law hasn't kept pace with what AI can do, particularly in aggregating publicly available data for surveillance purposes. Altman seemed to agree with this, stating in an internal memo to OpenAI staff that it shares Anthropic's "red lines" and wanted to help "de-escalate" the situation.

By Friday afternoon, however, he held a company all-hands meeting, telling employees the deal was taking shape. Around 70 OpenAI employees have separately signed an open letter titled "We Will Not Be Divided" expressing solidarity with Anthropic.

Anthropic was the first AI lab to deploy its models on the Pentagon's classified networks, through a partnership with Palantir. OpenAI had previously held a $200 million DoD contract for non-classified use cases. Anthropic said Friday it will challenge the supply chain risk designation in court, stating that "no amount of intimidation or punishment from the Department of War will change our position."

OpenAI announced this week that it has closed a $110 billion funding round backed by Amazon, Nvidia, and SoftBank — the largest private tech financing in history — valuing the ChatGPT maker at $730 billion pre-money, or $840 billion including the capital raised. Alongside the cash, the company secured major infrastructure commitments tied to Nvidia's next-gen Vera Rubin GPU architecture and a dramatically expanded partnership with Amazon Web Services.

The breakdown is $50 billion from Amazon, $30 billion from Nvidia, and $30 billion from SoftBank, with additional investors expected to join as the round progresses. Amazon's initial commitment is for $15 billion, with the remaining $35 billion contingent on unnamed conditions being met in the coming months. The new valuation marks a significant jump from OpenAI's $500 billion in secondary financing last October, and more than doubles the $40 billion raised last year, which itself was a record at the time.

Under the terms of Nvidia’s stake, OpenAI has committed to using 3 gigawatts of dedicated inference capacity and 2 gigawatts of training on Vera Rubin systems, Nvidia's successor to the current Blackwell architecture. That capacity is in addition to the Hopper and Blackwell systems that OpenAI already operates across Microsoft Azure, Oracle Cloud Infrastructure, and CoreWeave.

On the Amazon side, OpenAI is expanding its existing $38 billion AWS compute agreement by $100 billion over the next eight years and has committed to consuming at least 2 gigawatts of Amazon's proprietary Trainium AI chip capacity. AWS also becomes the exclusive third-party cloud distribution channel for Frontier, OpenAI's enterprise platform. The two companies are additionally developing a new "stateful runtime environment" that will allow OpenAI models to run natively on Amazon's Bedrock platform. According to Amazon CEO Andy Jassy, this will "change what's possible for customers building AI apps and agents."

OpenAI and Microsoft issued a joint statement Friday confirming the Amazon deal does not change their existing arrangement. Azure remains the exclusive cloud provider for OpenAI's APIs and first-party products, and Microsoft retains its exclusive license to OpenAI's intellectual property.

To justify the scale of investment, OpenAI cited its current user figures: more than 900 million weekly active ChatGPT users, over 50 million paid consumer subscribers, and weekly Codex users, which have more than tripled since January to 1.6 million.

“We are entering a new phase where frontier AI moves from research into daily use at global scale. Leadership will be defined by who can scale infrastructure fast enough to meet demand, and turn that capacity into products people rely on,” reads the official press release, which goes on to explain that this round of funding will enable the company to do both and ensure “AGI benefits all of humanity.”

Amazon and OpenAI have announced a sweeping multi-year strategic partnership, with Amazon committing $50 billion in investment, AWS securing exclusive third-party distribution rights for OpenAI's enterprise agent platform Frontier, and OpenAI agreeing to consume approximately 2 gigawatts of Amazon's custom Trainium compute capacity, according to a press release.

The investment is part of a $110 billion funding round that values OpenAI at $730 billion pre-money, with Nvidia and SoftBank each contributing $30 billion. The companies will also co-develop custom AI models for Amazon's own products, including Alexa, and jointly build a new stateful agent runtime on Amazon Bedrock.

Split between Trainium 3 and 4

Amazon's $50 billion commitment is structured in two parts: $15 billion upfront, with the remaining $35 billion contingent on conditions that, according to sources cited by The Information, may require OpenAI to complete an IPO or reach an as-yet-undefined "AGI milestone." The deal also expands OpenAI's prior $38 billion AWS compute agreement, struck in November 2025, by an additional $100 billion over eight years.

Meanwhile, the 2 gigawatts of Trainium compute will span both the current Trainium 3 generation and the upcoming Trainium 4, which is expected to ship next year. Amazon launched Trainium3 — a 3nm chip delivering four times the performance of its predecessor at 40% better energy efficiency — at its re:Invent conference in December 2025.

AWS has stated that customers can achieve cost savings of 30 to 40% running training and inference workloads on Trainium compared to equivalent Nvidia GPU configurations. Each Trainium3 UltraServer holds 144 chips, and up to 1 million of them can be linked in a single cluster.

Trainium4, meanwhile, is being designed with support for Nvidia's NVLink Fusion interconnect, which allows Trainium4-based systems to interoperate with Nvidia GPUs within the same server rack. Nvidia's CUDA software stack remains the de facto standard, which nearly all large AI workloads are built on, and migrating away from it means rewriting significant portions of a codebase.

Anthropic, in which Amazon has invested at least $8 billion, already trains its Claude models on Trainium at scale — Project Rainier, Amazon's largest dedicated AI data center, houses more than 500,000 Trainium2 chips running Anthropic workloads exclusively. But Anthropic is financially entangled with Amazon. OpenAI is not, which makes its decision to commit 2 gigawatts to Trainium a notably independent validation of the platform.

Let’s not forget that OpenAI also has a separate deal with Broadcom to develop its own custom ASICs, uses Nvidia GPUs through both Azure and AWS, and has committed to AMD chips — so, again, its willingness to stake 2 gigawatts on Trainium is a weighty decision.

Stateful Runtime Environment

Beyond compute commitments, Amazon and OpenAI have announced that they’re co-developing a so-called “Stateful Runtime Environment” (SRE) built on Amazon Bedrock and expected to launch within the next few months.

Most AI agents run on Retrieval-Augmented Generation (RAG) architectures that effectively use a model as an advanced search engine over a set of embedded documents. The issue with this architecture is that the agents can’t retain memory between sessions or carry context across different software tools, and they reset with every new interaction.

SRE, Amazon says, keeps context across calls, retains memory of prior work, integrates with AWS data sources, including S3 storage and IAM identity controls, and allows agents to operate persistently across ongoing projects rather than treating each call as isolated.

Frontier, OpenAI's enterprise agent platform for building and deploying coordinated AI agent teams across business systems, will be distributed exclusively through AWS as its third-party cloud provider, and the SRE on Bedrock is where that infrastructure will sit.

Microsoft-OpenAI partnership ‘strong and central’

Several initial reports following the announcement have framed the deal as AWS displacing Microsoft's position with OpenAI, but that’s not accurate. Azure, per Microsoft, “maintains its exclusive license and access to intellectual property across OpenAI models and products,” with Azure remaining the exclusive cloud provider for OpenAI's stateless API calls.

Microsoft also retains the option to participate in the current funding round, with both companies issuing joint statements affirming the partnership remains "strong and central." In terms of the Amazon-OpenAI deal, AWS gains the enterprise agent deployment side, while stateless API traffic stays on Azure.

AWS holds approximately 30% of the global cloud market heading into this announcement, compared to Azure's roughly 20% and Google Cloud's 13%. Despite that market position, Amazon had been widely characterized as trailing in the generative AI race relative to Microsoft's early OpenAI integration and Google's push with Gemini.

Exclusive distribution rights for Frontier, combined with Trainium's cost positioning, fit Amazon's consistent approach to cloud competition, whereby it prioritizes infrastructure scale and cost efficiency over all else. Amazon is now financially backing both of the leading independent frontier AI labs simultaneously — OpenAI and Anthropic — which positions it as infrastructure for the industry, regardless of which organization's models prove most durable commercially.

This comes with significant financial exposure for Amazon, which is spending approximately $200 billion in capital expenditure in 2026, the majority directed at data centers and AI infrastructure. Its stock had fallen about 8% on the year as investors weighed the return timeline on those outlays. Andy Jassy told CNBC today that he expected OpenAI to be "one of the very big winners" over the long term, but added that Amazon “still has a very strong relationship with Anthropic.” OpenAI CEO Sam Altman said the company now has more than 900 million weekly active users and more than 50 million consumer subscribers, and described an IPO as its "most likely path" given ongoing capital demands back in October.

The FTC issued subpoenas to Amazon, Microsoft, Google, OpenAI, and Anthropic in early 2024 to examine AI partnerships, with particular attention to exclusivity arrangements, and AWS's exclusive rights to distribute Frontier will no doubt give regulators a concrete point of focus. It’s still too early for a legal challenge, but it’s understood that the FTC’s investigation is ongoing, and these new terms could lead to a bite.

However, the $35 billion contingent tranche means a meaningful share of Amazon's headline commitment depends on a trigger — an IPO or AGI breakthrough — that comes with no known or guaranteed timeline. Until one of those conditions is met, the investment stands at $15 billion.

