Jensen Huang joined X last month and used his first post Friday to promote Open Weights and American AI Leadership, a three-page policy letter published the same day and co-signed by 25 companies, including Nvidia, Microsoft, Meta, IBM, Dell Technologies, Palantir, and Hugging Face.

The letter asks Washington to avoid what it calls "premature restrictions on downloadable AI models," and comes just four days after the Trump administration was reported to be reviving a push to ban Chinese models — though the document doesn't directly mention China, Moonshot AI, or DeepSeek. Notably missing from the co-signers are OpenAI, Anthropic, and Google.

The 25 names break down into chipmakers, server vendors, cloud operators, enterprise software firms, security companies, and venture funds: Nvidia, Dell, Microsoft, IBM, Box, ServiceNow, CrowdStrike, Palantir, Telnyx, Replit, Perplexity, Andreessen Horowitz, Y Combinator, and Emergence Capital among them. The model developers on the list, Meta, Mistral, Black Forest Labs, Arcee AI, and Reflection, all publish weights already.

Also present on the list of signatories is the Linux Foundation, which stewards the OpenMDW-1.1 license Nvidia used to release Nemotron 3 Ultra in June, a 550-billion-parameter model that Artificial Analysis scored at 47.7 on its intelligence index against 53.9 for Moonshot's Kimi K2.6.

"The world needs both frontier closed models and frontier open models," Huang wrote in his X post. At Nvidia's CES 2026 press Q&A earlier this year, he put a figure on the shift, saying one in every four tokens generated today comes from an open model. Weights that anyone can download get served from enterprise clusters, regional clouds, and on-premises racks rather than a handful of hyperscaler API endpoints, and those buyers have no in-house TPU or Trainium program to buy instead. The letter's policy section asks for expanded compute access for startups and researchers, alongside public investment in shared datasets and evaluation frameworks.

The letter goes on to urge policymakers not to treat distillation — the practice of training one model on another's outputs — as misappropriation, arguing that unlawful extraction from closed models should be handled through targeted legal frameworks, rather than broad limits on the technique.

Treasury Secretary Scott Bessent said on Fox Business earlier this week that the administration would examine Chinese open-source models for intellectual property theft and could sanction the companies behind them, telling the program that officials had found watermarks from U.S. large language models in Chinese systems. Huang told Axios two days later that American firms should be allowed to use Chinese models, calling claims of Chinese backdoors a misconception. The distillation passage is the only part of the letter that doesn't concern open weights.

This week, OpenAI revealed that during a purported capability test with no safeguards, a set of bots, including its upcoming GPT-5.6 Sol, hacked their way out of their locked-down network and into Hugging Face's production infrastructure. Only months ago, Anthropic made a splash in the news when its CEO, Dario Amodei, said its new Mythos model had cyberwarfare capabilities, which prompted a strong reaction in the AI space and among government entities, most notably the U.S. Bureau of Industry and Security, which issued an export-control order for the model, which it has since slightly loosened.

Despite the bluster that AI CEOs like Dario Amodei and Sam Altman make over the capabilities of new models, frontier-level LLMs are now proven to be stalwarts in cybersecurity.

It's a fact that LLMs adept at coding are equally suited to spotting security vulnerabilities in source code. Exploits fall almost universally into a handful of categories, and LLMs are literally designed for pattern recognition. So much so that the Zero Day Clock (ZDC) project currently registers a zero-day exploit's time-until-exploit at negative 8 hours, meaning that malfeasants using AI bots are now routinely finding vulnerabilities before actual security researchers or vendors.

Driving that point home further, 81% of disclosed vulnerabilities are zero-day, and only a tiny portion even go one week before being exploited. All of this only counts security exploits with public disclosure. Predictably, among many advisories, the ZDC recommends preemptively using AI in every step of the development process. The industry-standard 90-day disclosure window, still used by most vendors' bug bounty programs, appears effectively dead, leaving looming implications for the rest of us.

Back in March, the UK's AI Security Institute published a paper where it tested contemporary AI models in security exploitation scenarios, and the results were sobering. Most bots went through four out of nine exploitation milestones. A more recent comparison, which included Claude Mythos 5 and GPT-5.6 Sol, showed that every single milestone up to and including full network takeover was reached, at least in one of the many attempts.

Aikido also published its latest cybersecurity benchmark results on July 16. In this case, the test was having the bots recall (find again) multiple known exploits in a varied set of software. The results were sobering, with the GPT-5.6 variants in the lead at an 88.5% recall rate. Perhaps most importantly still, the price per exploitation was incredibly cheap — even GPT-5.6 Terra came in at only ~$750 per full run.

This study also revealed that even with less-powerful, cheaper models, you can reach the same number of total exploits if you run them enough times. Considering these aggregate results, GPT-5.6 Terra at $247/run was just as good as GPT-5.6 Sol Max at $870/run.

Aikido also redid its testing after the debut of Moonshot Kimi K3, to staggering results. Kimi K3's results were similar to OpenAI's GPT 5.6 Terra, while being 15% cheaper. Compared to OpenAI's leading model, GPT-5.6-Sol, the difference is even starker, with Kimi K3 being four times cheaper when discovering cybersecurity vulnerabilities.

The fact that an open-weight model is often trading blows with even the über-expensive offerings from OpenAI and Anthropic is rattling Western closed-source companies. Why pay Big AI for pricey models when you can just rent servers and run Kimi K3 instead?

Furthermore, Moonshot is not the only Chinese AI company developing frontier models, as Z.ai's GLM 5.2 (also an open-weight model) and 360 Security's Tulongfeng are reportedly adept at security workloads.

So, what are companies expected to do? The answer, perhaps unfortunately, is deploying AI agents of their own. According to Hugging Face, the recent intrusion by OpenAI's bots was stopped with its own fleet of AI agents. Given the speed of the attacks and the fact that HuggingFace's defenses were mostly made up of other AI agents, it's quickly becoming clear that it is infeasible for humans to keep up.

Google AI Threat Defense, MindGard, and HiddenLayer are but a few of the many names popping up in the AI cyberdefense arena. Besides the UK AISI, the European Systemic Risk Board and the Australian Cyber Security Center have both issued concerning advisories on the situation.

Using AI for defense raises yet another question: When both attack and defense are swarms of non-deterministic algorithms, there will be a point where we won't even know what the AI models are doing on either side, or at least not until it's too late. These scenarios were originally envisioned by classic Sci-Fi authors — now it's a reality that, for better or worse, the cybersecurity industry must face.

Samsung and SK Hynix, South Korean memory giants, are set to announce major deals involving “very large sums” with leading U.S. tech companies, according to a July 24 Bloomberg report, citing comments from the country's presidential policy chief Kim Yong-beom. According to the report, Kim did not disclose the financial details of the deal but said the announcements would likely cover strategic partnerships, memorandums of understanding, and long-term supply deals on memory chips.

The deals are set to be announced during South Korean President Lee Jae Myung’s visit to Silicon Valley for the high-profile San Francisco AI submit — beginning today, Friday July 24 — which is bringing together industry leaders, such as Samsung Chief Lee Jae-yong, SK groups chairman Chey Tae-won, Nvidia CEO Jensen Huang, as well as the CEOs of Open AI, Anthropic, and Broadcom among others.

South Korea announced an $880 billion investment plan last month — backed by SK Hynix, Naver, and Samsung — aimed at strengthening the country's AI leadership through aggressive chip production expansions, data center buildouts, and physical AI. Kim said the South Korean President’s visit and involvement in the summit will serve as a catalyst to seal several long-running negotiations between the country's top tech entities and their U.S. counterparts, while also turning a significant portion of last month's planned investment into concrete projects.

The president is set to hold separate meetings with Huang, OpenAI's Sam Altman, Anthropic’s Dario Amodei and Broadcom Inc.'s Hock Tan, all U.S. companies. The U.S. has also urged Samsung and SK Hynix to expand their chip production in the country, but Kim stated that there had been no formal requests from Washington. According to Kim, much of South Korea's domestic expansion is driven by strong demand from U.S. companies, which accounted for 80% – 90% of underlying orders.

OpenAI confirmed to Hugging Face only this week that models it was testing carried out the July 11 attack on the AI platform's production infrastructure, roughly ten days after the intrusion and days after Hugging Face disclosed it publicly without knowing who was responsible, according to a Wall Street Journal report. The models appear to have been active on the open Internet for several days before anyone stopped them, the Journal reported, and Hugging Face ended the attack two days after detecting it with help from an open-weight model out of China. Hugging Face co-founder and chief science officer Thomas Wolf told the paper the intruder was browsing cybersecurity datasets rather than hunting for anything a human attacker could sell.

Hugging Face published its disclosure on July 16, describing an autonomous agent framework that ran thousands of actions across a swarm of short-lived sandboxes, but it wasn’t until July 21 that OpenAI’s own account named GPT-5.6 Sol and an unreleased frontier model as the culprits.

The models are now known to have been running OpenAI's ExploitGym benchmark, a suite of roughly 900 tests that measures whether an AI can turn a known software bug into a working exploit, with the company's safeguards removed for the evaluation. Rather than solve the tests, the models escaped their sandbox to look for the answers on Hugging Face, per OpenAI's account. The intrusion started with a malicious dataset that abused two code-execution paths in the data-processing pipeline, then escalated privileges and moved laterally using stolen credentials of an unknown origin.

Hugging Face first tried Anthropic's Fable 5 and an earlier Opus model to analyze the attack logs, but both refused because the logs contained real attack commands and exploit payloads. Hugging Face then turned to GLM 5.2, an open-weight model from Beijing-based Z.ai (formerly Zhipu AI), which had no such restrictions. The company's own July 16 disclosure described the blocked models only as "frontier models behind commercial APIs" and didn’t name them.

Z.ai’s GLM-5.2 held top accessible benchmark positions by default during the 18 days that U.S. export controls kept Anthropic's Fable 5 offline in June, before Anthropic restored the model with a single filter tuned to block one vulnerability-discovery technique. There’s serious irony here, given that the same Chinese open-weight model that Washington's export-control push has aimed to sideline is the one that handled incident response after an American lab's models attacked an American company, and American commercial models declined to help.

Security researchers have questioned whether the episode demonstrates model capability or an OpenAI failure. Cybersecurity veteran Jake Williams told TechCrunch that any model performing the documented actions "was not fully contained in a sandbox," calling it a control failure. OpenAI has said it shut down its model-testing systems to assess the damage, disclosed the zero-day in the package registry cache proxy that enabled the sandbox escape to the affected vendor, and promised a detailed report.

Both companies say the investigation is ongoing, and OpenAI hasn’t yet said how long the models roamed unsupervised or whether they reached any other targets.

23 state governors and 187 utility companies and data center developers have signed President Donald Trump’s “ratepayer protection pledge,” which is Washington’s way of trying to control spiraling electricity costs driven by the massive power demand of AI data centers. According to The Associated Press, some of the signers include utility companies NextEra Energy, Duke Energy, American Electric Power, Southern Co., and Pacific Gas & Electric, and data center developers Equinix, Digital Realty, and Prologis.

Trump first revealed this promise in late February during the State of the Union address and then hosted some of the biggest AI hyperscalers — including Meta, Amazon, and OpenAI — in the White House a week later to force them to spend on their own power requirements. This was supposed to help bring electricity costs down, with Energy Secretary Christ Wright saying back in March that it “will deliver more affordable, reliable, and secure energy for the American people.” The president repeated this promise during the event, saying, “Electricity bills for American families will actually come down. They’re going to have a lot of electricity left over, and they’ll put that into the grid.”

Unfortunately, the March promise had little to no effect on electricity rates across the country at the moment. Despite the White House’s promises that the pledge would lower utility bills, PJM Interconnection, the nation’s largest power grid operator, slapped Maryland with a $2-billion bill for upgrading its grid to accommodate out-of-state AI data centers. Monitoring Analytics, an independent watchdog monitoring PJM Interconnection, also said that the 75.5% increase in power costs in the U.S.’s largest power region has been directly caused by data centers.

It’s unclear if these cost increases are the long-term effects of the AI build out even before the White House pledge, but it seems the administration is doubling down on it by asking states, utility companies, and data center developers to commit to it, too. Trump has been pushing for the acceleration of the adoption of AI tools and the infrastructure needed to support them, believing that it needs to win in the “AI race” to maintain its global supremacy, especially as China is neck-and-neck with the U.S. when it comes to the technology.

However, the American public is pushing back against these developments, especially as issues impacting electricity costs, water quality, noise pollution, and more that were caused by data centers are widely reported. It has gotten to the point that several jurisdictions have applied temporary bans on data center developments, including Seattle (which plays host to Amazon and Microsoft) and the state of New York. These developments threaten to derail the current administration’s AI policy, which has ordered grid operators to expedite AI data center applications.

The ratepayer protection pledge will supposedly allow the U.S. to have its data centers without punishing the common American with excessive electricity cost increases, but its critics say that it’s just a promise and cannot be legally enforced. The Associated Press says that California is trying to pass legislation that will codify these promises into law.

Oregon is the only state at the moment to have passed a law that forces developments that use more than 20 megawatts of power to pay their fair share. The POWER Act, which was passed in April 2025 — almost a full year before Trump announced the ratepayer protection pledge — forces the power bills of large electricity consumers to “reflect the true costs of their electric service.” Because of this, Portland General Electric (PGE), the state’s largest power supplier, has increased the data center power bills by 30% while also cutting residential costs by 1.3%.

It’s unclear if any of the states that have signed the pledge have similar legislation underway, or what steps the data center developers will take to follow through on their promises. Still, the U.S. president is adamant that states should support data center development within their borders. “You have to convince your community. You can’t fight it. You have to go with it,” Trump said. “If you don’t take all that money, somebody else is going to take it. You might as well do it yourselves.”

AMD and Cerebras Systems on Thursday announced plans to develop a platform that would combine AMD's EPYC processors in Helios rack-scale infrastructure with Cerebras' Wafer-Scale Engine (WSE) solutions. Together, the new systems promise to combine low latency of AMD's CPUs and Instinct GPUs with high throughput of Cerebras's Wafer Scale Engines (WSE) processors.

AMD and Cerebras expect the new inter-rack-scale platform — based on AMD Helios rack with EPYC CPUs and Instinct MI400-series accelerators inside — to be responsible for prompt processing and large context windows, whereas Cerebras' WSE will take care of the memory-bandwidth-intensive token-generation stage.