OpenAI spent much of 2025 trying to build its own AI data centers, only to find that it couldn’t secure financing on competitive terms. According to a report by The Information, that failure set off a cascading chain of negotiations and compromises that ultimately redirected ambitions down the stack. Rather than owning physical real estate, OpenAI reportedly pivoted to control what goes inside them while simultaneously assembling one of the most aggressive multi-vendor chip procurement strategies in the industry.

If you can’t build it, rent it

This is understood to have begun immediately following the White House’s Stargate announcement in January 2025. OpenAI employees fanned out across the country, scouting potential sites capable of supporting campuses between 800 megawatts and 1.2 gigawatts each, prioritizing locations where significant power would come online in 2026 and 2027. Executives reportedly floated spinning Stargate out as a separate entity that would construct facilities and lease them back to OpenAI, with some discussing using it purely as a financial vehicle to raise capital for chips and infrastructure.

Ultimately, none of this came to pass. According to sources cited by The Information, when OpenAI ran the numbers, it became clear the company would pay a significant premium to secure financing on its own. Lenders reportedly offered materially better terms when a more creditworthy tenant, like Oracle, signed the lease instead.

A blockbuster deal between Oracle and OpenAI followed this, with the duo agreeing to develop 4.5 gigawatts of data center capacity across multiple U.S. sites, with the two companies allegedly sharing economic risk on cost overruns and savings — a small but important detail that had not previously been publicly disclosed.

The Texas compromise

One particular site in Texas — a 1 GW facility in Milam County — was of particular interest to OpenAI. The company had reportedly hoped that the site would become its first self-built data center, while SoftBank, the other principal Stargate partner, wanted to develop and own it outright. Between September and October 2025, OpenAI's team made multiple trips to Japan to negotiate directly with SoftBank's Masayoshi Son, with talks reportedly stretching for hours across multiple sessions.

These meetings led to a compromise announced on January 9 of this year, when OpenAI and SoftBank each invested $500 million into SB Energy, with OpenAI selecting SB Energy to build and operate the Milam County campus. According to law firm Sullivan & Cromwell, which advised OpenAI, said in a statement that the deal “brings together OpenAI's first-party data center design with SB Energy's proven expertise in speed, cost discipline, and integrated energy delivery,” while OpenAI’s president Greg Brockman described the arrangement as combining SB Energy's "strength in data center infrastructure and energy development" with "OpenAI's deep domain expertise in data center engineering — in other words, SoftBank builds and owns the project, while OpenAI controls design.

According to The Information, design control covers cluster architecture, cooling systems, rack configs, and power infrastructure, four categories that together determine every meaningful hardware decision made inside a facility.

Control over cluster architecture means that it’s OpenAI, not SoftBank, that decides how GPUs or custom accelerators are grouped, how many form a single training or inference unit, and how they’re interconnected. So, while OpenAI doesn’t own the land or the physical building, it does have full say over all hardware decisions — that is, no doubt, what OpenAI wanted in the first place, even if the compromise meant that it doesn’t “own” the project on paper.

Late to the party

OpenAI has assembled a substantial silicon strategy since the Texas deal/compromise, with most of it formally confirmed. In September, OpenAI and Nvidia announced a letter of intent to deploy at least 10 gigawatts of Nvidia systems, with Nvidia intending to invest up to $100 billion in OpenAI as milestones are hit, with the first gigawatt targeting the second half of 2026 on the Vera Rubin platform.

That arrangement has since evolved: Nvidia is now reportedly moving toward a $30 billion direct equity stake in OpenAI, not tied to deployment milestones — as part of OpenAI's current funding round at a $730 billion pre-money valuation. As of December, Nvidia's CFO confirmed the definitive agreement had not yet been completed — and uncertainties remain — with OpenAI's purchases still flowing indirectly through cloud partners like Microsoft and Oracle. The actual figures, in other words, remain a work in progress.

Then there’s AMD, with whom OpenAI announced a definitive agreement in October. This covers 6 gigawatts of AMD Instinct GPUs, beginning with the MI450 series in the second half of 2026, with AMD issuing OpenAI a warrant for up to 160 million shares that vests as deployment milestones are reached. A week later, on October 13, OpenAI and Broadcom announced a term sheet covering 10 gigawatts of OpenAI-designed custom AI accelerators, with racks "scaled entirely with Ethernet and other connectivity solutions from Broadcom,” and in January, a confirmed $10 billion deal with Nvidia challenger Cerebras locked in 750 megawatts of Wafer Scale Engine 3 capacity through 2028 for low-latency inference workloads.

This split between training and inference makes sense because Nvidia’s GPU ecosystem remains extremely difficult to displace for large-scale model training, where CUDA's maturity introduces switching costs that can’t be eliminated. That’s not the case for inference, where Cerebras' wafer-scale architecture eliminates the inter-chip communication latency that constrains GPU clusters for latency-sensitive tasks, and custom ASICs highlight the same cost considerations that Google, Amazon, Meta, and Microsoft have already faced: at sufficient scale, the upfront cost of chip design is dwarfed by per-unit savings across hundreds of thousands of chips. Amazon, for example, claims 30% to 40% cost savings on specific workloads using Trainium versus equivalent Nvidia hardware.

The caveat is that OpenAI is arriving at this realization considerably later than its peers. Google began TPU development in 2013. Amazon launched Inferentia in 2018. Microsoft began its Maia program around 2019. Every company on that list will note that it’s not the chip, but the software stack that takes years to mature, and OpenAI is beginning that process now.

In November 2025, OpenAI confirmed the hire of Intel’s former chief technology and AI officer, Sachin Katti, to lead its infrastructure organization. According to The Information, his mandate is to develop OpenAI's data center intellectual property so future deals are built around the company's own hardware requirements. He reportedly oversees chip selection and the full compute roadmap, with the heads of data centers and industrial compute now reporting to him.

So, while it’s true that OpenAI still doesn’t own a single data center, it does have design authority over every campus it occupies, a confirmed custom accelerator program, production deployments on Cerebras hardware, and a hardware executive whose job is to close the gap, following the same path all the other hyperscalers have walked.

U.S. President Donald Trump announced in January 2025 that OpenAI, Oracle, and SoftBank would invest $500 billion and build several data centers across the U.S as part of their “Stargate” AI initiative. But instead of immediately supercharging the AI buildout, it devolved into a three-way tug of war between the three partners, according to a report by The Information.

OpenAI initially wanted to forge ahead on its own so that it owned the data centers. This would help it secure its own future without depending on third-party cloud providers, which can be more expensive in the long run. However, the company’s investors reportedly balked at the massive upfront costs required to construct AI infrastructure, especially as analysts are projecting that the company could run out of cash by mid-2027. This meant that it had to go back to the negotiating table and continue the discussions with its Stargate partners. This allegedly delayed OpenAI’s plan of acquiring 10GW of compute capacity over the next three years through its two partners.

Oracle was the first to secure a deal with OpenAI by the start of the latter half of 2025, when it said that it would build a massive Stargate data center with a projected compute capacity of 2 million chips. The AI developer then promised a few months later that it would purchase $300 billion worth of compute from the cloud provider over the next five years, although it’s still unclear how the two entities will be able to afford such commitments. Oracle did make two bond offerings in the third and fourth quarters of last year, but put itself in hot water with some investors after they complained about misleading statements during the initial offering.

These agreements do not include SoftBank, though. Instead, the AI developer and Japanese financial conglomerate were still locked in talks around the same time last year. OpenAI was already planning a 1GW data center in Texas, but this was put on hold in favor of the negotiations with Oracle. When it looked at the Texas site once more, it brought on SoftBank as a partner instead of going at it alone. This wasn’t a seamless transition, though, as the two parties vied for control. Sources say that the deal was only settled after a marathon session in Tokyo at the Japanese conglomerate’s headquarters. The two parties eventually settled on a compromise — SoftBank would get to own and develop the site, but OpenAI would control its design and would have a long-term lease on the facility. Despite that, SoftBank still had to pause its planned $50 billion acquisition of data center operator Switch due to regulatory hurdles.

After months of delays, it seems that Stargate is now back on track, and the three companies are making progress on several projects. According to The Information, this isn’t exactly ideal for OpenAI, as the best-case scenario for the company is that it owns and operates all of its own data centers. Unfortunately, that is out of the question in the near future because the costs are simply too much for an institution that has yet to reach profitability. Many investors still believe in its potential, though, with the AI firm hoping to secure $100 billion in its latest funding round and hit an $800 billion valuation.

OpenAI CEO Sam Altman took part in a wide ranging Q&A on Friday, answering dozens of rapid-fire questions during a 60 minute session hosted by The Indian Express. Not for the first time, Altman stoked controversy. This time, he bemoaned “unfair” comparisons between the efficiency of AI inference queries and human thought. In Altman’s view the comparison is skewed as humans have millennia of evolutionary smarts and technology teachings behind them, yet individuals require “like 20 years of life and all of the food you eat during that time before you get smart.”