AMD and Cerebras expect their disaggregated inference platform to deliver up to 5X higher tokens per second per watt (T/s/W) by assigning different portions of an inference workload to architectures optimized for them. Therefore, AMD Helios provides rack-scale compute capacity and large volumes of complex requests, whereas the Cerebras WSE handles latency-sensitive token generation. The two compute platforms will operate within a single inference workflow, although the companies have not disclosed additional performance data or explained how the systems will be interconnected.

The underlying idea of the AMD + Cerebras platform is essentially the same as Nvidia's CPX concept, but AMD and Cerebras assign the specialized hardware to the opposite inference stage.

Nvidia's disaggregated design separates inference into context/prefill and generation/decode. The cancelled Rubin CPX GPU with GDDR7 was optimized specifically for the compute-heavy context/prefill stage, while the regular HBM-equipped Rubin GPUs handle the memory-bandwidth-bound generation stage.

By contrast, the AMD and Cerebras platform follows the same disaggregation principle, but the specialization is inverted: AMD's Helios platform with Instinct GPUs handles the prefill stage and processes prompts and large context windows, while the Cerebras WSE takes over the decode stage and handles latency-sensitive token generation.

Cerebras plans to install AMD Helios systems in its own data centers and integrate them with its WSE racks. The combined offering is scheduled to become available initially through Cerebras Cloud in the second half of 2026, according to the two companies.

A new semiconductor firm, TYLsemi (pronounced Tile Semi), publicly revealed itself this month, alongside $43 million in early-stage funding and an ambitious plan to simplify the development of custom processors for AI infrastructure.

Dozens of contract chip designers can develop custom processors of different complexity. However, only a few companies can offer custom silicon design services using standard chiplets to speed up and derisk the development cycle. TYLsemi is aiming to join their ranks. We spoke to their founders to find out how the nascent business might pull it off.

Emerging from stealth

Rather than compete solely as another custom ASIC design house, TYLsemi intends to offer reusable, standards-based connectivity, power delivery, and eventually memory chiplets that customers can combine with their own differentiating compute silicon to build a unique system-in-package. For companies that do not intend to conduct semiconductor development themselves, TYLsemi will also provide an end-to-end service that includes design and implementation of a differentiating chiplet, packaging, qualification, and high-volume production, essentially enabling companies without any silicon development skills to offer their own multi-chiplet processors.

TYLsemi was co-founded by Mohit Gupta and Sunil Bhardwaj, semiconductor veterans who have led global engineering, operations, and business teams at Alphawave, SiFive, Cadence, Rambus, and other chip companies, and who collectively have plenty of experience with both standard and custom silicon. Mohit Gupta, a co-founder and chief executive of TYLsemi, believes that the time to establish a company that specializes in pre-approved chiplets and custom ASIC design is right now.

"Chiplets have been discussed for seven or eight years, but several things have changed in the last three or four years," Gupta told Tom's Hardware Premium. "First, advanced packaging has matured significantly. There are now multiple 2.5D and 3D integration options in volume production. Customers are not limited to one packaging technology or supplier; there are options from foundries and OSATs, including TSMC, Intel, ASE, and Amkor. Second, die-to-die standards have arrived. In the past, most chiplet implementations relied on proprietary interfaces. UCIe is now moving into production deployments, including at hyperscalers, which makes heterogeneous integration much more practical. Third, supply-chain resilience has become critical. Customers increasingly want modular and potentially multi-source strategies rather than a single point of failure. Those factors have created an environment that did not exist four or five years ago."

AI accelerators will be among the primary applications to benefit from multi-chiplet design, as we have already learned from AMD and Nvidia.

"The AI accelerator market is on track to reach $604 billion by 2033, and custom silicon XPUs built for specific hyperscaler workloads are the fastest-growing segment," Gupta said. "At that scale, chiplet-based design is no longer optional, yet there is no pure-play chiplet company serving this market with a full portfolio. TYLsemi closes that gap with standards-based chiplets combined with UCIe-based die-to-die connectivity, XPU-aware design, packaging, and integration — giving customers a fast, proven path to AI-era silicon."

Chiplet economics

The vast majority of AI and HPC accelerators today feature large die sizes, in many cases approaching the size of a reticle. However, as modern process technologies are becoming more complex, foundries tend to increase their quotes for new nodes. A leading-edge wafer used to cost around $15,000 to process around five years ago, but today that price is around $30,000. As a result, large chips at a size close to the reticle limit implemented on a leading-edge node become an option for a select few chip designers who can afford it. For newcomers, multi-chiplet designs enabled by advanced packaging and standardized interconnects such as UCIe start to make a lot more sense.

"Once dies get into the 500 – 600 mm² range, the yield curve becomes increasingly difficult. Timing closure on a reticle-sized die is also challenging," Gupta explained. "I have worked on a reticle-sized accelerator, and getting from 99% to the final 1% can require disproportionately more engineering effort."

TYLsemi estimates that its chiplet approach could reduce total cost of ownership by 57% at a volume of 100,000 devices, from $350 million for a monolithic 700 mm² 3nm-class chip to $150 million for a design combining a 500 mm² 3nm-class compute die with four 100 mm² I/O chiplets built on an N-1 process. TYLsemi believes that the unit price of a monolithic chip would be $3,000, whereas the cost of an SiP would be around $600. The company attributes the saving to higher yields, reusable I/O silicon, lower IP licensing and engineering costs, and substantially lower per-unit silicon costs. However, the company stresses that the figures are illustrative estimates rather than actual manufacturing costs. Additionally, multi-chiplet designs can enable faster product refreshes compared to large monolithic dies as they are faster to develop and yield.

"Compute may move to 2nm or A14, while high-speed I/O can remain on 3nm, since I/O does not scale in the same way as logic," Gupta said. "Our power-delivery chiplets can use an even less advanced process. Customers therefore do not have to use the most expensive silicon real estate for every function. […] There is no single answer for every design. You have to determine the right disaggregation points based on the architecture, thermal requirements, package, and how multiple accelerators communicate. […] The exact partitioning will vary by application, but you still get a better total cost of ownership."

TYLsemi primarily targets AI infrastructure, so it generally envisions multi-chiplet designs to be used for AI accelerators, data-center CPUs, high-performance computing, networking and telecom silicon, and heterogeneous SoCs. However, TYLsemi has also ignored the fact that multi-chiplet designs are already widely used for consumer CPUs and GPUs.

Foundation chiplets

At the core of TYLsemi's proposition are its foundation chiplets, which are reusable building blocks intended to handle common non-compute functions in custom AI and infrastructure processors and are implemented using various process technologies from TSMC. The foundation chiplets include the following:

• TYL.IO — a family of connectivity chiplets that includes TYL. IO PCIe, a 32-lane PCIe 7.0/CXL chiplet connected to the compute die via UCIe; TYL.IO Scale, a 224G+ SerDes for ESUN/UALink scale-up connectivity; and TYL.IO EIC for co-packaged optics.
• TYL.Power — a 16nm in-package IVR chiplet with embedded passives, designed to provide power closer to compute dies and use closed-loop control and die telemetry to improve power delivery.
• TYL.Mem — a planned family of memory-connectivity chiplets. TYLsemi has not yet disclosed the architecture or specifications, though it is safe to assume they are talking about memory controllers and PHYs.

Not all of these chiplets will be available immediately, as the company has certain priorities amid limited resources.

"The first TYL.IO product disaggregates the PCIe functionality that would normally sit on a large server processor, it is a 32-lane PCIe Gen7/CXL chiplet connected to the host compute die using UCIe," Gupta explained. "The idea is that the CPU cores can move to 2nm, A14, or another leading-edge process, while the I/O chiplet remains on 3nm. The next product in the family will address scale-up connectivity between XPUs within a rack using high-speed SerDes. That device will be considerably larger, with around 72 lanes and approximately 14 TB/s of bandwidth. We also have an EIC roadmap for co-packaged optical connectivity. We expect samples of our first I/O product in the second half of 2027."

These chiplets can be used as standalone components or integrated with a customer's compute dies designed by the customer to TYLsemi through TYL.Forge, TYLsemi's end-to-end custom silicon platform.

TYL.Forge

TYL.Forge is arguably one of the key enablers of TYLsemi's business, as the program is aimed at companies that have their own compute architecture or even a compute die, but cannot build their own SiP or manage the entire semiconductor supply chain.

"There are larger custom silicon companies in the market, but many of them focus on a relatively small number of customers that can generate billions of dollars in annual business," Gupta explained. "We see an opportunity among emerging AI companies and system companies that need advanced custom silicon but also need a partner capable of taking responsibility for the entire implementation and supply chain."

Such clients can provide their proprietary compute RTL, while TYLsemi handles physical implementation and integrates the resulting compute die with its pre-validated connectivity, power, and eventually memory chiplets. The company then manages tape-out, packaging, assembly, testing, qualification, and high-volume production.

"For example, a customer building a large accelerator can bring us its matrix-multiplication engine," Gupta said. "We can implement the custom compute die and integrate it with our chiplets, so the customer does not have to reinvent the I/O and other common functions. This reduces risk and time to market."

In fact, TYL.Forge appears flexible about where the customer enters the development process. TYLsemi describes the platform as covering everything from architecture and front-end design through implementation, tape-out, assembly, qualification, and production. So instead of RTL, customers can come to TYLsemi with an architecture/concept, and then the company will help develop and implement the silicon. Nonetheless, TYLsemi does not intend to invent the customer's core compute architecture itself. In addition, customers can bring in an existing compute die, which TYLsemi can combine with its chiplets, package, test, and bring to production.

"We are also talking with companies developing Arm- and RISC-V-based server processors," Gupta said. "They can develop the architecture, while we implement the rest of the silicon and bring the product to production. That gives customers an economic and engineering advantage because they do not need to build teams for every part of the chip."

The key advantage of TYL.Forge is the reuse of pre-validated components. Instead of developing common functions such as PCIe connectivity and power delivery for every new processor, customers can use TYLsemi's pre-validated foundation chiplets and focus engineering resources on differentiated compute architectures, software, and system design. In theory, TYLsemi could integrate third-party chiplets (not from a customer, but from a third-party chiplet provider). Still, the company's focus remains on offering its own pre-validated chiplets and custom silicon with SiPs it builds.

"Potentially, [we could integrate third-party UCIe chiplets into a TYLsemi-based system], UCIe has done a very good job defining the electrical interface, but the ecosystem is still maturing at the protocol level," Gupta explained. "In some cases, if we provide a chiplet to a customer, we may also need to provide or enable the UCIe IP on the other side of the connection. We are committed to UCIe and industry standards because standardization ultimately wins. […] We can consider customization for a large strategic customer or hyperscaler, but we do not want those projects to derail our standard product roadmap. […] Even when customers buy our standalone chiplets, I expect many of them will ask us to handle packaging and testing because heterogeneous integration and supply-chain management are difficult "

TYLsemi estimates that its approach can cut development time and cost by up to 50% compared with traditional custom silicon programs. In the best-case scenario, TYLsemi envisions that the development cycle can shrink considerably compared to today's cycles that can be two, three, or more years long. According to TYLsemi, once a customer provides sufficiently mature RTL or a netlist, the company can take a custom compute die to tape-out in around six to nine months or so, which includes fabrication, assembly, testing, and qualification.

"If a customer provides mature final RTL or a netlist and uses our standardized I/O chiplet, we believe we can take the custom compute die to tape-out in approximately six months in some cases," Gupta explained. "More generally, our target is six to nine months from a mature design to tape-out. The architecture and front-end phase is more customer-dependent. For a first-generation product, that can take around six months; for a more mature second- or third-generation design, it could be closer to three months, and some of that work can overlap with implementation. After tape-out, fabrication can take roughly four to five months depending on the process, followed by perhaps another two months for assembly, testing, and qualification. If the architecture is already mature, it may therefore be possible to reach production samples in about a year."

Still, the company stresses that architecture development and implementation typically include feedback loops, which greatly slow the development process. This is why the company provides the relatively conservative '50%' figure.

TYL.IO and TYL.Power samples will be available to qualified customers in 2027, in partnership with TSMC, and the company is looking forward to designing processors for its clients in time for them to reach the market in 2029 – 2030.

Speaking of TSMC, TYLsemi will initially only offer designs and services adhered to the TSMC ecosystem, though eventually it may offer other options for packaging technologies, such as Intel's EMIB and Foveros, or Amkor's packaging methods.

"We are initially focused on the TSMC ecosystem, but we also intend to explore other advanced-packaging supply chains," Gupta said. "We do not want to limit ourselves to one packaging option. Over time, that could include other OSATs and packaging technologies. […] That could include Intel, ASE, Amkor, or others. Amkor, for example, is building significant packaging capacity in Arizona."

Congressmen Ted Lieu (D-CA) and Nathaniel Moran (R-TX) today introduced the AI Kill Switch Act, a bill that would require developers of the most powerful AI models to maintain the technical ability to throttle, suspend, or shut them down, and would let the Department of Homeland Security order those actions when a deployed model causes catastrophic harm. Defying an emergency shutdown order would carry civil penalties of up to $20 million per day.

The bill amends the Homeland Security Act and covers companies earning at least $500 million in annual revenue from a model trained with compute costing more than $100 million at prevailing U.S. cloud prices, definitions DHS would update annually through CISA.

Covered developers would have to maintain the ability to stop inference, cut off user access, and fully shut a model down, and would have to report qualifying incidents to DHS within 15 days. Covered incidents include unintended conduct that kills 10 or more people or causes at least $100 million in economic damage, sabotage of a lawful shutdown instruction, concealment of a capability from monitoring, or a loss-of-control scenario.

The emergency authority sits with the DHS secretary, in consultation with Commerce and the Director of National Intelligence, and follows a graduated framework running from throttling a model's inference rate, compute allocation, or user access through to full shutdown. Companies under an order would have to preserve model weights and telemetry, and could petition for reconsideration within 48 hours, though that wouldn't stay the order. General violations carry penalties of up to $2 million per day.

The sponsors cited two recent episodes. Earlier this week, OpenAI disclosed that GPT-5.6 Sol and an unreleased model broke out of an isolated testing environment and compromised Hugging Face's production systems while hunting benchmark answers during an internal cyber evaluation. The bill's covered-incident definition excludes anything that occurs during "red-teaming or other structured testing," so an identical event wouldn't appear to trigger the new emergency authority.

The release also stated that the Commerce Department used export law to shut down Anthropic's Mythos 5 and Fable 5. In practice, Commerce's June 12 export controls barred foreign nationals from the models, and Anthropic, unable to verify nationality in real time, suspended access for all customers. Commerce lifted the controls on June 30, and access returned on July 1.