Chief Nerd clipped the eyebrow-raising Q&A segment for convenient sharing.

In the above video segment, the AI business torchbearer begins by stating “One of the things that is always unfair in this comparison is people talk about how much energy it takes to train an AI model relative to how much it costs a human to do one inference query.” But, according to Altman, it also takes a lot of energy to train a human.

“It takes like 20 years of life and all of the food you eat during that time before you get smart,” the OpenAI CEO said to the assembled audience awaiting gems of wisdom. Moreover, Altman wants to roll in the “evolution of the hundred billion people,” and humanity’s progress to “not to get eaten by predators and learn how to like figure out science and whatever,” into the equation. If we did that calculation, Altman appears to reason, “probably AI has already caught up on an energy efficiency basis… Measured that way.”

OpenAI tech also evolved - from the minds and technological feats of humans

We see a few leaps in Altman's expanded-timeline human vs AI efficiency comparison logic, that need to be addressed. For example, shouldn’t the AI computing world also roll in the prior ‘energy cost’ of human evolution, the Renaissance, and so on? Aliens didn’t provide the blueprints for ENIAC.

Some commentators have also argued that Altman is dehumanizing by reducing childhood, learning, and growth to their energy inputs. Others even wonder if Altman would prefer to see resources diverted from human to machine intelligence.

However, beyond the confines of this Tweet clip, to give it more context and be fairer to the OpenAI boss, he also takes the time to push for more sustainable energy solutions. Tapping more into sustainable resources would take massive consumers like OpenAI a little more out of the firing line as scarce resource competitors, as folks’ utility bills inch higher and higher.

The above Q&A took place in the wake of Altman, and other AI high rollers, meeting with PM Narendra Modi during a highly publicized week that underscored India’s importance as an AI growth engine.

OpenAI may be on the cusp of securing a new stockpile of cash to burn through, with the business looking to secure $100 billion in funding from a range of tech and investment firms, Bloomberg reports. The company is reportedly seeking a valuation of around $850 billion, according to the outlet's sources. With $1.4 trillion in pledged expenditure over the next eight years, OpenAI has enormous commitments to meet. While OpenAI is also preparing for an IPO later this year, its plans have not gone without criticism, with some reportedly coming from Nvidia CEO Jensen Huang himself.

AI's circular strategy

The circular nature of AI investment and infrastructure deals has been apparent since the earliest months of 2025, and it played out throughout most of last year. Nvidia invested in companies that often bought Nvidia chips, while tech firms investing in AI developers often sold those same companies cloud computing capacity.

This latest tranche of funding for OpenAI appears to be much of the same. The key players in the deal are Amazon, Nvidia, Microsoft, and SoftBank, all of which have a mix of deals with one another and OpenAI already.

Amazon's investment is rumored to be up to $50 billion, with the condition of that investment being that OpenAI uses more of its chips and cloud computing services. SoftBank is reportedly considering a $30 billion investment, while Nvidia's is said to be $20 billion, a fifth of the total $100 billion alliance made in September 2025.

Although Microsoft's name was also thrown around by Bloomberg's sources, there was no speculation on what kind of financial contribution it would make to this round of funding. All companies involved are expected to finalize their investments by the end of February.

As with other major investment deals, even the rumored news of this one sent stock prices rising. AI industry sceptics would suggest this is one more example of why the AI industry is in a bubble: Investments that haven't happened yet generate financial returns for these companies, even without any clear path to profitability.

Projecting confidence

It's been a while since any of the big AI companies have made any enormous, headline-grabbing announcements. Indeed, a lot of the meta-story surrounding AI in recent weeks has been a crisis of confidence. Microsoft CEO Satya Nadella was seeming wobbled by the "slop" moniker, and bubble name-calling; Nvidia's Jensen Huang specifying the details of its $100 billion partnership with OpenAI, and an increasing number of studies suggesting AI isn't helping improve productivity much, if at all.

OpenAI is even projected to run out of cash entirely by 2027. But this week, the AI Impact Summit in India is being held, where all the major players in the industry are speaking. OpenAI's Sam Altman is there, as is Anthropic's Dario Amodei. OpenAI also released GPT 5.3 Codex, which is the first of OpenAI's deployments that marks a move away from being reliant solely upon Nvidia chips.

According to Bloomberg, sources close to the matter suggested that this is just the first phase of funding supplied by major names in the industry, with more to come. Once the major tech firms have signed on, OpenAI is reportedly looking to generate further investment from venture capital firms, sovereign wealth funds, and other financial investors, leading to its eventual IPO.

Although it's not explicitly stated, the fact that OpenAI is targeting these other sources after its industry contemporaries have taken the first step feels rather deliberate.

Forever chasing its tail

With significant spending commitments already booked in, OpenAI is clearly looking toward how it might fulfill the titanic amount of spending. At the AI India summit, it's still making more commitments for new data center deals, expending more money it doesn't yet have, on the back of confidence generated by deals that have yet to be finalized, or in some cases even officially announced. So, despite the latest funding round, reportedly backed by some of the biggest companies in the world, OpenAI still has a long road to having enough cash to deliver on its promises.

Peter Steinberger, the creator of OpenClaw, (AKA MoltBot and ClawdBot), has joined OpenAI with his AI assistant tool set to be maintained as part of an open source foundation, Reuters reports. OpenAI is keen to integrate some of the real-world functions of his open-source AI tool, leveraging AI agents to act as personal assistants to act out day-to-day busywork for the user.

“Peter Steinberger is joining OpenAI to drive the next generation of personal agents,” OpenAI CEO Sam Altman posted on Twitter/X.

“He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings.”

A high-profile hire

Steinberger is a high-profile get for OpenAI, and likely one envied by some of the AI developers' rivals. Steinberger's OpenClaw AI assistant tool became a sensation in AI circles in January, when users created a social media platform named Moltbook, just for the AI agents to post amongst themselves. Originally launched in November 2025, the tool allowed users to create their own AI smart assistants and integrate them with apps like calendars and email accounts, while allowing interaction and command prompts to be sent via chat tools like Telegram and Whatsapp.

Designed to be easy to use, OpenClaw is compatible with over 50 external services, can fill in forms, and execute scripts. It's designed to work without a dedicated app, but there are companion apps for macOS, iOS, and Android that add additional features like camera access and audio recording capabilities.

Through its mix of easy-access through local hardware compatibility and viral news stories about the Moltbook platform, OpenClaw generated significant interest. By the start of February this year, OpenClaw had created 1.5 million agents, with running costs quickly reaching $20,000 a month.

That's less likely to be a problem now. Although OpenAI or Steinberger has disclosed how much he's being paid, OpenAI isn't above spending big to acquire specific individuals, so he's likely been well compensated.

Unsurprisingly for an AI developer headhunted by one of the world's leading AI companies, Steinberger is bullish on the move. He plans to focus on developing agentic AI that are so simple "even [his] mum can use [them]."

“When I started exploring AI, my goal was to have fun and inspire people. And here we are, the lobster is taking over the world,” Steinberger said in a blog post.

Missing opportunities, dodging bullets

Although we'll have to see how much impact Steinberger makes at OpenAI, it's clear that Anthropic shot an own goal on this one. The tool was originally called ClawdBot, inspired by Anthropic's Claude. Steinberger even recommended users connect Claude Opus 4.5 to ClawdBot. However, their relationship soured. Instead of hiring the man making the viral AI tool, Anthropic threatened him with legal action.

“Anthropic asked us to change our name (trademark stuff), and honestly? Molt fits perfectly – it’s what lobsters do to grow,” Steinberger said of the news at the time. ClawdBot became MoltBot before he switched it to OpenClaw, deciding that the Molt name didn't really fit - though it did inspire the name of Moltbook when it was launched shortly after.

But while there's a chance Anthropic could have capitalized on Steinberger's success, popularity, and potential expertise on agentic AI integration, there's also a chance it's avoided a security and headline minefield. Security experts have warned that the deep integration of OpenClaw without robust security has left several glaring holes in its defences. These could reveal private conversations and allow for the theft of identities and personal information. Depending on what apps and services they give OpenClaw access to, there's the potential to allow near-complete access to your digital life without proper authentication.

Chasing relevance

The sceptical take on this new hire, though, is that it's just one more example of OpenAI chasing relevance. It might have the largest user base and the most impressive chatbot mindshare with ChatGPT, but Gemini is coming up fast with its own huge user base, and OpenAI's financials look increasingly precarious.

It launched Sora 2 in a big drive towards AI-driven social media, but interest in that tool has cratered since. Its Code-Red response to Gemini nipping at its heels didn't suddenly change its fortunes, and advertising in ChatGPT has gone down about as well as you might expect.

Can bringing in Steinberger make such a difference? It's hard to imagine OpenAI doesn't have the kind of developer talent it needs to develop an OpenClaw-like tool itself. Steinberger is supposed to have vibe-coded the first version of OpenClaw in about an hour.

Can't OpenAI do better? Perhaps now with him onboard they can. We'll have to see.