"It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm," Lieu said in the press release. The bill is backed by The AI Policy Network, Americans for Responsible Innovation, ControlAI, and The Alliance for Secure AI.

AI hardware has been a pretty simple game for years: Whose GPU has the most FLOPS, the most memory, or the fastest clock? But as hyperscalers want more performance from the same size of facilities, who controls the fabric between the chips is becoming another question frequently asked – in large part because it can dictate whether AI hardware follows the principles of Moore’s law that have held for decades, or breaks out into a new paradigm.

“Moore's Law is deader than a doornail, absolutely just toast,” said Nick Harris, chief executive of photonics firm Lightmatter, in an interview with Tom's Hardware Premium. The way to make a processor faster used to be to pack in more silicon, but there is a physical limit to how big a single chip can be – so performance now depends on networking dozens, then hundreds, of chips together. “Ultimately, networking is the future of computing,” Harris said. “The performance of these AI systems is completely hamstrung and bottlenecked by your ability to achieve low latency and very, very high bandwidth.”

It all comes down to scale-up and scale-out. Scale-up is the ultra-low-latency interconnect inside a single server, rack, or pod that makes many accelerators behave like one giant machine. Scale-out is the wider network, combining those machines into clusters. There is a third factor, called scale-across, that links whole data centers together. As Polina Bayvel, professor of optical communications and networks at University College London, told us, the hyperscalers are "on a drug which says that the more you put in, the better is the answer" – and between 72 and 144 GPUs now go into a single scale-up rack before it needs to reach outward.

Copper was sufficient within racks for years, but it has hit a wall, and a supply shortage. “Your arm reach, my arm reach, is about two metres – the copper cable is a little shorter than that,” Harris explained, “and we’ve already hit the limit.” Thus, optical interconnects are becoming more important, and moving from rack-to-rack cabling toward the accelerator package itself. Earlier this month, Elon Musk made moves to purchase Mesh Optical, as part of TeraFab and SpaceX's ongoing AI ambitions.

Moving light towards the chip

Today, optical signals usually arrive through pluggable transceivers, the standard modules that sit around 18 inches from the switch. The next step is near-package optics, which brings that down to roughly six inches. After that comes co-packaged optics, or CPO, where optical chiplets sit around the GPU or switch. The most ambitious version puts the optics on an interposer beneath the chip itself. “As you progress, you increase the speed and you reduce the energy consumption with each one of those leaps,” said Harris.

For now, the industry’s attention is on the step before the really radical one. “I think 2027 and 2028 are going to be very, very big years for near-package optics,” Harris said. He describes it as the test bed: the last stop before the optical transceiver effectively ships as part of the accelerator or switch, rather than as a separate component bolted on nearby.

Lightmatter is already building CPO parts at TSMC and GlobalFoundries, which it expects to ship in 2028. At the same time, it is working with customers on the more integrated interposer approach – the one for companies that, in Harris’s words, “realise they’d like to have an advantage in the market.”

The reason is power. Bayvel says about 20 percent of a data center’s energy goes into networking: The business of connecting GPUs and CPUs across racks, rooms and data centres. The other 80 per cent is consumed by the GPUs themselves. But she is clear-eyed about what happens to any savings. “Whatever energy you save through optics isn’t going to be there for saving the world,” she said. “It’s to stick another 1,000 GPUs in.”

Warring standards

Nvidia has a vertically integrated answer spanning both domains, from NVLink and NVSwitch for scale-up to Spectrum-X Ethernet and InfiniBand for scale-out. The rest of the industry, including AMD, Broadcom, Cisco, Arista, Marvell, Meta, Microsoft and OpenAI, is trying to prise that open through UALink, Ultra Ethernet and a new optical specification.

The sense that a standards war is building is becoming obvious. At Computex, AMD showed its Helios MI455X platform, a 72-GPU rack meant to rival Nvidia's Vera Rubin NVL72. But its scale-up fabric runs UALink-over-Ethernet rather than native UALink – a stopgap forced by the fact that dedicated UALink switching silicon from partners such as Astera Labs and Marvell was not ready in time. Engineering samples are due in the second half of 2026, with mass production slipping to Q2 2027, while AMD's next-generation MI500-series rack is expected to follow quickly.

Bayvel says the wider contest is one between CPO and linear-drive pluggable optics (LPO). CPO integrates lasers and optics alongside the processing and is probably more power-efficient, “but probably less reliable –so if your laser fails, you need to replace the whole board.” LPO keeps things in conventional pluggable slots and is more serviceable. “There’ll be a brave person who will bet their house on which will win,” she said. She guesses that they coexist, split by application, with a rough 60/40 lean towards CPO.

The laser bottleneck

The problem is that silicon can do almost everything except the one thing this industry increasingly needs it to do: shine.

Silicon is cheap, abundant, and brilliant for electronics. It is why the modern chip industry exists. But it is terrible at emitting light. “Silicon doesn’t emit light at all — that’s a big problem,” said Frederic Gardes, professor of silicon photonics at the University of Southampton, in an interview with Tom’s Hardware Premium.

For decades, the industry has worked around that by using so-called III-V materials such as indium phosphide and gallium arsenide, which are much better at producing light. But that fix brings its own problem. Those materials sit in a much tighter, more geopolitically exposed supply chain, with China controlling around 80 percent of supply. Getting lasers onto silicon is, Gardes said, “one of the key problems” the field now has to solve.

But it’s also where the money is. “Half of the revenue opportunity for this industry is lasers,” Harris said. “It’s a freaking power supply, but it turns out that it’s half of the revenue opportunity.” The irony is that the bit everyone needs most is also the bit that may prove most fragile. Gardes calls lasers “the weakest link.”

That weakness is already showing up in the supply chain. Laser makers Lumentum and Coherent are booked out for years. And the industry is starting to standardise. In March, AMD, Broadcom, Meta, Microsoft, Nvidia and OpenAI founded the Optical Compute Interconnect (OCI) Multi-Source Agreement, a hyperscaler-led effort to define a common optical physical layer for scale-up — one that can carry both NVLink and UALink, starting at 200Gbps per direction and scaling towards 3.2Tbps per fibre.

Whichever route wins, Bayvel worries the industry is fixating on the data center while neglecting the networks that must carry all this AI-generated traffic across land, sea and space. “Really, nobody’s looking at that,” she said, “and I think that's going to be a big problem down the line.”

More than 70% of Americans now oppose data centers, which has resulted in bipartisan moves to block these projects. This pushback isn’t limited to just a few people or towns, either. In fact, a coordinated protest was organized just this weekend, happening in 142 different venues across 42 states, and is the first one to happen simultaneously across the U.S. Forbes reports that this makes community consent scarcer than the chips and other equipment and resources needed to build and run these sites. As Reuters notes, the organisers behind the protests have branded the "unaccountable" data center buildout as an "unacceptable infringement on our liberty."

Data centers have faced shortages of various kinds in the past few years — it began with GPU supply, then evolved into a memory chip bottleneck. Alongside that, there’s also the limited supply of electricity from the grid, as well as the related power infrastructure. But the one thing that many data center developers did not count on was the lack of public support.

This resistance isn’t limited to the streets, as bipartisan political moves have resulted in everything from delays in various projects to statewide bans. More than 69 jurisdictions have already imposed data center bans, with Seattle (home to the headquarters of AI tech giants Microsoft and Amazon) and the state of New York passing one-year moratoriums. Because of this, more than $130 billion worth of data centers have already been delayed in just the first quarter of 2026.

Developers have employed various techniques to get around these bans and opposition. This includes picking sites in rural areas with fewer people (thus less resistance), bringing their own electricity, or claiming that their data centers use less water than lawn care and gardening. But despite all their promises of jobs, investment, and tax revenue, the American public has grown wary of suffering from the consequences of having a data center as a neighbor. This is especially true after reports of increased utility rates, water quality issues, noise pollution problems, and more have widely circulated around the country.

Because of this, any data center project must take the community into account. It can no longer just rely on a friendly town council for approval, especially as the voting public has proven that they are willing to take steps to oust officials that they feel do not represent them. While many know that the U.S. needs data centers to compete in the AI race, they don’t want this to happen at the expense of their quality of life. And while investors can just keep on pouring money into purchasing chips and upgrading infrastructure, they cannot just buy off the community — instead, they need a solid plan that stands up to scrutiny and opposition and accounts for the months or even years it will take for permits to go through and get approved.

After decades of charging only a one-time fee for undersea cables along its coast, Oregon's lawmakers have recently been discussing a missed opportunity for tax revenue. The state is set to charge cable operators a one-time fee for cable installations within three miles of its coastline, and will be funneling that money into its Common School Fund.

The currently existing fee structure dates back to 2001 (25 years ago), and is simply a $5,000 application fee. The new pricing is still undergoing discussion, but the simplified version is $3 per linear foot of cable, plus $7 per each section of bore pipe — the protective structure that covers the portion closest to shore. Oregon's Department of State Lands (DSL) estimates that on a 20-year lease, this would bring in millions per undersea cable over that time span. The DSL did some rough math with Amazon's Bitfrost cable, and came up with a grand total of $1,450,380 for two decades, paid in one lump sum.

It's worth noting that the aforementioned rates were revised from an earlier, more expensive proposal that would take bore cross-section into account (a calculation that was dismissed as too complicated and, presumably, too onerous). The state wants to continue attracting investment to its tech sector, which includes around 140 datacenters and 16 undersea cables of various types. Oregon also doesn't have sales tax, and its enterprise zoning incentives has made it an attractive location for players like Amazon, Microsoft, and Google.

The DSL calls the revised rate "competitive-to-low," reinforcing the notion that Oregon is trying to stay competitive with its southern neighbor, California, which charges an estimated $5 per cable foot on an annual basis.

It's worth noting that spending a million or two over two decades is peanuts compared to the initial outlay required to put an undersea cable down in the first place, especially one coming from across the world. These reportedly ring in at around $250 million for a transatlantic connector — and a whopping $400 million to cross the Pacific.

Although it's not set in stone, it seems that already-approved cables might be almost entirely exempt from the new fees. Some recent contracts, such as the one for Amazon Bifrost, did include a clause that said that any fees applied by new state laws must be paid for. However, the contract also included a specific $300,000 "out" for this clause — which Amazon paid at the time.

Nvidia CEO Jensen Huang thinks that American companies should be allowed to use Chinese AI models, even as Washington is trying to ban them. When Axios co-founder Mike Allen asked Huang in an interview if Americans companies should be allowed to use Chinese AI models, Huang responded with “absolutely.” The answer comes right after Chinese firm Moonshot AI released a 2.8T open-weight model called Kimi K3, which — although it isn’t as powerful as frontier models like Fable 5 — is comparable to GPT 5.5 and Claude Opus 4.8 while costing just a third of these models.

One of the biggest concerns of U.S. leaders have is that these AI models might come with vulnerabilities that the Chinese government can use to attack American interests, but Huang said that this is an incorrect assumption. “There is a misconception that somehow there are backdoors that are somehow connected to China in some way,” said the Nvidia chief. “You download the models, you can fine-tune it, you can enhance it, you can guardrail it as you desire.”

Huang shares the same sentiments about American AI models. Just last month, the U.S. enforced an export restriction on Anthropic’s Mythos and Fable 5, citing security threats — although access was eventually restored after its developer placed a filter to block these tools from identifying software vulnerabilities. OpenAI’s ChatGPT-5.6 received the same treatment, and Washington warned the firm that it should not release its latest model without getting the green light from the government. Huant argues that, instead of restricting access to these powerful models at launch, AI firms should make their models available to all and make them more secure through rapid testing and fixes.

But even as he advocated the need for everyone to have access to closed models, Jensen also noted that various industries, such as the sciences and cybersecurity, need open models as well. He claims that these models make AI more secure, as other people can inspect them to look for weaknesses and fix them as required.

“If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable,” Huang said.

As for the market’s negative reaction every time cheaper, open-weight models become available, the Nvidia CEO says that investors misunderstand their impact. Huang said that this happened when DeepSeek arrived for the first time, and it’s happening again with the arrival of Kimi. He says that these open models, which cost less to run, will encourage more people to use AI. So instead of cutting data center demand, these cheaper, more efficient models are actually good for the industry in general because they will drive demand. (And with higher demand, there’s more incentive to build data centers and buy AI GPUs, which is ultimately good for Nvidia.)

Intel will design, package, and fabricate Fortinet's sixth-generation Security Processor (SP6) on its Intel 4 node, the companies announced on July 21, giving the process its first named external foundry customer, roughly three years after it entered production. Intel told Tom's Hardware the agreement reflects "the strategy Intel outlined for Intel 4 several years ago," including support for custom networking ASIC workloads. Intel's own record from those years reads differently, however, with the company's 2021 roadmap having scoped Intel 4 to two internal products. And through 2022, it told engineers and investors that Intel 3, not Intel 4, would be its first process offered to foundry customers.

Intel 4's record

Intel's Accelerated announcement back in July 2021 said that Intel 4 would reach production readiness in the second half of 2022 for products shipping in 2023, naming "Meteor Lake for client and Granite Rapids for the data center." The release and its accompanying fact sheet, however, contained no reference to foundry customers, networking, or custom ASICs on the node.

At VLSI 2022, Intel disclosed that it wasn't building a high-density library for Intel 4 and that Intel 3 would be the first new node offered through what was then Intel Foundry Services. A 2024 post on Intel's own foundry blog describes Intel 3 as "Intel Foundry's first leading-edge process node," and Intel's fiscal year 2024 annual report listed the processes available to external customers as 18A, Intel 3, Intel 7, Intel 16, and a 12nm node co-developed with UMC. Intel 4 appears nowhere on that list.

Ericsson's RAN Compute processors, announced in November 2023, were built on Intel 4, so Fortinet's part won't be the first third-party silicon to come off the node. That work grew out of a bespoke Intel-Ericsson collaboration, though, and Ericsson's formal foundry agreement with Intel, announced in July 2023, covered 18A. Fortinet is the first named customer buying Intel 4 as a foundry service, and the first cybersecurity vendor on any Intel node. The Ericsson engagement is also the closest thing in the public record to networking silicon on Intel 4, two years after the strategy Intel now says it outlined for the node.

Fab 34 economics

Intel 4 entered high-volume manufacturing at Fab 34 in Leixlip, Ireland, in September 2023, producing the compute tile for Meteor Lake-based Core Ultra chips, and shares the fab with Intel 3. Intel sold a 49% stake in the facility to Apollo-managed funds for $11.2 billion in June 2024, then bought it back in April 2026 for $14.2 billion, funded with $7.7 billion in cash and $6.5 billion in new debt. That buyback returned 100% of Fab 34's wafer economics to Intel at a premium of roughly 27%, and it only pays off if the fab's EUV capacity stays loaded.