As for OpenClaw itself, it will remain an independent foundation that OpenAI will support. It's not clear how much focus Steinberger will give it, as OpenAI will likely be keen for him to focus on improving its agentic AI offerings instead. But forks and branches will be possible for anyone who wants to pick up the mandible mantle.

SoftBank swung back to profitability in its fiscal third quarter, reporting net income of 248.6 billion yuan — roughly $1.6 billion — for the three months ending December. These numbers, as reported by the Financial Times, represent a stark reversal from the 369 billion yuan (53.4 billion USD) loss the company posted in the same quarter last year.

This turnaround of the company’s fortunes was primarily driven by a $4.2 billion gain tied to its stake in OpenAI, which is part of a broader $17 billion in gains the company says it has recorded on the ChatGPT-maker in the financial year to date — strip out that contribution and the quarter would have looked very different.

Concentration by design

SoftBank has invested more than $34 billion in OpenAI and is understood to be in talks to commit another $30 billion in a funding round that could push its valuation to $750 billion. Following OpenAI’s restructuring last year, SoftBank holds roughly 11% of the company, compared with Microsoft’s near-27% stake.

To fund this, SoftBank has aggressively reshaped its balance sheet, exiting its entire position in Nvidia for $5.8 billion and selling $12.7 billion worth of T-Mobile shares. In doing so, Masayoshi Son effectively traded SoftBank’s exposure to public, liquid assets for a private AI platform and its associated infrastructure.

The company’s Vision Fund is reflective of why Son might have been keen to do this. Split into multiple parts, Vision Fund 1 recorded a $4.1 billion loss in the quarter, dragged down by declines in publicly-traded holdings like Coupang, which suffered a significant data breach. Meanwhile, Vision Fund 2, which includes OpenAI, generated a $6.6 billion gain, more than offsetting those losses.

The older portfolio, heavy with late-stage tech companies that are now exposed to public market volatility, is under serious pressure while the newer fund, anchored by OpenAI, is delivering gains at a scale large enough to dominate consolidated earnings. In effect, SoftBank’s quarterly performance is now increasingly tied to how OpenAI is valued.

Growing hardware strategy

In addition to its heavy investment in OpenAI, SoftBank also retains a majority position in Arm, whose CPU architectures underpin much of the world’s mobile computing and an expanding share of the server silicon market. Arm is also known to be interested in manufacturing its own chips.

Over the past two years, SoftBank has also acquired Ampere, the Arm-based server CPU company founded by former Intel executive Renee James, and Graphcore, the UK AI accelerator startup that once claimed its Intelligence Processing Unit could be an alternative to GPUs in AI training and machine learning applications. SoftBank has said that these companies are now housed under a new unit — the AI Computing Segment — alongside Arm.

Anyone looking at this collection of assets can easily come to the conclusion that it resembles what appears to be a developing hardware strategy. Arm’s designs are already central to hyperscale data center applications, and Ampere’s server CPUs target cloud-native and AI workloads with high core counts and power efficiency. Graphcore, despite facing commercial headwinds, developed architectures purpose-built for machine learning. Overlay OpenAI’s model development needs on top of that, and you’ve got everything you need to support AI training and inference at scale. In fact, reports claim that Arm is developing a custom CPU to be used by OpenAI itself.

Cost of scaling AI

OpenAI’s valuation is tied to growth in both model capability and enterprise adoption — and both require infrastructure. Training frontier models demands clusters built around high-performance accelerators and high-bandwidth memory, and serving them at a global scale requires data centers with robust power delivery and cooling capacity.

SoftBank is playing a leading role in that buildout, with Son leading financing efforts for the $500 billion Stargate project — antitrust challenges ongoing — in the United States, which is aimed at scaling AI data centers and infrastructure in partnership with OpenAI, Oracle, and Abu Dhabi-backed DGX. It’s impossible to understate the capital intensity of such an undertaking, which involves land acquisitions, grid interconnections, power contracts, and vast quantities of silicon.

SoftBank’s shares have nearly doubled as a result over the past 12 months as investors seek exposure to OpenAI through the Japanese conglomerate. This has effectively turned the company into a publicly traded proxy for private AI valuation. However, the stock price has fallen 29% from its October peak, reflecting growing concerns surrounding the so-called AI bubble and intensifying competition from the likes of Google and Anthropic, and, naturally, the enormous, seemingly endless amounts of capital needed to sustain frontier model development.

As enterprise buyers grow increasingly sensitive to pricing and competition heats up from Anthropic’s Claude and Google’s Gemini models, which themselves are pushing performance boundaries, OpenAI’s ability to maintain competitive pricing will depend on continued technical leadership and efficient deployment of compute resources. That dynamic creates risk for SoftBank, given that a large portion of its recent profit derives directly from mark-to-market gains on OpenAI. If investor sentiment around AI reprices downward, those gains could be wiped out as quickly as they appeared.

SoftBank CFO Yoshimitsu Goto said that the group has made “over $40 billion of investments in the last nine months,” including commitments to OpenAI, while maintaining net debt at roughly 20% of the value of its equity holdings and a cash position of $24 billion. That, at least on paper, suggests some degree of financial discipline.

OpenAI on Thursday released GPT-5.3-Codex-Spark, its first AI model served on chips from Cerebras Systems, marking the ChatGPT maker’s first production deployment on silicon outside its long-standing core stack with Nvidia. The new model is a streamlined, lower-power variant of Codex designed for fast, interruptible coding tasks, and is initially rolling out as a research preview to ChatGPT Pro subscribers.

According to OpenAI, GPT-5.3-Codex-Spark is tuned for interactive development workflows such as editing specific sections of code and running targeted tests, and the model is optimized for high throughput when served on ultra-low latency hardware. The company claims it can exceed 1,000 tokens per second under the right configuration, while also defaulting to minimal edits, and will not automatically execute tests unless instructed.

The hardware behind all this is Cerebras’ third-generation Wafer Scale Engine. Unlike conventional GPU clusters built from many smaller chips connected over high-speed interconnects, Cerebras uses a single wafer-scale processor with hundreds of thousands of AI cores and large pools of on-chip memory. The architecture is designed to minimize data movement and reduce latency, which is often the bottleneck in interactive inference workloads.

OpenAI said last month that it had signed a deal to deploy Cerebras hardware for low-latency inference, and that it plans to bring 750 megawatts of Cerebras-backed compute online in phases through 2028. While that capacity will not replace Nvidia’s role in OpenAI’s training infrastructure, it gives the company a dedicated tier optimized for responsiveness rather than training.

Earlier this month, Sam Altman took to X.com to say that OpenAI loves working with Nvidia, and that “they make the best chips in the world,” adding, “We hope to be a gigantic customer for a very long time.” This came following a controversial report from Reuters that claimed OpenAI is unsatisfied with some Nvidia chips.

OpenAI has also described the partnership with Nvidia as “foundational” and said the company is anchored on Nvidia as the core of its training and inference stack, while also expanding the ecosystem around it through partnerships with Cerebras and others. OpenAI’s most powerful models continue to be trained and served on Nvidia systems.

OpenAI has also agreed to deploy 6 gigawatts in chips from AMD over multiple years and has also struck a deal with Broadcom to develop custom AI accelerators and networking components.

Codex itself now has more than 1 million weekly active users, according to OpenAI, and will expand beyond Pro users in the coming weeks as the company evaluates performance and demand.

Nvidia and OpenAI inked a memorandum of understanding (MOU) in September 2025, signaling the former’s intent to build data centers capable of delivering 10 gigawatts of compute power for the latter to lease, and to invest $100 billion in the AI company to help fund it. However, several months have passed since that announcement, and it seems the deal is going nowhere. OpenAI had hoped to seal the deal a few weeks after the initial agreement, but sources told The Wall Street Journal that the talks have largely stalled since then.

It seems both parties are rethinking the deal terms, although OpenAI could still receive tens of billions of dollars from Nvidia through an equity investment. According to the publication, Nvidia CEO Jensen Huang has been privately telling his peers that the $100 billion MOU isn’t binding and is still subject to change.

Furthermore, although Huang said that OpenAI is “likely going to be the next multi-trillion-dollar hyperscaler company,” he has also been quietly criticizing the AI company’s approach to its business. People familiar with the matter say that the Nvidia CEO is concerned about the company's lack of business discipline and competition from Anthropic and Google. An analyst actually seems to agree with the leather-clad CEO, predicting that the company could run out of cash by mid-2027.

Despite that, Huang still believes in financially supporting the maker of ChatGPT, partly because it’s one of Nvidia’s largest customers. Although it has other big clients like Oracle and xAI, OpenAI’s failure would still put a big dent in the AI chipmaker’s order book. Aside from that, Google uses its own TPU chip. In contrast, Anthropic uses a combination of Google’s TPU and Amazon’s Trainium chips, meaning Anthropic does not rely on Nvidia for its hardware needs.