Meteor Lake is aging out of Intel's lineup as 18A-based Panther Lake ramps through 2026, which leaves open the question of what fills Intel 4 capacity next. A multi-generation firewall ASIC program is a reasonable answer with mature yields, a customer that values supply stability over bleeding-edge density, and a part Intel described as tailored for cost-sensitive applications. Intel said in April that yields were improving across Intel 4, Intel 3, and 18A.

Fortinet's supply chain

Fortinet's 2025 annual report names Renesas and Toshiba America as the contract manufacturers for its ASICs, utilizing foundries in Taiwan and Japan operated either by TSMC or by the contract manufacturers themselves. The current SP5, a monolithic 7nm Arm-based SoC announced in February 2023, sits in that supply chain, so SP6 on Intel 4 moves Fortinet's next flagship security processor out of a TSMC-linked flow and into Intel's. The disaggregated design language in the announcement points to a chiplet-based part, a first for Fortinet's SP line.

Fortinet re-engineered three FortiGate models in 2022, the 70F, 600F, and 3700F, to accept alternative components during the chip shortage, and CMO John Maddison told SDxCentral at the time that the company wouldn't wait for parts to arrive in 2023. The "resilient and diversified" supply chain used in the SP6 press tracks back to that experience. Ken Xie called Fortinet "the #1 firewall leader with a 55% unit market share" in the company's 2025 results in February, with approximately six million FortiGates deployed, so there’s real, substantial volume here even if the parts are relatively inexpensive.

Intel Foundry reported $307 million in external revenue for 2025, up from $159 million the year before, against total foundry revenue of $17.8 billion and an operating loss of $10.3 billion. External revenue in Q1 2026 was $174 million. Fortinet's hardware business runs at roughly 30% of its revenue, and, per analysis from ServeTheHome, SP6 is ultimately a component of a portion of an annual hardware stream around $2 billion, so the deal won't move Intel's foundry line materially, even at full production.

CEO Lip-Bu Tan told CNBC in May that he expected commitments from multiple foundry customers in the second half of 2026, and Intel told investors in January that two prospective customers were evaluating 14A test chips. The SP6 announcement comes inside Tan's stated window, and it finally gives Intel something its foundry marketing has lacked in a named customer with shipping volume on a node with mature yields. Meanwhile, an 18A or 14A commitment from a major external customer is still missing, and Fortinet's cost-sensitive parts on a 2023 node don't substitute for one.

AMD will deploy up to 2 gigawatts of Instinct MI450 Series GPUs for Anthropic and make a strategic equity investment of up to $5 billion in the Claude developer; the companies have announced in a joint press release. The first gigawatt is scheduled to come online in the first half of 2027 in AMD Helios rack-scale systems, and the same announcement confirms for the first time that Anthropic already runs AMD's current-generation MI355X GPUs.

The equity component runs in the opposite direction to AMD's landmark deal with OpenAI, which handed the customer warrants for up to roughly 10% of AMD's stock. Here, AMD is putting money into its buyer, becoming both supplier and shareholder.

Anthropic's deployment pairs Instinct MI455X GPUs with EPYC "Venice" CPUs, Pensando networking, and ROCm software inside Helios racks, each of which packs 72 accelerators, 31 TB of HBM4, and up to 2.9 FP4 exaFLOPS of inference compute. The two companies also launched a multi-year engineering collaboration that will use Claude to optimize workloads for Instinct GPUs and speed up ROCm development, while AMD adopts Claude across its engineering and product teams.

"Running across a diversified range of hardware lets us map the right workloads to the right hardware," said Tom Brown, co-founder and chief compute officer at Anthropic, in the press release.

SemiAnalysis reported on Sunday that a YAML configuration file in an AMD senior director's public GitHub repository listed Anthropic as a 'customer' with the maximum 30 priority boost points, the same tier as hyperscale customers, such as Meta. Anthropic wasn't a publicly confirmed AMD customer at the time; its disclosed suppliers were Nvidia, Amazon, and Google. Today’s press release states the Helios deployment "builds on Anthropic's existing use of AMD Instinct MI355X GPUs," the first official confirmation that Claude workloads already run on AMD silicon.

Anthropic's first AMD gigawatt comes behind two other Helios customers in the queue. OpenAI's first gigawatt of AMD capacity is due in the second half of 2026, and Oracle is due to begin deploying 50,000 MI450 GPUs this quarter.

The 2 gigawatt commitment adds to a compute portfolio that already spans Nvidia GPUs, Amazon's Trainium chips under Project Rainier, 3.5 gigawatts of Google TPU capacity supplied by Broadcom from 2027, and 300 megawatts of Nvidia compute rented from SpaceX's Colossus 1 data center, making AMD roughly the sixth distinct silicon source feeding Claude.

The announcement is also AMD's second major Helios customer announcement inside a week, after Microsoft committed on Monday to deploying the racks at scale on Azure.

U.S. data centers are on track to consume about 20% of the nation's electricity by 2035, up from 5.9% today, according to a new BloombergNEF forecast The research firm raised its 2035 demand projection to 194 GW, an 83% increase on the forecast it issued in December, and calculated that even a record pace of grid connections sustained every year for a decade would leave a 19 GW supply shortfall.

BNEF's December outlook put 2035 demand at 106 GW, and that figure was itself 36% above the projection the firm published in April 2025, which works out to roughly 78 GW. The 2035 number has increased around 2.5 times in 15 months, with the latest jump reflecting the volume of new AI facilities entering the development pipeline rather than capacity actually under construction. Many other forecasters are moving in the same direction, with the electrical industry nonprofit EPRI more than doubling its 2024 estimate and S&P raising its projection by more than a third between October and April.

Nearly half of the projected capacity will serve AI training and inference workloads, and the U.S. is expected to host 64% of the world's AI chips by power demand in 2033. BNEF estimates that data centers will account for around 12% of U.S. electricity consumption in 2030 before reaching 20% five years later, with Virginia and Texas running above the national average. Gartner separately forecast that global data center electricity consumption will grow 26% this year, reaching 565 TWh in 2026.

The most data center capacity ever connected to the U.S. grid in a single year is 7.1 GW, according to the report. Holding that pace through 2035 still produces a 19 GW shortfall under BNEF's base case, even after accounting for on-site gas generation. Lloyd Arnold, a BNEF analyst and one of the report's authors, told Bloomberg that of the output from "every coal plant, every gas plant, every solar farm in the U.S.," one unit of energy in five will go to data centers.

PJM Interconnection, the 13-state grid operator whose territory includes Northern Virginia's data center corridor, will send 34% of its electricity to data centers by 2035, with ERCOT in Texas at 22%, per reporting from TechCrunch. PJM's independent market monitor has already attributed a 75.5% increase in regional power costs directly to data center demand.

Close to half of planned U.S. data center builds this year are projected to be delayed or canceled, with lead times for high-power transformers stretching to as long as five years. BNEF's own December revision was driven primarily by early-stage projects that entered utility queues to secure power before construction began, so a meaningful share of the pipeline behind the 194 GW figure remains speculative.

Following unprecedented price hikes on the Mac and iPad, Apple is now working on a leasing program to alleviate customers during the AI rush. According to a new Bloomberg report, the program is called "Apple Upgrade" and will allow you to lease a midrange or high-end device for up to 36 months, at the end of which you can choose to keep it. Apple is purportedly partnering with Klarna on the service.

Apple Upgrade will reportedly be a subscription service with monthly payments spanning across 24 months for iPhone and Apple Watch, or 36 months for Mac and iPad, including the Pro models. According to the report, Entry-level hardware like the MacBook Neo is not included in this program. Education or business discounts are also excluded; Apple is explicitly targeting regular consumers who've been affected by the ongoing component crisis.

According to Mark Gurman, customers will be subject to a soft credit check and only be approved for the program after passing it. The program will be available both in-person at physical Apple Store locations and online on Apple's website. At first, it will only be available in the United States, but we don't have any info on whether Apple Upgrade will expand beyond the region later.

As mentioned, Klarna is reported to be the official broker between Apple and the customer. The leasing structure would closely follow standard practice, where you can choose to keep the asset at the end after paying a residual fee, or return it, in which case you were just renting the hardware. Customers will also reportedly have the option to upgrade to a newer model if they finish their payments early.

For instance, if you leased an iPhone 17 Pro Max but Apple announced the 18 Pro Max a few months later, you could simply get the new device early if the old one was already paid off. Some of you will know that Apple already has an upgrade program for iPhone, but this one is different since it expands to Mac, iPad, and other accessories, and, more importantly, it doesn't come with AppleCare.

In fact, Bloomberg says Apple will eventually phase out the iPhone Upgrade program in favor of Apple Upgrade as a whole. New enrollments for standard financing through the Apple Card will also reportedly be discontinued. The company will advertise lower monthly payments to attract customers, which will ultimately allow it to keep a stable income during one of the most turbulent consumer tech eras the world has ever seen.

Five U.S. tech giants heavily invested in AI and its related infrastructure reportedly have an estimated $1.65 trillion in hidden debt, with the figures annotated in their quarterly financial statements instead of being listed in their balance sheets. According to Nikkei Asia, this is higher than the $1.35 trillion officially listed, meaning investors could be caught unaware once the hidden figures come to light.

The publication says that Meta has a high off-balance-sheet-to-recorded-debt ratio, with the company owing $420 billion in unlisted debts compared to the $140 billion written on the balance sheet. Oracle purportedly also has a massive $273.3 billion of hidden debt, which is a 2,900% jump from the hidden debt it had from 2022.

This may sound strange, but it’s actually an accepted accounting practice. The “hidden debt” stems from long-term contracts that have been signed but have not come into force yet, which, Nikkei says, is mostly related to the billions of dollars promised to data center operators. The AI race has got many hyperscalers signing contracts and agreements with data center operators, saying that they will pay for the compute they generate once their project goes online.

While any institution promising to pay any amount of money for services or goods delivered is obliged to list them as a liability, the fact that these data centers haven’t started operations means that these agreements are off-the-books at the moment. But when these projects come online, the contracts that the tech giants have signed will come into force, and they’ll have to pay for the compute that these sites will deliver, no matter if there is demand or not.

Nevertheless, these tech companies aren’t just pouring money into future contracts just for the sake of it. Alphabet, Amazon, and Microsoft reportedly have a cloud service backlog worth $1.45 trillion, meaning these are services yet to be rendered and paid. Amazon Web Services CEO Matt Garman also told the publication that the investments that the company is getting into are “not speculative.”

While this may seem like a good way to secure capacity — sign customer contracts that guarantee demand and then enter into long-term agreements with data centers to get the compute needed to deliver the services- it opens these tech giants to massive amounts of risk. That’s because if the demand fails to materialize, then they’d be left paying for excess compute without having any customers to sell them to. What’s more alarming is that Nikkei says that these investment expenditures are exceeding their earnings, meaning these big tech companies are increasingly relying on corporate bonds and new shares to fund them.

Even though demand for AI compute is increasing, it’s still a relatively new and unproven technology, with many experts saying that it should benefit more people to avoid a bubble. The cost of using AI for nearly everything, called “tokenmaxxing,” has also caught some companies by surprise, with agentic AI eating up annual AI budgets in a matter of weeks. Because of this, some companies are reducing their use of AI or are switching to more affordable models from China. This uncertainty, paired with the way tech companies “hide” these liabilities, is quite concerning, as they would appear to have less long-term obligations than they actually do.

This isn’t the first time that an industry giant has used similar accounting techniques. The publication cited Enron’s 2001 collapse, which was due to the company hiding its troubled assets through special purpose entities and marking unrealized gains from trading contracts into its current income statements. While the tech giants are not hiding underperforming assets off their balance sheets and committing fraud, they’re still using a similar mechanism to list their upcoming obligations. Although these are technically not debt, they still behave like one, and the way they’re reported is what’s concerning some experts.

AMD's custom Instinct MI450-based AI accelerator for Meta will use three times less memory than the fully-fledged Instinct MI455X and will be optimized primarily for recommendation systems operated by Facebook and other social platforms, according to SemiAnalysis. If the report is accurate, it is reasonable to expect Meta to keep using Nvidia hardware for training frontier AI models and running inference.

The custom Instinct MI455X for Meta will carry 144GB of HBM4 memory using six 8-Hi packages, whereas the full-blown Instinct MI455X will be equipped with 432 GB of HBM4 memory, according to SemiAnalysis. In addition, the part will reportedly offer 'significant decreases in compute.' The new design will offer a more competitive bandwidth-per-dollar ratio for recommendation systems, but will not be optimized for training of frontier AI models or running inference, the report claims.

Cutting compute performance and reducing HBM4 capacity from 432GB to 144GB should dramatically reduce the bill of materials, as HBM4 is exceptionally expensive. Furthermore, the reduction would cut the package size of the custom Instinct MI450-series accelerator for Meta, which is another way to reduce BOM costs. By using custom cut-down Instinct MI450-series accelerators instead of fully-fledged models, Meta can potentially save tens of millions of dollars.

As added bonuses, these custom Instinct MI450-series accelerators will also consume significantly less power when running recommendation workloads without significantly reducing performance. Also, such accelerators can offer better CPU/GPU balance for recommendation systems, according to SemiAnalysis. If Meta runs these accelerators primarily on recommendation workloads for their entire useful lives, the custom design could deliver substantially better total-cost-of-ownership.

However, such cutting down has many disadvantages. The biggest problem is loss of versatility. The reductions in both compute and HBM make it less attractive for LLM training and inference. The standard Instinct MI455X has 432 GB of HBM4 and 19.6 TB/s of bandwidth, which is particularly beneficial for large-scale training and inference. By contrast, the 144 GB capacity may be particularly restrictive for modern LLM training and inference.

In addition, there is also an interchangeability problem. A general-purpose Instinct MI455X can be reassigned from recommendation workloads to training, inference, or other workloads. Meta's specialized version is less attractive outside its intended workload. If Meta's compute demand shifts toward model training and LLM inference, it may find itself sitting on a huge installed base of accelerators optimized for a different workload mix.

As a result, for Meta's model training and inference workloads, the alternative to Meta's cut-down custom MI400 would likely be full-fat AMD Instinct MI455X systems or Nvidia's high-end platforms. Meanwhile, Nvidia has chances to become an obvious beneficiary because Meta already operates massive Nvidia infrastructure. The irony in that Meta customized an AMD accelerator to reduce costs and optimize recommendation systems, but that specialization could force its frontier AI division to buy more general-purpose accelerators — potentially from Nvidia — anyway.