It’s unclear how the supposed partnership between these two AI giants will proceed. While we might no longer see the $100 billion blockbuster deal between them, there’s still a chance that OpenAI will get a massive infusion of cash even as it seeks to soothe investor concerns about its future profitability ahead of its planned IPO.

Despite Elon Musk's announcement on Monday that xAI's Colossus 2 data center had reached a 1 GW scale, the supercomputer is not even close to that, a satellite image published by Epoch AI researchers reveals.

Based on 550,000 Nvidia Blackwell AI accelerators, xAI's Colossus 2 is advertised as the industry's first AI facility that consumes one gigawatt of power for AI inference and training. But for now, the data center codenamed 'Macrohard' purportedly only has 350 MW of cooling capacity — not nearly enough to cool down 550,000 Blackwell GPUs at full power, even in winter. As a result, Musk's Jan. 19 announcement may have been premature, to put it mildly. Epoch AI expects the supercomputer to reach 1 GW by May.

Interestingly, when Grok, xAI's AI bot, was asked about Colossus 2, it confirmed that the launch of the supercomputer may be phased. Furthermore, it recalled media reports claiming that xAI may be using unpermitted gas turbines for extra power.

At the pace that Colossus 2 is being equipped with cooling systems right now, the new supercomputer will become a gigawatt-class machine sometime in May, according to the research. Meanwhile, the machine was once predicted to use a million GPUs, and then Musk said that it could scale to 1.5 GW or even 2 GW when the time comes. When this could happen is not known because xAI needs to get enough AI servers, procure enough power, and then get cooling systems.

Even though xAI's Colossus 2 will hit its 1 GW milestone later than expected, it is still projected to be ahead of rivals from Amazon and OpenAI, according to a graph by Epoch AI. The company will have more resources for AI training, AI inference, and agentic AI workloads than its rival for some time.

Based on the graph's reference lines, the power consumption of the whole city of San Diego averages ~800 MW, Amsterdam consumes around ~1.6 GW, and the power consumption of Los Angeles is about ~2.4 GW, which puts modern frontier AI data centers in the same class as major cities. When fully equipped and ramped, xAI Colossus 2 at roughly 1.3 GW – 1.4 GW, would consume about 1.7× San Diego's power, slightly less than Amsterdam, and around 60% of Los Angeles.

This week, a blog post went up on the OpenAI website, broadly discussing the company's financial outlook. The write-up is authored by Sarah Friar, OpenAI's Chief Financial Officer, where she claims that there's a direct relationship between available computing power and revenue generation, and thus, adding more AI accelerators equals more money coming into the firm's coffers.

Friar's figures say that OpenAI's computing power grew threefold every year between 2023 and 2025, from 0.2 to 0.6 and 1.9 GW, respectively. Meanwhile, the firm's revenue purportedly followed the same pattern, at $2B, $6B, and over $20 billion at the end of last year.

Not only does Friar conclude that this growth pattern was never witnessed before at this scale, but she also goes on to state that if OpenAI had had even more computing power available, it would have "led to faster customer adoption and monetization," a strong claim in the face of the company's current finances. In the famous words of Jensen Huang: "the more you buy, the more you save."

Compute becomes currency

Friar goes on to make the point that in this scenario, compute ultimately transforms from a "fixed constraint" to a portfolio, or de facto currency. That assessment aligns with similar assertions from other executives highly invested in the field, such as Nvidia's Jensen Huang, former Stability AI CEO Emad Mostaque, Microsoft's Satya Nadella, and, of course, Sam Altman himself.

The missive proceeds to make predictions, remarking on the evolution of AI usage patterns over time and describing how agent-based workflows "move from novelty to habit." For Friar, this shows a pattern of predictability that "strengthens the economics of the platform and supports long-term investment." The firm also expects that "new economic models will emerge" as AI becomes entrenched in research-heavy fields. However, over half of CEOs currently report seeing little benefit from AI deployed in the workforce.

OpenAI's words may sound reassuring at face value, but there's no shortage of virtual ink spilled on how the AI market is, at least for now, somewhat comprised of a circular economy. One company's investment feeds the next one in a cycle that tends to have comparatively little external revenue entering the loop. Many analysts predict that OpenAI is a particularly high spender, as its current burn rate is billions per month.

The company's blog post, on the contrary, makes it sound like OpenAI is actually playing it really safe. In its own words, "capital is committed in tranches against real demand signals", letting the company "lean forward when growth is there without locking in more of the future than the market has earned." Those are strong statements, but ripe for analysis.

OpenAI and expenditure

According to investor data shared with the Financial Times last November, OpenAI's estimated 2025 expenditure was of $22 billion, or an average of $1.83 billion every 30 days. That wouldn't be of concern if revenue were in the same ballpark, except that sales for the period were reportedly $9 billion, meaning OpenAI actually lost $0.69 on every incoming dollar. Altman's planned IPO cannot come soon enough, then.

A cynical reader can argue that OpenAI's post exists to assuage investor fears — a form of damage control, or at the very least, reframing the situation as a controlled burn for future crops rather than a raging wildfire. The current AI economy is so tightly woven together that a loss of confidence in its largest player could bring down the house altogether.

Profitability for the firm is expected in 2030, but many analysts believe the money well will run dry well before that happens. Not only do expenditures still far outweigh income, but OpenAI's consumer market share saw a drop from around 90% in 2024 to 60-70% in 2025, a chunk seized by Google Gemini and Perplexity.

It's worth keeping in mind that none of the pure-play AI firms are known to be profitable yet — and that, unlike traditional companies, they have no other ventures to fall back on. Should one go bust, usually it'll have only data centers and IP for creditors to repossess. In November, JP Morgan called out climbing spend on the ongoing AI buildout, which Jensen Huang says might take up to 50 years.

Elon Musk's xAI is burning close to $1 billion a month, with only around $500 million in 2025 income to show for it. Anthropic (makers of Claude) seems to be playing the game more conservatively, expecting profitability in 2028. This displays a stark difference in financial approaches, with OpenAI and xAI clearly betting the proverbial server farm on future gains, while Anthropic moves in a comparatively steady manner.

To the AI firms' credit, though, it's not like any of this has happened in history, or at least in this manner. Estimates put the amount of money already earmarked up for AI datacenters up to 2030 at $7 trillion, a figure that's difficult to picture but can be put into words: enough to run the entire U.S. government for one full year, the combined valuation of Microsoft and Amazon, or 1.5 times the entire GDP of Germany.

With the number of players in the game, it's hard to make the argument that absolutely everyone is wrong about the viability of their investments. While it can be argued that OpenAI might be flying too close to the sun, there's also no precedent for such explosive growth; what appears to be extremely long-term bets now may well prove to be the highest-risk, highest-reward scenario in the tech industry.

Following Trump's calls for AI tech companies to 'pay their own way', and in the footsteps of Microsoft, AI giant OpenAI has announced it will do exactly that as it continues in its quest for AGI through its massive Stargate buildout. In a press release, the company stated that every Stargate site will have a Stargate Community plan. Using Trump's own words, the company committed to 'paying our own way.'

Stargate confirmed that it is well on its way to its targeted 10GW U.S. AI infrastructure expansion by 2029, and says that it is "well beyond halfway" in terms of planned capacity. As sites come online in places like Texas, New Mexico, Wisconsin, and Michigan, OpenAI said, "We are committed to working with communities to ensure that our Stargate campuses are built and run in a way that strengthens communities and demonstrates that we’re being good neighbors."

The impact of AI data centers on local communities has become an increasingly hot-button issue with the explosion in AI buildouts. Reports of vast increases in energy prices, water shortages, and other issues have clouded progress made by big companies, with energy consumption and cost a particularly pressing concern.

To that end, OpenAI says every Stargate site "will have its own locally tailored Stargate Community plan," which will be driven "by community input and local concerns."

The headline commitment, as mentioned, is that OpenAI says "we commit to paying our own way on energy, so that our operations don’t increase your electricity prices." Measures proposed include bringing new dedicated power and storage funded by its projects, to increasing energy generation and transmission resources, again paid for by OpenAI. The company listed example approaches, including "Funding the incremental generation and grid upgrades our load requires," planning with local utilities companies, and developing "flexible loads" to reduce consumption of data centers during peak periods or forecasted grid stress.

The company further stated aims to minimize water use and protect local ecosystems, use innovations in cooling water designs "that drastically reduce the water use compared to traditional datacenters," and invest in local jobs.

Announced exactly a year ago, Stargate is a $500 billion OpenAI project funded by the likes of SoftBank, OpenAI, Oracle, and MGX. Tech partners include Arm, Microsoft, Oracle, and, of course, Nvidia. It is a large part of OpenAI's thrust towards AGI, artificial general intelligence, a hallowed benchmark generally best described as AI that is generally smarter than humans. Only time will tell whether OpenAI can stick by these Stargate commitments, but signals from Washington and Trump's rhetoric make it clear that this is an administration priority. Trump warned AI data center builders that Americans must not "pick up the tab" for their power consumption.

Court files from Elon Musk’s lawsuit against OpenAI revealed that Greg Brockman, one of its co-founders, wanted to get the company out from the Tesla founder since 2017. According to a document from the United States District Court for the Northern District of California, posted on X, case discovery revealed that Brockman didn’t just want to remove Musk from OpenAI, but also tried to convert it into a for-profit company without him. In a publicly-posted rebuttal of some aspects of the journal, OpenAI has acknowledged that the journal is real. However, the firm's comments don't address the following excerpts.

“This is the only chance we have to get out from Elon. Is he the ‘glorious leader’ that I would pick? We truly have a chance to make this happen. Financially, what would take me to $1B?” Brockman wrote in his personal files revealed during the lawsuit. “Accepting Elon’s terms nukes two things: our ability to choose (though maybe we could overrule him) and the economics.”

In another extract, Brockman wrote, "can't see us turning this into a for-profit without a very nasty fight. i'm just thinking about the office and we're in the office. and his story will correctly be that we weren't honest with him in the end about still wanting to do the for profit just without him."

Expand the above tweet to see the court documents. The supposed revelation appears in documents related to Musk’s lawsuit seeking to stop OpenAI’s move to become a for-profit company. Elon Musk first sued OpenAI in early 2024 in a California state court, alleging that its plan to become a for-profit entity breaches its founding agreement. However, he dropped the case one day before hearings were supposed to start, only to refile the case a few months later — but this time at a federal court.

Relations between the two camps have been testy ever since the OpenAI founders wanted to make the firm for-profit. This has gotten to the point that when Altman called out Tesla for not being contactable via its reservations@tesla.com email address on X, Musk replied with, “You stole a non-profit.” Naturally, we will have to wait for court proceedings and the jury’s decision to determine if this is true.

According to the court files, “Brockman wrote after the meeting [with Musk to reaffirm OpenAI’s commitment to the non-profit structure] that the ‘conclusion is we truly want the b-corp. Honestly, we also want to get back to work, but it’s super clear how we get there.’ He also continued, ‘cannot say that we are committed to the non-profit, don’t want to say that we’re committed, if, three months later, we’re doing B-Corp, then it was a lie.” The document also revealed that Brockman did not like the situation and the “the true answer is that we want [Musk] out.”

With the way things are shaping up, this case seems set to be an epic court fight between two AI tech bros, with billions of dollars at stake. Elon Musk is reportedly seeking damages ranging from $79 billion to $134 billion, as well as an unspecified punitive penalty.

Given how quickly the evolution of AI has upended technology across the globe and is affecting various markets, it's nigh impossible to accurately predict where anything might be headed. There's no shortage of predictions, ranging from utopia to ultimate doom for established industries. An NYT columnist, however, has one specific bet: OpenAI will be destitute in 18 months in the wake of its AI endeavors.

According to an external report last year, OpenAI was projected to burn through $8 billion in 2025, rising to $40 billion in 2028. Given that the company reportedly predicts profitability by 2030, it's not hard to do the math.

Altman's venture projects spending $1.4 trillion on datacenters. As Sebastian Mallaby, an economist at the Council on Foreign Relations, notes, even if OpenAI rethinks those limerence-influenced promises and "pays for others with its overvalued shares", there's still a financial chasm to cross. Mallaby isn't the only one thinking along these lines, as Bain & Company reported last year that, even with the best outlook, there's at least a $800 billion black hole in the industry.

The financial guru contextualizes the situation adeptly, broadly stating that it's not a matter of whether end-user AI will become technologically entrenched, but rather whether the economics of developing it will make sense in the mid- to long-term.

The analyst points out that in theory, investors would "bridge the gap between the emergence of a great technology and eventual profits", except that many AI companies seem to be burning cash far faster than they can generate income. Mallaby remarks that the newcomers are in a much worse position than "legacy" companies like Microsoft or Meta, given that the old-timers already had money-making businesses before AI came about and can (literally) afford to wait out the necessary period until the clankers deliver the fruits.

According to him, the majority of people are using free services and have no qualms switching to a competitor once their usual bot adds ads or usage limits, a fact corroborated by the myriad options available right now for all sorts of tasks.

He sees this as a temporary problem for AI providers, though, as agentic AI becomes more entrenched in daily people's lives, it'll become harder to switch, as the bots should eventually have all your shopping preferences, aspirations, and emotional profile mapped out —perhaps even better than yourself.

Mallaby does praise OpenAI's CEO Sam Altman's dollar-attracting gravitational field that raised $40 billion in investment, an amount bigger than any other private funding round in history — even more than Saudi Aramco's $30 billion. The difference is that Aramco, along with other IPO'd enterprises, had a business model and profitability, neither of which OpenAI currently enjoys.

The AI financial ouroboros certainly looks poised to eat its own tail, but there's an argument that the ophidian will only lose its newer part. There would be some irony in the AI market losing one or more of the players that started it all.

OpenAI has signaled its intentions to become a major player in brain computer interfaces (BCIs). The scale of the firm’s first round of investment in Merge Labs, as it emerges from stealth mode, places it among the most heavily funded BCI efforts in the U.S., second only to Neuralink.

That’s because Merge Labs, co-founded by Altman, will be going forward with $252 million in its tech advancement war chest, reports Bloomberg. However, it admits there's a long road ahead.

OpenAI wasn’t the only contributor in this investment round, but it was the biggest. Another notable investor was Gabe Newell, co-founder of Valve, which owns the gaming storefront Steam. Newell’s hat is already in this ring with his own brain tech company, Starfish Neuroscience.

OpenAI’s interest in Merge Labs BCIs could result in further public sparring matches between two of the biggest personalities in tech. Altman’s Merge Labs will be making ripples in Musk’s Neuralink pond. However, their approaches to BCIs, as we currently understand them, are quite different. These differences will likely be pivotal to their relative successes.

The limited amount of Merge Labs' currently public materials confirms that the fledgling BCI outfit will be developing fundamentally new approaches⁠ to this technology.

“We believe this requires increasing the bandwidth and brain coverage of BCIs by several orders of magnitude while making them much less invasive,” explains a blog penned by the freshly uncloaked firm. “To make this happen, we’re developing entirely new technologies that connect with neurons using molecules instead of electrodes, transmit and receive information using deep-reaching modalities like ultrasound, and avoid implants into brain tissue.”

No to invasive implants, yes to AI operating systems

Merge Labs also claims that the most recent breakthroughs in biotechnology, hardware, neuroscience, and computing will be adopted. The resulting BCIs, according to the company, will be “equal parts biology, device, and AI,” mixed into an accessible form factor.

So, in brief, Merge Labs BCIs will contrast with Neuralink’s approach because they will avoid implants into brain tissue. The key will be whether the firm’s technology can achieve workable results from “AI operating systems that can interpret intent, adapt to individuals, and operate reliably with limited and noisy signals.”

Distant horizons

The $252 million investment in Merge Labs sounds like quite a gamble, as Bloomberg’s report suggests the money will effectively establish a research lab to fix the disadvantages of the non-invasive BCI route. In other words, the money raised appears to be for a pre-prototype outfit, not a product-ready company. Meanwhile, Neuralink is pretty deep into testing its BCIs with humans, as are various Chinese competitors.

Chinese tech companies ByteDance and Tencent are spending massive amounts on bonuses and salaries to attract and retain AI talent. According to South China Morning Post, the former has increased its budget for employee bonuses by 35% over the previous year, while also allocating 150% more for future salary increases. On the other hand, the latter is reportedly hiring senior staff from competitors, even offering double their current salary just to get them to move.

This is part of the ongoing trend of increasing competitiveness in the AI jobs market, with companies in both the East and West looking to hire and retain the most talented AI engineers and developers to gain an edge over everyone else. While these companies are spending a lot of money on their people, this pales in comparison to what Meta is allegedly offering.

OpenAI CEO Sam Altman said that the social media giant is enticing his employees with $100 million in bonuses and even bigger salaries, but none of them have taken the bait. One founder even claimed that he received a billion-dollar offer from Meta for four years, which he declined.

However, Tencent apparently succeeded where Zuckerberg has failed: the Chinese tech giant has recently hired Yao Shunyu, a former OpenAI researcher, as its chief AI scientist. He reports directly to its president, Martin Lau, and also heads the company’s AI Infrastructure and LLM Development units, reporting to Technology Engineering Group Head Lu Shan.

All this news of skyrocketing pay and bonuses might be a bitter pill to swallow for the rest of the tech industry, though. Over 100,000 individuals in the tech sector have already been hit by layoffs in the first half of 2025, with Intel taking the lead, cutting over 20,000 jobs as October 2025.

Increasing AI adoption is also expected to disrupt every other professional field, with an MIT simulation showing that 11.7% of U.S. workers are likely to be replaced by the technology, leading to a loss of $1.2 trillion in salaries in benefits across the board. This though has been echoed by other analysts and business leaders, with Anthropic CEO Dario Amodei and Ford CEO Jim Farley saying that it will wipe out half of entry-level white collar jobs in the U.S.