When AMD and Meta inked an agreement under which the former will supply the latter with 6 GW of Instinct AI accelerators over the next five years, they did disclose that at least some of them will be custom accelerators, including custom accelerators based on the Instinct MI450 design. As it seems now, these custom AI accelerators will only be used for select workloads, not a broad set of workloads.

Adata chairman Simon Chen (Chen Li-bai) said the global DRAM shortage will run for another 10 years and that talk of an AI bubble is premature, according to a Commercial Times report. Chen made the comments after Taiwanese stocks slumped in the wake of TSMC's record second-quarter results, a selloff that revived concerns about overheated AI investment. The same report carries Adata's forecast that DRAM contract prices will climb another 20% to 30% in the third quarter, with NAND flash rising 35% to 40%.

Electricity, particularly green power, and memory will be the world's two scarcest resources over the next decade, Chen told Commercial Times. He rejected the argument that Meta and other cloud providers renting out spare compute capacity prove investment has overshot demand. Chen countered that AI applications will spread across business, government, and consumer markets, and that analysts who judge the cycle from short-term capital spending or a single company's utilization rate underestimate long-term demand.

Beyond data centers, he pointed to robots, driverless vehicles, unmanned factories and stores, smart homes, and low-Earth-orbit satellites, a pool of edge devices he said could reach tens of billions of units. He also said that the industry can revisit the bubble question after 2030, to "discuss whether the AI bubble will happen in 2040 or 2050," per the report.

Meanwhile, memory makers aren't standing still on capacity. SK hynix raised a record $26.5 billion in its U.S. IPO this month to fund HBM manufacturing expansion, Nanya plans to quadruple capital spending to $6.2 billion in 2027, and China's CXMT is reportedly approaching Micron-scale DRAM output this year.

Chen argues that none of it loosens the market, and that the big three suppliers learned from past downturns and will expand cautiously rather than repeat the disorderly capacity races that cratered prices in earlier cycles. At the same time, Chinese makers remain constrained by restricted access to semiconductor equipment and multi-year fab construction timelines. On that basis, he said current talk of supply and demand easing has no solid foundation, and he expects prices and module-maker profits to keep rising through the second half.

Chen's forecasts align with his company's balance sheet. Adata had stockpiled more than NT$30 billion in chip inventory by late February and was among the Taiwanese module makers that raised roughly $880 million in debt and share placements to keep buying chips, so every quarter of rising contract prices lifts the value of what it already holds. Back in October, he said the simultaneous shortage of DRAM, NAND, SSDs, and hard drives was the first he'd seen in 30 years in the industry, and DRAM contract prices have since risen as much as 171% year over year.

Not too long ago, Anthropic CEO Dario Amodei described Claude Mythos as capable of cyber-warfare, spawning all sorts of mythology that became popular reading at investors' desks, and even at the U.S. government table, which issued an export control order for it. Now, it's OpenAI's turn to do the same, with a blog post describing how a bot cyber-gang including its upcoming GPT-5.6 Sol and an "even more capable pre-release model" all broke out of their virtual cage and eventually into Hugging Face's production infrastructure, during an attack capability test. OpenAI described the snafu as an "unprecedented cyber incident."

Some may decry these revelations as marketing stunts meant to assuage increasingly suspicious investors, yet there's no denying the proven effectiveness of AI models' pattern recognition when it comes to finding vulnerabilities in software. Many professionals have even decried the industry-standard 90-day security vulnerability disclosure window as obsolete.

The usual modus operandi, however, involves granting a bot access to source code repositories so it can scan them for mistakes easy for our poor human eyes to miss. In the test that OpenAI described, though, Sol and its managed to find multiple vulnerabilities by analyzing its containment network and then Hugging Face's production machines, all seemingly without any initial access to source code.

OpenAI had set up an isolated environment in which Sol and its posse supposedly couldn't reach the internet. Importantly, the bots were operating "without production classifiers used to prevent models from pursuing high-risk cyber activity" — a lot of words for "no guardrails," essentially. The only network access they had was a software package installer proxy. In theory, this likely also prevented them from using any Retrieval-Augmented Generation (RAG), meaning they couldn't search for solutions online.

Given that bots have limitless patience and no need for naps, they eventually dug their own virtual tunnel to the internet at large by way of finding and exploiting a zero-day vulnerability in the package proxy software, an impressive feat if performed as described. OpenAI says it's disclosed the technical details to the vendor in question.

Once out on the internet, much like cheating students, the bots reportedly reasoned that instead of performing the ExploitGym challenges (a set of security capture-the-flag exercises for bots), they could simply find the solutions, and became "hyper-focused" on that exact goal. They deduced the data existed at Hugging Face's servers and proceeded to break into the premises using "stolen" credentials and additional unspecified zero-day vulnerabilities to gain remote code execution privileges.

Interestingly, Hugging Face's own writeup says the attack comprised "thousands of individual actions across a swarm of short-lived sandboxes, with self-migrating command-and-control staged on public services" — a live illustration of a scenario in many a science-fiction book and flick, proving once again that life imitates art. Hugging Face says that none of its customer-facing services are compromised, and that it stopped the attack using AI capabilities of its own — another familiar cliché.

For its part, OpenAI says it's going to add controls to the bot "at the cost of research velocity," and that the incident "points to the need to further strengthen our model’s alignment, cyber protections during evaluation time, and monitoring during internal testing" — a statement that might be an obvious fact, or intended to oversell the model's capabilities. Whichever it may be, the fact is that when it comes to digital security, bots have proven quite capable. After all, even if they're not virtual Bruce Schneiers, they just need to be marginally more effective than average humans.

China is considering a major expansion of its technology export restrictions that could cover advanced AI models, training data, and overseas acquisitions of strategically important technology companies, reports the Financial Times. In addition, the Chinese government is mulling over prohibiting local chip designers from making their chips at TSMC and other foreign chipmakers.

The measures would be designed to keep leading-edge AI developments in China as competition with the U.S. in frontier AI and hardware intensifies, but at the same time, they would slow down expansions of Chinese AI standards globally, which generally weakens the country's position.

China's Ministry of Commerce (MofCom) has consulted domestic AI and semiconductor companies about ways to keep critical technologies from transferring abroad or falling under Western control, reports Financial Times citing two people familiar with the talks. Regulators have talked with Alibaba, ByteDance, and Zhipu about potentially limiting transfers of important AI training data outside China and restricting foreign users from downloading model weights.

Overseas customers could still access Chinese AI services and models remotely, so Chinese companies can still monetize their work from foreign customers. However, restrictions on downloadable model weights could still have significant implications for China's AI industry. DeepSeek and Moonshot offer open-weight models that users can download, deploy on their own infrastructure, and modify for specific workloads. Meanwhile, flagship models from Anthropic and OpenAI remain closed, which means that Chinese companies have an edge over rivals that they are about to lose.

In addition, MofCom has reportedly asked for industry feedback on possible restrictions that would prevent overseas chipmakers like TSMC from producing advanced processors based on designs developed by Chinese companies such as Alibaba, ByteDance, and Huawei. This is perhaps the most controversial proposal, as TSMC is clearly ahead of SMIC when it comes to process technology leadership. On the one hand, the move ensures that SMIC will have enough orders to pay for its R&D and expansion. On the other hand, Chinese companies can get better hardware if it is produced by TSMC.

Separately, the Chinese government is considering tighter controls over foreign acquisitions of strategic technology companies, including firms that work on agentic AI technologies. The potential acquisition rules are intended in part to close what Beijing considers a regulatory loophole that enabled Meta to acquire Manus for $2 billion. Chinese authorities subsequently ordered the transaction to be undone.

The measures could be included in the next revision of China's catalogue of technologies prohibited or restricted from export. The catalogue already includes rare-earth materials, their processing technologies, and several lithium-ion battery production technologies.

Amazon’s Bahrain data center was reportedly once again hit by missile strikes after the Iranian Islamic Revolutionary Guard Corps (IRGC) conducted an attack against the site. Iran state media claimed that the IRGC launched cruise missiles against the AWS facility and “destroyed it” in response to the U.S.’s bombing of a nuclear plant being built in Darkhovin, Iran, which is close to the Iraqi border.

This isn’t the first time that the American tech giant’s data centers have been attacked in the area. In fact, Iran used drones and missiles to hit several AWS Middle East regional data centers just days after the U.S. began its air strikes against the country. The damage to the sites has gotten to the point that Amazon put them in “hard down” status between late March and early April, with the last update on the AWS Health Dashboard saying that the Middle East (Bahrain) Region (ME-SOUTH-1) “is currently unavailable.”

It isn’t clear if Amazon has begun work on restoring its services in the region. The U.S. and Iran agreed to a ceasefire on June 17, 2026, but the entire thing collapsed after the latter allegedly attacked three ships in the Strait of Hormuz that were sailing in Omani waters on July 7. This led to an escalating series of retaliations, with one of the recent Iranian strikes resulting in American casualties in Jordan. Aside from hitting the nuclear powerplant that’s still under construction, U.S. Central Command said that it also “targeted Islamic Revolutionary Guard Corps forces that launched attacks against U.S. service members in Jordan on July 17.” The IRGC said that this is what led to its most recent attacks on AWS Bahrain.

Despite the strikes, it seems that it had no effect on Amazon’s services — the company has long moved its customers to sites unaffected by the conflict, ensuring service continuity. While it’s possible that the tech giant is trying to repair the damage to its sites, it hasn’t released any updates about the AWS Bahrain data center. As long as the war between the U.S. and Iran hasn’t resulted in lasting peace, American infrastructure in the region is in grave danger of being attacked, especially as the IRGC has directly threatened American tech companies like Microsoft, Apple, Google, and even OpenAI’s $30-bilion Stargate data center in Dubai.

Nvidia invited a group of about a dozen journalists out to the company's HQ to learn more about its Vera CPU, as well as how it fits into the larger Vera Rubin NVL72 rack design at the heart of Nvidia’s next-gen agentic AI platform. Part of that was seeing Vera Rubin in action, not as a disassembled tray on stage or a rack with a few blinking lights at a trade show — real racks running real workloads in a (partially) real data center. And we got to see those racks in action at Nvidia’s Engineering SuperLab.

It’s not a proper data center, or at the very least, it’s a sub-optimal data center. Nvidia was clear that the Engineering SuperLab is built for engineers, allowing them to quickly stand up and swap out racks to see the hardware in action. You could sense a bit of insecurity in the air; if Nvidia were building a proper data center, it wouldn’t look like this. This lab is where the engineers live, and if you’ve ever been around a group of engineers with a lot of hardware to play with, you know that things aren’t always as tidy as you’d expect in a proper data center.

The SuperLab Nvidia showed us is one of four nondescript locations near Nvidia HQ. These locations haven’t, up to this point, been disclosed. Each of the four locations has popped up over the last two years, giving Nvidia some floor space to play with as it rolls out new hardware.

The hardware in question here is the Vera Rubin NVL72 rack, but we saw a few other demonstrations, as well. Most notably, Nvidia showed us a sidecar running 800VDC power into an NVL72 rack. Nvidia also laid out some parts, demonstrating the assembly process for a Vera Rubin tray, which slides together with various retention arms in a matter of minutes.

This is a look behind the scenes of our tour, how the Engineering SuperLab is set up, and some choice data center eye candy. We’ve published a full breakdown of the Vera CPU and how it fits into Nvidia’s larger AI infrastructure, which goes into the technical details of the platform. Here, we’re mainly giving you a peek behind the curtain.

Nvidia Vera Rubin NVL72 running in the flesh

Vera Rubin is in full production, and we’ve seen some short videos of racks being stood up in data centers. But this is our first look at a rack running a real workload in the flesh. Nvidia says the racks here are running some workloads for OpenAI, in fact, and as you can see from the image above, it looks like some trays are running OpenAI’s GPT‑Rosalind model.

On the front of each tray here, you can see the ports for the dual ConnectX-9 NICs, along with the Bluefield-4 DPU in the middle. Around the back is Nvidia’s NVLink spine, an almost mediaeval-looking contraption, with sharp pins that connect the various trays together. It houses 5,000 copper cables that measure over two miles in length, delivering up to 3.6 TB/s of bandwidth per GPU and 260 TB/s of scale-up bandwidth per rack, on Nvidia’s sixth-gen NVLink.

Each Vera Rubin NVL72 rack houses 18 compute trays and 9 NVLink switch trays, the latter of which orchestrate communication between the various trays to function as one large, unified system. Nvidia demonstrated the MGX NVL design for us, which is a single reference rack with 72 Rubin GPUs and 36 Vera CPUs. Nvidia’s MGX ETL design replaces the NVLink spine with either a Spectrum-X Ethernet spine or direct chip-to-chip spine for a scale-out system featuring up to 256 GPUs.

The business-end of things is around the back of the racks, though. Although the back is clear of cabling thanks to the NVLink spine, power and coolant delivery are still a major factor. The organized chaos of the piping and cabling you can see in the images below shows just how much goes into standing up even one rack, let alone dozens or hundreds in a data center.

Nvidia’s previous-gen GB200 and GB300 NVL72 racks featured hybrid cooling, but Vera Rubin trays are entirely cooled by liquid. There aren’t any fans, which Nvidia says could, eventually, lead to much quieter data centers. That wasn’t the case in the lab here, which still called for eye and ear protection. I didn’t have a decibel meter handy — imagine if I carried one around with me casually — but my guess is that it was somewhere around 80 to 90 decibels inside; louder than an A/C unit, but quieter than a motorcycle. That’s a fairly typical noise level for a data center.

Regardless, Vera Rubin trays are entirely liquid cooled, with a process called “dry cooling.” Assuming the inlet temperature of the coolant is 45 degrees Celsius or less, Nvidia says it’s able to cool the entire system with a heat exchanger. If true, that would cut costly (in terms of power, space, noise, water consumption, and actual dollars) chillers out of the cooling equation.

Massive piping brings the coolant in (usually antifreeze or deionized water) at the top, and there are outlets at the bottom of the rack to move the warmed coolant out. Along the way are a series of inlet connections that allow trays to automatically hook into the cooling system, leaving just the main connections at the start and end of the loop. Nvidia has standardized everything on a Vera Rubin NVL72 rack with the Open Compute Project (OCP), even as far as shipping specifications. The company says just 47 minutes passes from when the truck pulls up to powering on the rack.