Because of this, some lawmakers are concerned about AI’s impact on our society. Sen. Bernie Sanders (I.-Vt.) has called for a halt to all AI data center construction while Elizabeth Warren (D.-Mass.) and two other Democratic senators have asked several big tech companies to explain how their power consumption has affected nearby communities. Nevertheless, it seems that AI development will continue surging forward, and unless something big happens that disrupts its progress (like the AI bubble bursting), we will likely see the disparity between the AI industry and other workforces to only grow bigger.

OpenAI is allegedly still working on adding ads to ChatGPT, with sources saying staff are discussing ways to bake them into the chatbot’s responses. According to The Information, the AI company is looking to create a new type of digital ad rather than simply copying what existing search and social media companies are running. This is possible because OpenAI can use historical chat data to serve ads that are highly relevant to users' interests.

“As ChatGPT becomes more capable and widely used, we’re looking at way to continue offering more intelligence to everyone. As part of this, we’re exploring what ads in our product could look like,” OpenAI told The Information. “People have a trusted relationship with ChatGPT, and any approach would be designed to respect that trust.”

Staff discussions on ad implementation have ranged from prioritizing sponsored content in the chatbot’s answers to adding a sidebar that shows ads related to the user's query. They’ve also considered showing them only when the conversation moves toward shopping or similar activities, or as a secondary step where ads are displayed only when someone clicks a link in ChatGPT’s results.

It’s been reported that OpenAI is shifting its focus away from ads, especially after CEO Sam Altman declared a ‘Code Red’ for the company following the latest version of Google’s Gemini, which outpaced ChatGPT in several benchmarks. Altman said that OpenAI needed to improve the AI chatbot’s personalization, speed, and reliability, and cover a broader range of topics, so the company is pausing work on all other projects to focus on these capabilities. However, it seems to be continuing progress on ChatGPT ads, despite the recent change in focus.

ChatGPT has three main revenue streams at the moment — subscriptions to ChatGPT Plus, Pro, and Business, API access for developers, and enterprise solutions. Aside from that, OpenAI said it will start earning revenue from non-paying users by 2026, projecting $2 per user annually, which will grow to $15 by 2030. Despite that, OpenAI has yet to turn a profit since its founding in 2015. Even though its annualized revenue hit $10 billion earlier this year, it’s still expected that the company’s operating losses will hit $74 billion annually by 2028. Nevertheless, investors continue to pour money into the company, even as some are starting to ask how its long-term profitability will look.

For comparison, Google’s ad business accounted for $237.8 billion in revenue in 2023, representing 77% of the company's total revenue. This amount is more than enough to cover OpenAI’s estimated losses, and it seems it wants to follow the search giant’s playbook by baking ads into its results as well. However, this also raises privacy concerns, especially since ChatGPT likely has much more information about its users than Google does. Furthermore, there’s the question of how OpenAI will ensure its LLM gives the best answer to the user, especially if it stands to make money by showing ads instead of organic results.

The Michigan Public Service Commission (MPSC) has given DTE Energy permission to deliver 1.4 gigawatts of power to a planned data center that will be built in Saline Township, about 40 miles southwest of downtown Detroit. This data center project is part of the Stargate data center project by OpenAI and Oracle, which aims to deliver more than 5GW of computing capacity. According to the Detroit Free Press, DTE filed an “ex parte” motion with the MPSC to get the nod for its contract without contest, with the commission voting 3-0 for its approval.

This move meant that DTE Energy did not have to go through a lengthy hearing, wherein opposing groups could seek expert testimony and present evidence to challenge DTE’s claims. Many residents are concerned that this massive project would cause electricity prices to spike and affect the water quality in the area — issues that several U.S. senators are also looking into. On the other hand, the utility company argued that it didn’t need to prove that this project would affect energy prices for local residents, especially as a 2024 state law prevents data centers from passing on energy costs to the community to qualify for preferential tax treatment.

Aside from this, DTE’s contract with Green Chile Ventures, an Oracle subsidiary, requires that the latter will pay 80% of the contracted capacity, whether or not it actually uses it. This ready-to-serve provision in the power agreement helps ensure the power utility’s expenditure on additional infrastructure to support the AI data center will not be passed on to the average consumer. The contract will also last for at least 19 years, giving the power provider ample time to recoup its expenses from the data center and not from the community. Furthermore, if the data center fails to pay for its energy costs, DTE cannot pass them on to other consumers.

Despite these mechanisms, opponents of the plan were still dismayed about not being heard. “I, along with everyone here… probably don’t feel like they’ve been publicly served today. When you make these decisions, you’re eliminating the very necessity of what public service bodies are all about,” Wendy Albers, a resident of nearby Augusta Township, told the MPSC. “I appreciate the protections you put in place, but I just don’t trust them.”

Oracle has revised delivery schedules for several large AI data centers it is building for OpenAI from 2027 to 2028, Bloomberg reports, citing people familiar with the matter. The delayed facilities are parts of Oracle's commitments under the Stargate AI infrastructure program jointly announced in January by Oracle, OpenAI, and SoftBank

Bloomberg's sources familiar with the projects reportedly blame shortages of skilled labor and 'materials' for the setback, though it is unclear whether by 'materials' they mean building materials, shortages of data center equipment, or shortages of materials for building out infrastructure surrounding AI data centers. Despite the delay from 2027 to 2028, the overall scope of the projects that Oracle will build for OpenAI remains unchanged, the report says.

The delayed facilities are part of a contract between Oracle and OpenAI inked in July, under which the two parties plan to increase Stargate AI data center capacity to two million AI accelerators and 5 GW of power. Despite the revised timelines, the U.S.-based AI campuses are still planned on an unusually aggressive scale, and some of them are designed to rank among the largest data centers globally once completed. In fact, given the scope of the projects, delays and slipups are inevitable.

Earlier this week, Clay Magouyurk, chief executive of Oracle, said that despite obvious bottlenecks, Oracle believes its global expansion plans were realistic. He pointed to 147 active regions, 64 more in development, and roughly 400 MW of data center capacity delivered in one quarter. One of the examples of how Oracle can manage building out a large data center is its SuperCluster in Abilene, Texas, which houses nearly 200,000 Nvidia GPUs, and which was built in several months.

"Our SuperCluster in Abilene, Texas, is on track with more than 96,000 Nvidia Grace Blackwell GB200 delivered," Magouyurk told analysts and investors. "We also began delivering AMD MI355 capacity to customers this quarter."

The head of Oracle also stressed that the company continues to see strong demand for AI capacity, but emphasized that the company only accepts orders if it feels confident that it can fulfill them.

"Our pace of capacity delivery continues to accelerate," said Magouyurk. "We continue to see strong demand for AI infrastructure across training and inferencing. We follow a very rigorous process before accepting customer contracts. This process ensures that we have all the necessary ingredients to deliver to customer success at margins that make sense for our business."

Many of the world's largest AI tech companies are going to start working together on some of their shared problems. Microsoft, Google, Anthropic, OpenAI, and a number of other related companies are going to team up as part of the Agentic Artificial Intelligence Foundation, as reported by The Information. Managed by The Linux Foundation, the group will work on developing key open source tools and standards for AI agents, and will share their findings with each other on solving key technical problems.

However, as signs mount that agentic AI is not particularly effective at replacing workers, and rumors of the AI bubble stretching to its limits continue to swirl, agents need to impress, especially if they're being hailed as the next big thing in the AI landscape.

AI agents have long been seen as a next-generation development for the latest large language models that would finally realize their potential. They could laser-target larger tasks by breaking them down into smaller pieces, which a larger AI model could use to create or complete a larger project or goal.

That's how it works in theory, but as the Harvard Business Review highlights, they rarely achieve the end goal in practice. Especially when it comes to customer-facing roles, AI agents just aren't ready to replace real-world workers as they can't be trusted to complete their tasks effectively enough, or at all. Hallucinations are still a real problem, and the public has little tolerance for abject failure in basic tasks or wild shifts in tone.

That doesn't mean there's no potential there, though. It's the basket the main AI companies are putting their eggs in at the moment, anyhow, hence this new initiative to pool their efforts to create something more effective, and maintain standards that they have a greater say in developing.

The group's first goal will be to develop three existing open-source tools, according to people familiar with the matter. These include: a model context protocol developed by Anthropic called MCP, to standardize how AI agents connect to other applications; an OpenAI format for giving instructions to coding agents, known as Agents.md, and an open source AI agent invented by Block that can run locally on a single computer without networking, called Goose.

MCP is already in use at OpenAI, Microsoft, Google, and Cursor, so it's no surprise that it was chosen as one of the group's main goals. As it stands, it can connect ChatGPT to a company's Slack, for example, which would allow a manager to quickly summarize conversations. But IT managers speaking to The Information claim there are serious security concerns, especially when it comes to prompt injection attacks, so MCP needs continued development, and the developers need to agree on the best way to patch discovered security holes quickly and effectively.

Cementing the industry

This foundation also has the potential to cement its participants as the premier AI companies. Although it's not just the big tech firms that have joined this foundation, and it is being organized by a long-standing organization with a strong reputation for keeping software development as its main focus, the potential is there for exploitation and, arguably, stagnation.