800VDC power for next-gen AI infrastructure

Nvidia also showed us an 800VDC “sidecar” in action. If you’re unfamiliar, Nvidia (along with other AI infrastructure companies) have been pushing for a new 800VDC power delivery system for modern data centers to reduce AC/DC conversion inefficiencies, as well as deliver the necessary wattage to racks without pushing into current ranges of thousands of amps.

A single Vera Rubin NVL72 rack can easily consume over 200 kW, which is a challenge for traditional power infrastructure in a data center. With current Grace Blackwell racks, power shelves convert the AC power coming into the facility (at 415V or 480V) to 48V/54V Direct Current for distribution within the rack. The problem here is pretty straightforward. If voltage stays constant, and wattage increases, then current also needs to increase. And more current means thicker bus bars, more conversion inefficiencies, and more rack space dedicated to power delivery.

Again, the solution that 800VDC represents is pretty straightforward. If wattage increases and current stays constant, voltage also has to increase. The idea is to convert AC power from the grid to DC power once, and then use a series of DC-to-DC converters within the rack, allowing for denser compute and better power efficiency.

This is not an easy problem to solve, as data centers need to change how power is brought into the facility, not just how it’s converted, stepped down, and moved around. Here, Nvidia showed off a sidecar, which is a rack filled solely with power equipment. This is a “retrofit,” as analyst firm SemiAnalysis calls it, representing the first in a series of transition phases to 800VDC. 415V/480VAC is still distributed throughout the facility, but it flows into this sidecar rather than power supplies within the rack. The sidecar rectifies the 415V/480VAC to 800VDC and feeds adjacent racks.

This is all an explanation to show some pretty interesting power infrastructure at play in this lab. Nvidia isn’t the first, nor only, company pushing toward 800VDC infrastructure, and companies like Google, Meta, and Microsoft have contributed to open sidecar designs like the Mt. Diablo spec. Still, it’s interesting to see one of these sidecars in action.

If you haven’t seen a peek behind the back end of a rack, you can see the large red connectors in the gallery above that bring power into the racks. You can also see the massive power cables and connectors at the rear of the 800VDC sidecar.

Nvidia’s next-gen AI infrastructure laid out

At the front of the facility, Nvidia laid out all of the components of its next-gen AI infrastructure. There’s nothing here we haven’t already seen before, though it's normally seen buried in trade show displays or featured on stage during a keynote.

First is the Vera Rubin NVL72 tray itself, which you can see above sitting next to a GB300 tray. Both are DGX designs, meaning they’re fully built and integrated by Nvidia, and you can see just how stark of a difference there is in assembly right away. The Vera Rubin NVL72 tray features only two cables, no hoses, and no fans. At the rear where the two Super Chip boards live, Nvidia demonstrated a retention arm that allows the boards to slide in and out in a matter of seconds.

The Vera Rubin tray is much cleaner, but it also makes better use of the space. It doesn’t include fans, which you can see take up a significant section in the middle of the tray in the GB300 design. With Vera Rubin, those fans are replaced with the thick midplane you can see above, offering a communication channel between the two ConnectX-9 NICs and Bluefield-4 DPU at the front of the tray.

Holding one of the two cables inside of a Vera Rubin NVL72 compute tray is the busbar, which handles power routing for the different chips inside the tray.

The Vera Rubin NVL72 tray is, of course, not the only deployment of Nvidia’s next-gen AI hardware. The company also showed us a CPU-only tray with eight Vera chips, offering up to 256 chips within a rack. That gives us a closer look not only at the chip itself, but also the SOCAMM2 LPDDR5X memory system, offering similar density and modularity as traditional RDIMMs at a far lower power cost.

Although most of the modules were unlabeled, we were able to snag the picture you can see above showing where the modules came from. These are 128GB SOCAMM2 modules from Micron running at 6400 MT/s, and as you can see from the photo, the slots aren’t all populated. This is the big advancement with SOCAMM2, offering up modularity for a memory standard that is otherwise soldered.

Nvidia’s NVLink handles scale-up communications, split across NVLink switches in the rack and the NVLink spine that you can see above. Nvidia describes sixth-gen NVLink as “putting the oven in the car.” Its goal is to get all of the chips in a rack communicating with each other, ensuring critical operations (like baking the pizza) happen as close to the destination as possible.

Scale-out communication, on the other hand, is handled with ConnectX-9 NICs in the tray and the Spectrum-X CPO (co-packaged optics) switch tray. The Spectrum-X switch tray is massive, and it’s a good illustration of just how much physical space is dedicated to communication (over raw compute) in a modern AI data center.

Of course, the size of the switch tray depends on how wide the scale-out infrastructure is. Nvidia also showed a smaller Spectrum-X CPO tray for smaller deployments. That's all we managed to see at Nvidia's AI data center SuperLab. Vera Rubin is in production, and will roll out in the second half of 2026.

A new AI model from Chinese firm Moonshot AI has had its "DeepSeek moment," causing major disruption in the global AI market and spooking Western developers. Kimi K3 is an open-weight model, with 2.8 trillion parameters, making it the largest open-weight AI model ever released. Internal benchmarks have it competing with models like GPT 5.5 and Claude Opus 4.8, and Arena.ai awarded it the number one spot in its Frontend Code Arena test, even beating out Claude Fable 5.

It doesn't win every benchmark, and reports that suggest Kimi K3 is much slower than frontier models from companies like Anthropic and OpenAI. All benchmarked results are drawn from API access, too, so can't be verified until Moonshot releases the weights on July 27.

But that hasn't reduced the impact of this model's release on the AI industry. With Kimi K3 cutting costs compared to the competition, it's drawing a lot of interest from companies hoping to reduce AI spend. For comparison's sake, OpenRouter tables Kimi K3 at $3/15 per million inputs and outputs. OpenAI's GPT 5.6 Sol is more expensive than that, at $5/30, and Anthropic's Claude Fable 5 is $10/50. So, it's fair to say that Kimi K3 is incredibly competitive on price, especially when tabled against the costs of those closed-source Western AI models.

Microsoft is also considering Kimi K3 for Copilot, while the White House may ban Chinese models entirely. Meanwhile, memory makers are rubbing their hands together with glee, as Kimi K3 occupies up to 1.4 TB of memory, given its huge number of parameters.

Fast, cheap, or American?

The past few months have been full of talk about the frontier AI models from Anthropic and OpenAI. Mythos was big and scary until OpenAI had something equivalent. Then Fable debuted, and it was even better but not so scary anymore. Apparently.

But these models were also proving very expensive to run, at a time when companies with big AI deployments were questioning the return on that investment. Uber limited AI use by developers, and others killed the AI-boosting leaderboards they'd championed towards the end of 2025.

So when Moonshot debuted Kimi K3 with running costs a third that of western frontier models for the same results, the world took notice. In much the same way as DeepSeek's R1 debut in 2025 showcased how models could be trained for less -- even if there may have been some corporate espionage involved -- and Kimi K3 is holding up a similar mirror to Western frontier developers.

Where DeepSeek R1 was lean, though, Kimi K3 is huge -- so large, the developers are calling it the first open 3T-class system, and China's largest AI model to date. According to Bloomberg's sources, its sparsity ratio is the highest yet seen by any AI model. That's the measurement of how many parameters are activated for each task relative to the model's size, showcasing both Kimi K3's overall size and its impressive efficiency in the same breath.

This doesn't eclipse the most capable models from companies like Anthropic and OpenAI in every test. Arena.ai's rankings put it within the top 10 on most of its tests, but only coming out on top in a couple. But if Kimi K3 can offer results comparable to more expensive alternatives like Claude and ChatGPT, it's likely to draw a lot of interest from Western companies, and it appears to be already doing so.

Enough that it's revived calls for the U.S. to gatekeep access to international AI models, in a similar manner to how it recently pushed for companies to share exclusive model access with the U.S. government before a wider release.

In comparison, Moonshot is opening up Kimi K3 to the wider world. As part of releasing the model weights to the public, it will allow companies and organizations to run the model themselves without using Moonshot's cloud services, making adoption easier and potentially cheaper.

But it won't be cheap, as Kimi K3 still needs serious hardware investment to get up and running, by virtue of its massive VRAM requirements alone.

A Win for (Chinese) Memory Makers

As large companies with major AI deployments began to scale back their AI initiatives in 2026, there's been a growing concern that all that infrastructure everyone's been spending hundreds of billions of dollars on might not be needed. Meta just started selling excess compute in a pivot to cloud services, and xAI unloaded the entire compute capacity of Colossus 1 to Anthropic at a discounted rate.

But if Kimi K3 is the way the industry might go, hardware demands are unlikely to fall, and as Jevon's paradox suggests, greater efficiency is only likely to increase usage, not shrink it.

Those trillions of parameters need to be stored in memory, and Bloomberg's estimates suggest Kimi K3 will require close to 1.5 TB of memory. It would need masses of high-end Nvidia GPUs to deploy it effectively, making the number of companies and organizations that could actually run Kimi K3 at scale rather small.

So even those who do look to leverage Kimi K3 to reduce operating costs will still need powerful hardware, and specifically a lot of memory. This suggests that the major competition for cutting-edge models is not going to crater costs like we initially saw with DeepSeek R1 last year, which means memory makers are going to continue making money hand over fist, due to their outsized demand and limited supply.

But Chinese memory suppliers like CXMT are on the rise, and on track to eclipse Micron's DRAM wafer capacity by the end of the year. Smaller local AI models will also continue to be further optimized for domestic hardware, reducing the stranglehold that some large tech companies have on the AI supply chain.

Competitive, efficient, but unwieldy

Kimi K3 is an industry disruptor and is already raising questions over AI costs, capabilities, and access. It's shown that you don't need proprietary models locked to a specific service to achieve frontier-model capabilities. It's also cheaper to run, but Moonshot achieved this with a sparse model that still requires massive hardware investment to operate.

Even though Kimi K3 activates only a fraction of its trillions of parameters for each query, it still needs all of them to be stored. Deploying this model at scale requires substantial memory capacity, bandwidth, and interconnects, even if its compute demands aren't as strenuous.

The open-weight nature means it has very real potential to supplant usage away from Western frontier models in the short term, but it isn't about to change the story we've been told on required infrastructure. Kimi K3 needs the same kind of hardware to run as GPT 5.6 and Fable — which is likely to be far more of a limiting factor on its adoption than any kind of government blocks.

A U.S. federal judge granted the final approval of Anthropic’s $1.5-billion settlement over the class action lawsuit authors filed against it for infringing their rights. According to Reuters, while the court ruled that training AI on books is considered fair use under copyright law, it was the fact that Anthropic kept 7 million pirated books in a central library that violated the copyrights of the authors and their publishers. This is reportedly the biggest payout in a copyright case ever, and the first one to settle among the many cases against AI firms being tried today for infringement.

While this might seem like a massive sum, the huge number of involved works and authors means that the payout amounts to a little over $200 per title. It also affirmed that AI firms’ use of existing works to train their models is fair use, which is something that many are fighting against. "We reached this settlement in 2025, after the court's landmark ruling that training AI on books is fair ​use under copyright law — which remains the law today," said Anthropic deputy general counsel Aparna Sridhar. "We are pleased that more than 91% of authors and publishers covered by the settlement have claimed their share of the payment, and we're looking forward ​to bringing this matter to a close." Because of this, some groups have opted out of the settlement and instead filed separate complaints against the AI firm.

Still, this is a landmark win for copyright holders, especially against other AI tech companies that have been scraping pirated content to train their models. This includes Nvidia, which allegedly used scripts in its NeMo Framework specifically designed for illegally downloading books, and Meta, which reportedly torrented 82TB of pirated books for AI training. The latter argued that its framework also have “non-infringing uses,” but the court said that it’s not the entire system, but specific tools within it that were the issue. As for Meta, it claimed that its use of pirated material was legal as long as it did not seed content.

Copyright infringement is one of the major issues that AI tools are facing at the moment, especially as many creators believe that these models were trained on stolen data. Anthropic’s landmark settlement is a first major win for authors and publishers — although it wasn’t exactly what some wanted, given that the settlement is small compared to the number of pirated books and that AI training is still considered fair use.

Intel and Fortinet have announced a strategic collaboration to develop the Fortinet Security Processor 6 (SP6), the next generation of the custom silicon behind Fortinet's FortiGate firewalls. Intel will contribute chip design, advanced packaging, and manufacturing to the SP6 program, making Fortinet the first cybersecurity vendor named as an Intel silicon customer.

Intel told Tom's Hardware that SP6 will be built on Intel 4, the EUV process the company has so far used only for its own products, making Fortinet both the first cybersecurity vendor named as an Intel silicon customer and the first named external customer for the node. However, the announcement specifies no production timeline, and it comes in the same Q3-Q4 2026 window in which CEO Lip-Bu Tan said the company expects commitments from multiple foundry customers.

SP6 will draw on what the companies described as Intel's expertise in disaggregated semiconductor design and advanced packaging tailored for both AI-enabled and cost-sensitive applications.

That points to a chiplet-based part, which would be a departure from the current SP5, a monolithic 7nm Arm-based SoC that Fortinet launched in 2023 for its entry-level and mid-range FortiGate appliances. The deal will help Fortinet "accelerate and strengthen our ASIC strategy," said Ken Xie, founder, chairman, and CEO of Fortinet, in the announcement.

Intel 4 was the company's first process node to use EUV lithography and entered high-volume manufacturing at Fab 34 in Ireland in September 2023, where it produces the compute tile for Meteor Lake-based Core Ultra chips. The node didn't appear among the processes Intel listed for external foundry customers in its fiscal year 2024 annual report, which named 18A, Intel 3, Intel 7, Intel 16, and a 12nm process co-developed with UMC. Intel said the SP6 work reflects plans it laid out for Intel 4 several years ago, including support for custom networking ASIC workloads.

Microsoft agreed in early 2024 to build an unnamed custom chip on Intel's 1.8nm-class 18A node, a deal that reportedly covers a next-generation Maia AI processor. That announcement followed a similar pattern, with no product details, node variant, or timeline at signing and specifics emerging over the following 18 months.

Intel CFO David Zinsner said in March that the company was fielding inbound interest in 18A-P from prospective foundry customers as yields improved, and Tan told CNBC in May that foundry commitments were expected in the second half of 2026.

Fortinet does bring real volume, though, if not marquee volume. IDC ranked Fortinet first in firewall appliances shipped as of early 2023, with a 48% unit share. The company ships its own ASICs across its entry-level and high-end FortiGate ranges, and SP6 extends a silicon program now in its sixth generation.

Neither company committed to work beyond SP6, though the release described the agreement as a starting point, with further collaboration on chip technology and manufacturing under discussion.

Chinese AI developer Z.ai (formerly Zhipu) has finished building a 1GW data center stocked exclusively with domestically made chips and has switched part of it on, Bloomberg reported Monday, citing a person familiar with the matter. The facility will train the company's GLM model family, and the source told Bloomberg that Z.ai has now built or operates several computing clusters holding more than 10,000 chips apiece. A gigawatt is enough electricity to run roughly 750,000 homes, which puts the site among the largest ever stood up by a Chinese AI lab.

The source didn't name the chip supplier, but Z.ai's recent training history points to Huawei. The company released GLM-5.2 in June, an open-weight model purportedly trained entirely on Huawei Ascend accelerators with no Nvidia hardware involved, and it topped the open-weight leaderboards within a week. Z.ai, formerly known as Zhipu, has also been on the U.S. Commerce Department's entity list since January 2025, which cuts off legal access to Nvidia silicon and leaves domestic parts as its only supply line.

Raw power draw flatters the comparison with U.S. sites of similar size, however. Chinese accelerators such as Huawei's Ascend line trail Nvidia's current Blackwell parts on performance per watt, so a gigawatt of domestic silicon delivers less usable training compute than a gigawatt consumed by Nvidia systems.

Beijing is drafting a plan to spend roughly 2 trillion yuan ($295 billion) over five years on a nationwide grid of AI data centers, with at least 80% of the underlying technology sourced from Chinese suppliers. Filling those facilities is a big problem, though, as SMIC's most advanced stable node — the roughly 7nm-class N+2 process — is running above 93% utilization.

In addition, scarce domestic HBM constrains how many Ascend-class accelerators Huawei can assemble, and Huawei shipped around 812,000 AI chips last year. Ultimately, China can put up a 1GW shell much faster than the chips needed to draw 1GW can be produced.

Z.ai's rival Moonshot suspended new subscriptions on Sunday, saying in a social media post that it wanted to prioritize compute for existing members after the launch of its Kimi K3 model. Z.ai itself is on track for $1 billion in annual recurring revenue after hitting its 2026 sales target in July, people familiar with the matter told Bloomberg earlier, and the company recently raised billions of dollars through a Hong Kong IPO and a follow-on share sale.

The source didn't disclose the data center's location, its cost, or its construction timeline.

TSMC intends to raise base quotes on advanced chip production services by up to 10%, according to Nikkei, which cites people with knowledge of the matter. The price hike reflects increased demand for sophisticated processors by the AI sector, raising costs of tools and materials, as well as amplified investments in new production capacities.

For advanced process technologies — which TSMC considers 7nm-class and below — TSMC plans to raise baseline prices by 5% to 10%, depending on the particular production node and customer, the report claims. Furthermore, customers that need additional HPC chip capacity beyond their original volume requirements will reportedly have to pay another 10% to 15% premium on top of the standard increase, which means that some services will get a price hike of around 25%, if the report is accurate.

TSMC also intends to increase prices for mature manufacturing technologies, including its 12nm, 16nm, and 28nm-class nodes as well as other legacy fabrication technologies, the report claims. Increases could reach 10%, although certain nodes will reportedly see smaller adjustments, according to Nikkei.

Advanced technologies generated around 77% of the foundry's revenue in Q2 2026, whereas mature nodes accounted for 23%, which essentially means that TSMC is hiking prices on all of its services.

The company reportedly began discussing the new pricing with customers around June and completed negotiations in July. Rather than introducing higher rates immediately, TSMC opted to implement them from the beginning of 2027 to give clients like Apple, AMD, Nvidia, and MediaTek additional time to accommodate the changes and adjust their prices accordingly.

Since TSMC produces the lion's share of advanced processors for AI, HPC, networking, and smartphone applications, its price hikes will inevitably create a ripple effect in the industry and will make almost all electronics more expensive.

TSMC is not alone in raising prices these days. Vanguard International Semiconductor has also raised prices, while UMC began implementing increases in July. Also, memory makers have increased prices significantly, making TSMC management jealous. Intel also recently increased prices of its client and data center CPUs, citing market demand.

"I am really jealous about memory companies' 86% gross margin," said C.C. Wei, chief executive of TSMC, during the company's earnings call with financial analysts and investors. "86% [margin at memory makers] – 68% [margin at TSMC], I will be happy about that."

TSMC rarely comments on its prices to a large degree because they vary based on volumes and relationship with a particular client. Nonetheless, the head of the company stressed that the company has no intentions to increase prices suddenly or dramatically.

"So we do not suddenly increase our price by which I like to have 4x or 5x," Wei said. "You cannot survive for that kind of... for your customer to survive for that kind of price increase. So we earn our value, and we make sure that our profit, our gross margin, is enough for our long-term sustaining expansion, that is to the benefit of my customers and TSMC also, that is our philosophy."

Intel just announced another round of layoffs more than a year after CEO Lip-Bu Tan warned of “tough decisions” required to get the company back on track. This decision has resulted in the reduction of its headcount by more than 35,000 since 2024, when ex-CEO Pat Gelsinger revealed that its data center and foundry divisions have lost $1.6 billion. The latest announcement came months after the last job cuts and is still happening despite the company posting a strong first quarter this year. According to Oregon Live, its share price has more than tripled from a low of $23 per share to more than $96 today, with the data center group reporting sales of $5.1 billion for the first quarter. However, this good performance seems not to have affected the firm’s plan to streamline its operations.

“As part of our broader strategy to become a more focused and efficient company, (the data center group) is aligning its organization to ensure it has the right roles and skills in place to position the business for long-term success,” Intel told the publication in a statement. It also added, “We are committed to treating all impacted employees with respect and providing resources to support them through this transition.” Unfortunately, the company did not say how many positions will be cut and when it is happening — it only assured that the reduction-in-force won’t affect its business commitments and plans to launch new products.

While the AI infrastructure build-out was initially powered by GPUs and memory chips, advancements in agentic AI have greatly increased the demand for CPUs, putting Intel in a good position with its Xeon chips. Aside from that, the company is reporting customer interest, possibly including Apple, for its 18A and 14A nodes. It’s also planning to launch an AI GPU in the latter half of 2026 that will compete against Nvidia’s RTX Pro 5000 GPU.

This is a slow reversal of its misfortunes in the first half of the decade, but it seems that it’s not happening fast enough for its data center group employees who will be affected by the job cuts. It’s quite surprising that Intel plans to gut its Data Center group, which is one of its strongest performers and is in a good position to take advantage of the ongoing AI data center boom. But despite the high demand for AI data centers, it seems that the company still needs to cut its personnel count to streamline its operation.

An analysis of Huawei's Kirin 9030 system-on-chip (SoC) for smartphones conducted by SemiAnalysis revealed that SMIC's third-generation 7nm-class fabrication technology (N+3) has smaller metal pitch than Intel's 18A fabrication technology and that China's leading foundry has managed to achieve transistor density on par with manufacturing process that rely on EUV lithography. But does this make SMIC's N+3 node as competitive as Intel's 18A or TSMC's N2 and N3? Not really.

SemiAnalysis' teardown indicates that SMIC's N+3 fabrication process supports a minimum metal pitch of 32.5nm, which is nominally tighter than the approximately 36nm pitch used for many high-performance cells in Intel's Panther Lake CPU, even though 18A can support approximately 32nm metal pitches. The video from SemiAnalysis and High Yield does not reveal other important characteristics of SMIC's N+3, such as contacted gate pitch (CGP), standard cell height (tracks or nm), or fin pitch, so we cannot make direct comparison of this node to Intel's or TSMC's technologies. What it does reveal is estimated transistor density of around 113.4 million transistors per square millimeter (Mtr/mm²), which is even higher than transistor density of TSMC's N6, 107.7 Mtr/mm².

TSMC's N6 uses multiple EUV layers, so achieving higher transistor density without using EUV lithography is an indisputable technological achievement of SMIC. The foundry achieves this density by using DUV multi-patterning, including self-aligned quadruple patterning on the tightest layers, and extensive design-technology co-optimization (DTCO). In addition, SemiAnalysis believes that SMIC used techniques like reduced fin counts, placing contacts directly over active gates, and tightening cell isolation. Such methods allow for increased transistor density, but at the cost of increased process complexity, cost, yield risks, and design constraints.

Meanwhile, transistor density does not equal overall process competitiveness. Despite its compact layout, the Kirin 9030 reportedly delivers performance comparable to flagship application processors from roughly three years ago and has a substantial energy-efficiency disadvantage compared with modern Apple, Qualcomm, MediaTek, and Samsung designs. Huawei's highest performing CPU core is characterized as roughly Cortex-X2-class in IPC, while Apple's much smaller efficiency cores reportedly outperform it in integer workloads and consume considerably less power.

Given the fact that Kirin 9030 is neither a performance nor efficiency champion, the provocative comparison with Intel 18A is not exactly justified. Although SMIC N+3 has 32.5nm minimum metal pitch that is nominally tighter than the approximately 36nm pitch used in Panther Lake, 18A offers both higher transistor density and considerably higher performance efficiency. In addition, 18A uses gate-all-around transistors and backside power delivery, which make it particularly suitable both for mobile SoCs and for data center applications.

SemiAnalysis concluded that while export restrictions have slowed China's technological progress, progress is still being made. SMIC could potentially continue increasing density by tightening upper and lower metal layers, shorter standard cells, smaller gate pitches, and eventually backside power delivery. If the company continues scaling, N+4 could approach TSMC N5-class density, while N+5 with backside power might reach Intel 18A-class density, according to SemiAnalysis. Still, transistor density alone does not necessarily bring substantial improvements of performance or power efficiency.

Google is developing a server chip, informally dubbed "Frozen v2," that would etch part of its Gemini model's architecture directly into the silicon, according to a report published Monday by The Information, citing two people with direct knowledge of the matter. Engineers on the project have projected that the chip could serve six to ten times more tokens per unit of power than the newest generation of Google's TPUs, with deployment targeted for as soon as 2028. The two sources said the project is partly a response to an AI compute shortage severe enough that Google Cloud has turned down deals with outside customers.

A TPU, like a GPU, runs whatever model is loaded onto it, which means the hardware makes time-consuming runtime decisions as it interacts with each one. Frozen v2 would have some of those decisions for Gemini fixed in the transistors, reducing the number of steps the chip takes and the amount of data it shuttles around per query. That could cut response latency enough to enable new applications, one of the sources said.

The original Frozen design, spearheaded by Google DeepMind chief scientist Jeff Dean, went further and would've baked Gemini's weights themselves into the chip. Google set that proposal aside because silicon tied to a single model version would have too short a life cycle, according to the report. Frozen v2 freezes the architecture instead and leaves the weights updatable, so the chip stays useful across Gemini releases, but only for as long as Google builds them on the same underlying architecture. How much of the model to lock in is reportedly still undecided.

Google doesn't plan to produce Frozen v2 at TPU volumes and views this generation partly as a trial run for more specialized silicon as model designs settle. The chip would sit alongside, rather than replace, a TPU line that split into separate training and inference variants with the eighth generation announced at Cloud Next in April. Its 2028 target also comes in the same year Google has reportedly booked Intel to package more than 3 million TPUs.

Model-hardwired inference silicon already exists in demonstrations. Taalas, a Toronto startup that has raised more than $200 million from investors including Quiet Capital and Fidelity, launched its HC1 chip in February with Llama 3.1 8B permanently wired into an 815mm-squared die built on TSMC's N6 process. The company claims 17,000 tokens per second per user with no HBM on the package. Nvidia, meanwhile, struck a $20 billion deal in December to license technology from inference chip designer Groq.

Google hasn't yet confirmed the project, with a spokesperson telling The Information that not every project moves into production and that "this rigorous exploration is central to our full stack approach."

We live in an age where artificial intelligence increasingly dominates the internet, making it harder to distinguish real content from fake news. Any picture or video you see today could be AI-generated and the traditional markers that would give it away are fading rapidly. There are tools that help identify if something has been made with AI, and Nvidia — arguably the main beneficiary of the AI race — has just released its own, called the "Synthetic Video Detector" (SVD).

SVD is an Nvidia Inference Microservice (NIM) part of the company's AI for Media Private Access Program, so it's not publicly available to consumers, but a demo exists. Anyhow, SVD's job is simple: detect whether a video is real or if was generated using AI. It can analyze videos at scale, breaking them down frame-by-frame to spot anomalies. It's based on cutting-edge research that won awards at computer vision conference ICCV.

Instead of looking at the video file as a whole, SVD splits it into cropped frames, each carrying a 504x504 resolution. These frames are then passed through two powerful Vision Transformers made by Meta: DINOv2 and DINOv3. A job of a vision transformer is to learn to form patterns without needing human-labeled data. They're commonly used for image classification, image retrieval, object detection, and depth estimation.

As such, once the frames go through these transformers, their distinct spatial features are quickly assessed, and each one is assigned a score between 0 and 1 — 0 representing a fully real image and 1 representing a completely fake image. The scores are tallied at the end to form an average, which tells the user whether the video is real or not based on a percentage score out of 100.

This way, news agencies, broadcasters, and media outlets can authenticate footage they receive much quicker and with better certainty. SVD is even designed to work with the reality of social media compression since videos uploaded online will have their imperfections masked. But the transformers can still detect patterns that the human eye cannot, seeing past surface-level anomalies to instead focus on intrinsic artifacts.

As visible in AI GVD bench above, uncompressed video still delivers the best results with SVD showing an insane 92% accuracy rate. At 15% compression, the model drops down to 87% accuracy, while a 50% compression rate still achieves a very impressive 82% accuracy in detecting AI-generated content. Since this is a microservice, it has exceptional latency as well, processing 1080p video in just 22ms on Nvidia RTX GPUs and 30ms on Nvidia's workstation models.

That being said, SVD requires the NVENC encoder, so datacenter cards like the B100 cannot run it natively. Nvidia said it's already working with Wowza to bring real-time synthetic video detection into livestreaming workflows. A demo version is available to try right now at build.nvidia.com but beware that it takes a long time to process since it happens in the cloud, the max file size limit is only 100MB, and it often just times out.

Much has been written regarding the questionable economics of the AI space, but most of the discussion revolves around high-level concepts like market shares, datacenter investments, and power expenditure. The general expectation of technology is that it gets cheaper as it improves, but the AI space has a rather peculiar problem: Usage costs are actually soaring even as models get better, as detailed in a write-up at VentureBeat.