The largest companies are likely to have the largest input on the direction of these open standards, which could allow them to shape the future of Agentic AI in a way that benefits them. With enormous investment capabilities, larger companies are capable of pivoting toward new efforts at the drop of a hat. If any breakthroughs are made in Agentic AI that require heavy investment or access to hardware and software to take advantage, those larger companies will be in a prime position to reap the rewards.

Indeed, the very financing of the entire AI industry has been rife with major companies pumping each other's stock prices with promises of future revenue and long-tail investment pledges that won't be realized for years. The major companies all collaborating to advance the industry have some strengths, but it could also be taken as a further example of the major firms propping each other up for the foreseeable future.

Unproven, unrealized, unprofitable

At their core, the major AI tech firms have the same problem: They aren't making any money from any of this, yet. AI costs far more to run than it generates for the companies developing it, and there's no sign of that stopping any time soon.

This foundation could be a way for them to collectively try to solve this issue. Someone needs to make a killer AI app or a way for agentic AI to fix real problems, or rapidly enhance productivity, so that these massive companies can make good on their equally large investments.

Shareholders and early investors are going to come calling for the promised profits over the coming years. Accelerating the development of their tools and standards through this foundation could be one way for these major firms to also accelerate their path toward profitability.

OpenAI and Anthropic have released a pair of new reports on how the use of their AI products helps to grow enterprise productivity. The reports serve as the AI industry's latest response to a wave of recent academic studies amid a sea of public discontent pushing back on the AI data center boom, as the big AI firms seek to stow doubts in the value of enterprise AI spending.

OpenAI's report released today, "The State of Enterprise AI", hinges on two major points: companies are using AI more, and workers are saving time as a result. OpenAI claims that in a survey of 9,000 workers across 100 companies, workers reported having saved 40 to 60 minutes of work per day on professional tasks with the use of ChatGPT. Of these 9,000 workers, 75% of respondents reported that AI has improved either the speed or quality of their work.

Because OpenAI's report appears to be more focused on marketing to enterprise than performing scientific resaerch, there is no way of knowing beyond the most favorable published numbers how this 75% metric breaks down. Much of the data isn't very specific.

The OpenAI report also makes a case that companies are using AI more, stating that "frontier firms" and "leaders" are sending 6x more prompts to ChatGPT than "laggards", or the median AI-using firms. However, all this "6x" number proves is that some companies use ChatGPT more than others, saying nothing about the quality of the work done or how this usage affects business numbers.

OpenAI may be looking to contradict studies from educational institutions published earlier this year. An August study from MIT showed that 95% of organizations that invested in AI business products "found zero return" despite corporate investments of $30-40 billion. The study shows that the "vast majority" of AI pilot programs stall, delivering little to no measurable impact on profit. Shortly after, a research initiative from Harvard Business Review found that most professional AI use constituted little more than "workslop", or work content that "masquerades as good work, but lacks the substance to meaningfully advance a given task.”

In late November, Anthropic published its own research to respond to these allegations. The internal survey, submitted without peer review, found that using Claude, Anthropic's AI assistant, cuts down the time it takes people to complete work tasks by 80%, from an average of 90 minutes down to 18 minutes, based on a look at 100,000 private Claude conversations. But as the company admits, buried deep in the website copy, these numbers have no promise of actually reflecting real-world efficiency. "This doesn’t account for the time that humans might spend on these tasks beyond their conversation on Claude.ai, however, so we think these estimates might overstate current productivity effects to at least some degree," says Anthropic's own study.

Regardless of these self-admissions of weak methodology and the cherry-picking of numbers, the AI industry is still publicly bullish on its own claims of enterprise profit increase.

In a statement to Bloomberg, OpenAI COO Brad Lightcap directly addressed the MIT and Harvard Business studies. "There’s a lot of studies flying around saying this, that and the other thing. They never quite line up with what we see in practice."

The AI industry doesn't just have academia to reckon with in the new year, however. The physical realities of the needs of the AI industry to keep up with data center expansion are catching up, with a copper shortage expected to hit data center buildouts in the next decade, matching the current RAM shortage crisis currently caused by AI data centers. Add this to rising public fear and outrage over data center expansion's health risks and rising electricity prices, and the AI juggernaut will have more pictures to paint than one of productivity.

OpenAI chief Sam Altman said in an internal memo that the company is in a “Code Red” status, meaning all other projects will take the backseat in favor of ChatGPT. According to The Wall Street Journal, Altman said in the memo that the company needed to improve its flagship AI LLM’s personalization, speed, and reliability, as well as allowing it to cover a wider range of topics.

OpenAI popularized the LLM with the release of ChatGPT on December 1, 2022, almost exactly three years ago. However, its competitors have since caught up with their own releases. Its current biggest threat is Google, which just released Gemini 3 in early November, and is baked into most, if not all, of Google’s products. Aside from this, Anthropic has also released Claude Opus 4.5 late last month, which is gaining market share among enterprise and business users. There are also several other challengers, like Meta’s open-source LLaMA and China’s DeepSeek.

This tightening competition has made it more crucial for OpenAI to stay ahead of the curve, especially as it continues to burn through capital. Microsoft, which has invested over $13 billion in the startup and holds around 27% of its for-profit OpenAI PBC, reportedly lost $3.1 billion on the AI company in its fiscal first-quarter earnings released in late October. Redmond previously hid its losses on OpenAI as part of its $4.7 billion “other” expenses.

Despite not turning a profit since its founding more than 10 years ago, OpenAI has been continuously spending and investing billions of dollars on massive data centers, with its president envisioning a 10-billion-GPU future. It even signed a deal with AMD in early October to secure 6 gigawatts in chips in exchange for 10% of AMD shares. All this talk of billions of dollars has led some experts to ask how it will make good on these investments.

Even though OpenAI’s rivals have taken strides in advancing their own LLMs, it still leads in several aspects. In fact, the company is expected to drop a new model next week that will challenge Gemini’s latest release. However, the WSJ reports that the response to its last major release in August 2025, GPT-5, was less than stellar, with users complaining that it felt clinical and was less capable in math and geography versus previous versions. OpenAI updated the model around three months later to fix these issues.

OpenAI CEO Sam Altman released an internal memo last September 2025, stating that he plans to build up to 250 gigawatts of compute capacity by 2033. According to Truthdig, this is equivalent to the electricity required to power the entire nation of India and its 1.5 billion citizens. It would also emit twice the carbon dioxide that ExxonMobil produces, which the report says is the current “largest non-state carbon emitter” in the world.

Aside from carbon emissions from its data centers, the writer, Alistair Alexander, estimates that the 250 GW capacity is enough to support 60 million Nvidia GB300 GPUs. This means that OpenAI would have to order 30 million GPUs annually to ensure continuous operation. While Alexander claims this is because the cards are run 24/7, 365 days a year, and therefore have a short lifespan, the two-year life cycle of a GPU actually refers to its economic value, which drops with the release of newer generations of products. Either way, the number of GPUs required to power OpenAI's ambition will be staggering.

This estimate is for just one company: other tech giants are also planning their own massive AI data centers — for example, Elon Musk’s xAI wants to have 50 million H100-equivalent AI GPUs by 2030, which will require around 5GW of power. This massive build-out, which is happening all over the globe, is squeezing power supply, causing a spike in electricity prices, and reducing power quality for households. There’s also the question of its impact on the water supply, especially as the massive compute capacity requires astronomical amounts of cooling.

Alexander also looked upstream and investigated the impact of chip manufacturing on the environment. The massive demand for AI processors and the billions of dollars companies are willing to spend on them has led to an explosion of new fabs. In the past two years, construction on 97 new fabs has started all over the globe, with Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung among them. Much like the data centers, these chip factories also require metric tons of water and massive amounts of electricity to operate. More than that, the processes for making the most advanced chips also often require toxic chemicals.

For example, the report says that TSMC’s Fab 25 would require at least 1GW of power — enough power for 750,000 households in Taiwan. Environmental Rights Foundation deputy CEO Po-Jen Hsu also said that Fab 25 would also use 100,000 metric tons of water daily, or about the same amount that 196,000 Taichung residents would use in a day. SHARPS, the semiconductor labor organization in South Korea, has also said that some workers in Samsung factories were suffering from various forms of cancer, all of which were linked with the chemicals used in their workplaces.

The race for AI supremacy is putting a lot of demand on the limited resources we have on Earth. The issues aren’t just limited to data centers, too. Instead, it goes all the way up the supply chain — from manufacturing the most advanced chips to the mining operations needed to extract the rare earths and other materials needed to make these GPUs. Tech companies are pouring billions of dollars into hardware and infrastructure to extend their capabilities, while nations are engaging in a trade war to limit the advance of their rivals.

The report concludes: “As Silicon Valley CEOs anxiously figure how much computing it will take to propel artificial intelligence forward, the real question we should be asking is how much more artificial intelligence the planet can take.”