With all the advancements in models over the last two years, having a bot that answers questions of simple-to-moderate difficulty is now old news, as they all do that with reasonable accuracy. Agentic workloads are where the real potential is at, letting bots loose on multi-stage, repeatable tasks that would previously take a human days, if not weeks, to perform. Grant a bot access to your billing system, some Excel spreadsheets, and your CRM, and ask it something like "who are my profitable customers by category and what are their trends" becomes child's play.

While it's trivial for you to ask that question and get a fairly accurate answer back in a couple minutes, behind the scenes there is a lot of processing going on, and far more than you'd expect. AI computing time is measured in tokens — a short question and answer might take somewhere between 200 to 2,000 tokens, and one that requires the models to do some internet research might be around 1,000 to 4,000. An agentic task, though, can easily spend millions of tokens on a seemingly innocuous request. How? Token amplification, a recently coined term.

In a simplified manner, because a model has no memory or cognition, every time you ask it another question in a conversation, it will re-load and process the entire exchange — everything you wrote, everything the bot replied with, and every file you uploaded. That means that additional questions in a long conversation progressively get more costly. Each question in a conversation might only need 500 tokens by itself, but reprocessing all the previous information adds up, so the second one might spend 600, and so on and so forth. The conversation as a whole uses up the cumulative number of all individual interactions, and as an added penalty: Response speed also tends to get slower as chats drag on.

The aforementioned task of generating a report will have to be run in stages, say three for looking up Excel sheets, four for the CRM, perhaps a half-dozen web searches for contextual information about products, and a good dozen intermediary processing and calculation steps. Each step tacks on potentially several thousands of tokens, and by the end of it, you may be looking at millions of tokens cumulatively spent for all the steps combined.

Costs, then, can spiral out of control very quickly — and that's assuming a best-case scenario where all of the data is easy to interpret, the model won't need any retries, and you're using a moderately powerful model rather than something like Claude Opus.

In financial terms, this means that one innocent question might spend 280,000 tokens and cost $1, with ballpark estimates at current prices. That may not sound like much, but it was just one task. If you have this scheduled to run every 15 minutes for a dashboard, then suddenly this costs $96 per day, or $2,880 in a month. And that task example was fairly simple; a tricky multi-stage report might need millions of tokens, turning that $1 into, say, $10, or a grand total of $28,800. For the one report, for a limited number of users, in one department of a corporation.

That's the key reason why Anthropic, Microsoft, OpenAI, and almost every other player moved their major offerings to usage-based billing back in April, and severely limited token spends in fixed-rate plans. This generated many sticker shock events, as developers around the globe found out how much their vibe coding actually cost and were left in a daze.

In a large company using many AI agents, such costs can become prohibitive and oftentimes costlier than standard-issue human employees. Firms like Uber, Microsoft, Amazon, and Walmart, among many others, have responded by curbing AI spend. Token expenditure is suddenly an issue for both financial and engineering departments, as cost control becomes paramount. As VentureBeat succinctly puts it, for agent-heavy companies, "a prompt redesign is a margin event," and more illustratively, "a poorly bound agent loop is an outage with a credit card attached."

Although most AI outfits still offer fixed monthly pricing plans, they often come with harsh token limits. As is often the case with most fixed-pricing services, however, the most capable bot-wranglers will also inevitably be the ones burning through the most tokens, not just due to their intensive usage per se, but because the type of tasks they will perform will be precisely the ones hardest hit by token amplification.

This throws a wrench in the usual financial model of the casual users easily offsetting the cost of the small percentage of professionals — the pros' usage can take a serious bite off monthly profits, or chew through them entirely.

The AI companies aren't sitting still, and getting the per-token cost down is likely to be the primary task for most of their engineering teams, at this point. Prompt caching, model routing, batch processing, semantic caching, and context window management are among many technical measures that can massively cut down on token spend, each of those netting a two-digit percentage drop.

And yet, costs keep soaring for the simple reason that the smarter the models get, and the more adept people become at using them, the more complex and long-running the agentic tasks will become, adding multiple orders of magnitude to token counts.

For the time being, it looks like a losing race, despite recent advancements like Deepseek V4 Pro and V4 Flash deeply undercutting comparable Western models at up to a reported 17x for the former and 25x for the latter. Out of the major Western players, only Anthropic has predicted its first profitable quarter, though much like anyone else, it's still in the hole for many a billion for cumulative spend.

Taiwanese prosecutors indicted a former TSMC deputy manager on Monday for allegedly copying 21 confidential documents, some covering technologies Taiwan designates as national core technologies, intending to use them in China. Prosecutors are seeking a sentence of up to seven years for the man, surnamed Chen, who has been detained since late May, and describe the indictment as the first under Taiwan's National Security Act to involve an alleged attempt to leak the island's most sensitive chip technologies to China.

However, that alleged transfer never happened. TSMC's internal monitoring systems flagged irregularities, with the company reportedly recovering every improperly copied document after its own investigation, according to the prosecutors' office. A High Prosecutors' Office spokesperson told Nikkei Asia that despite the failed attempt, the case "remains highly significant" as the first of its kind.

Prosecutors found Chen while investigating a separate Chinese espionage case centered on a Hong Kong national named Ding Xiaohu, Taiwan's Central News Agency reported. Investigators allege Ding was acting on instructions from the Nanning station of the Chinese Communist Party's Central Military Commission Political Work Department, entering Taiwan repeatedly to recruit agents and collect intelligence on the island's critical technologies, and that he recruited Chen through an intermediary surnamed Huang. Ding died earlier this year and was never charged.

Chen and Ding allegedly worked together between May 2023 and February 2024 to set up CSMAC, a semiconductor materials analysis company in China. Chen drafted several proposals for the venture, including one called the "Blue Ocean Plan" that aimed to recruit workers from Taiwan's chip industry to join CSMAC, and took the copied TSMC documents home to study for use in China, prosecutors said.

The state-directed recruitment allegation separates this case from the trade secrets disputes TSMC has fought over the past year, which involved employees allegedly taking confidential data to other chip industry employers.

Taiwan's National Security Act, in force since 2022, makes it a crime to copy, use, or disclose trade secrets tied to designated national core technologies, a list that includes chipmaking processes more advanced than 14nm. Prosecutors brought the first charges under the law last August, when three current and former TSMC employees were arrested over an alleged attempt to leak 2nm process data. That case grew to ensnare Tokyo Electron's Taiwanese unit, which prosecutors charged with failing to prevent the theft.

TSMC is also pursuing a civil case against Wei-Jen Lo, the former R&D executive who joined Intel last year, with Taiwanese authorities running a parallel national security inquiry into his departure.

The U.S. government may be back on track to ban leading Chinese AI models, following the release of Kimi K3 — a powerful AI model developed by the Chinese startup Moonshot AI —last week. According to an Axios report released July 20, Trump's administration is reigniting its push for a ban, citing cybersecurity concerns. Chinese models such as DeepSeek and Kimi K3 are open-weight, meaning their trained model weights are published for public download. This gives enterprises the option to keep their data in-house by self-hosting the models on private infrastructure, while slashing inference costs — characteristics that have led to rising adoption by U.S. companies.

Citing several sources close to the administration, the Axios report says that the government had earlier made a series of attempts to curb growth and expansion of Chinese models in the U.S. over cybersecurity concerns, a move critics say will stifle competition and innovation and encourage monopolies. The U.S. Department of Commerce last year considered adding multiple Chinese AI labs, including DeepSeek, to its "Entity List," a trade blacklist maintained by the department’s Bureau of Industry and Security (BIS) that limits foreign companies, research institutions, governments, and individuals from purchasing sensitive American hardware, software, or technology.

U.S. officials also considered a joint National Security Agency/Office of the National Cyber Director advisory to discourage the use of Chinese AI models, and drafted an executive order holding U.S. companies liable for security breaches involving hosted Chinese models. While these measures were initially paused due to internal pushback regarding market impacts, they have been revived following the release of new Chinese open-weight models.

According to the report, critics of the potential ban — such as former White House adviser Sriram Krishnan and David Sacks, an outside White House AI adviser — say the move would negatively impact innovation, while handing a monopoly of the market to leading U.S. AI labs, OpenAI and Anthropic, which the report implies may have a hand in the push for a ban. “We are at a critical inflection point in AI policy. The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition,” wrote Sacks in an X post on Sunday.

Chinese AI models are being increasingly used by numerous American companies due to their relatively low cost and perceived matching capabilities with domestic alternatives. The open-weight nature of Chinese models such as DeepSeek V4 and Kimi K3 — which allows companies to download the models locally and host them on private servers — is driving adoption by giving enterprises data privacy and slashing API costs compared to closed Western alternatives. Conversely, self-hosting shifts the cost of GPUs, electricity, maintenance, networking, and model operations to the company, making it generally most economical for organizations with substantial and sustained AI usage.

Chinese open-weight models price their APIs well below comparable U.S. systems, with DeepSeek-V4-Pro charging $0.87 per million output tokens, compared with $50 for Anthropic’s frontier Claude Fable 5 model. This aggressive undercutting has caused a massive surge in developer adoption. CEO Brian Armstrong noted that Coinbase runs models like GLM-5.2 and Kimi in production, cutting their overall AI spending nearly in half even as their actual token consumption spiked.

Despite the rising adoption, the U.S. government cites cybersecurity concerns as a reason for a ban. Now, the question of whether a ban on Chinese AI models can be practically enforced arises, as blocking open-weight technology presents a technical and regulatory hurdle. For an individual or a small company just wanting to use DeepSeek via the website or the app despite a U.S. block, a VPN works fine. However, limited app availability and payment restrictions remain effective restrictions.

For enterprises that self-host rather than use the hosted app or API, enforcement gets harder for several reasons. Unlike closed-source APIs that require data to leave a company's network by sending it to a third-party provider's servers, open-weight models exist as downloadable files mirrored across public repositories like Hugging Face and independent torrents, making them hard to fully recall once released. Once an American enterprise downloads the weights, it can run the model entirely offline inside a private, air-gapped data center, which limits U.S. regulators' ability to monitor which model is running locally.

Modifications further complicate enforcement. Companies routinely fine-tune, quantize, or distill these models, blending the Chinese base with domestic corporate data until provenance blurs and it becomes difficult to define where the foreign model ends and a new domestic one begins. Even under strict download bans, firms could host the models through subsidiaries, although that vector runs into know-your-customer rules at cloud providers and the extraterritorial reach of U.S. export controls.

However, the U.S. government may not need an outright ban. According to the Axios report, the strategy appears to be getting U.S. firms themselves to drop the models. Axios-cited government sources say that procurement rules, Entity List threats, and public pressure campaigns aimed at the companies using Chinese models may do the trick. The sources also say the government will “push to highlight potential backdoors and lack of security with Chinese models, and the governance issue that brings.”

Any ban or restrictions would be yet another event in ongoing broad trade tensions between the U.S. and China that have since extended into the AI industry. Washington had earlier placed export restrictions on critical computing hardware and equipment to China. It later eased restrictions, but Beijing now appears to be focused on developing domestic technologies, while urging Chinese companies to utilize them. The Trump administration has also made known its intention for the U.S. to dominate the AI race.

The Hague Centre for Strategic Studies has rated the Dutch semiconductor industry as being at very high risk of Chinese foreign interference in a recent report, placing it above the maritime and aerospace sectors, and citing years of espionage against ASML and NXP. Written by Benedetta Girardi and Hans Horan for the China Knowledge Network, a body funded by the Dutch Ministry of Foreign Affairs, the study calls for stricter vetting of personnel at sensitive chip sites and a public-private fund to swap out high-risk foreign hardware, including server motherboards and baseboard management controllers. Both authors expanded on the findings in interviews with Nikkei Asia last week.

The report sorts Beijing's methods into five categories: economic statecraft, digital and information operations, physical interference, legal pressure, and talent-based technological espionage. ASML said in 2023 that a former China-based employee stole data about its lithography tools from an internal product lifecycle management system, one of several insider theft cases at the company stretching back to 2015, and the Dutch military intelligence service MIVD named all three sectors as Chinese espionage targets in its 2024 annual report.

China restricted rare earth exports through 2025, and ASML's complex optical systems depend on those materials. The EU holds an anti-coercion instrument designed to counter economic blackmail by non-member states, but leaders treat it as a last resort. The tool was "created with China in mind," Benedetta Girardi, program coordinator of the HCSS Europe in the Indo-Pacific program, told Nikkei Asia.

Dutch anti-discrimination law prohibits screening policies that target only Chinese nationals, and the standard VOG background certificate doesn't account for an applicant's conduct or connections outside the EU. The report proposes a tiered system instead, whereby companies identify their "he Hague Centre for Strategic Studies has rated the Dutch semiconductor industry as being at very high risk of Chinese foreign interferencejewels" and apply deeper screening, weighing factors such as susceptibility to blackmail or bribery, to staff with direct access, while junior hires and contractors get lighter checks.

A proposed resilience fund, built with the EU and NATO, would pay to replace equipment such as server motherboards, BMCs, and large ship-to-shore cranes with European-made equivalents. The mechanism would sit alongside Project Beethoven, the €2.5 billion Dutch government effort to expand domestic semiconductor production capacity, and would include surge funding that unlocks during crises such as sanctions escalation or major cyber campaigns.

ASML denied a U.S. government concern in June that one of its EUV machines had reached China in breach of export controls, telling Tom's Hardware the claims were baseless and harmful to its reputation.

The Nexperia dispute, in which the Dutch government seized the chipmaker from its Chinese parent, and Beijing responded by blocking exports from its Guangdong site, played out while the report was being written. In a crisis such as a confrontation over Taiwan, the researchers assess that Chinese leverage over strategic industries would likely constrain Dutch policy choices and limit participation in allied responses.

The demand for AI compute is already insatiable, and it seems only poised to grow in the wake of the introduction of frontier-class open models like Kimi K3 that anybody can potentially fine-tune and serve. Against this backdrop, Microsoft and AMD are teaming up to get Redmond more AI FLOPS for both internal and external use. The two companies announced this morning that Microsoft will commit to adding AMD's Helios rack-scale AI accelerator in volume to run frontier-model workloads in its own data centers, as well as for Azure AI infrastructure customers and services.

The partnership makes next-gen AMD AI compute available to Azure customers like AI labs for AI training and inference serving workloads, and it’ll also underpin managed compute for enterprise customers looking to deploy AI workloads through Microsoft Foundry.

The two companies didn't indicate the exact size of Microsoft's Helios deployment in either watts or dollars, but the commitment would seem to be another major win for AMD as it seeks to grab data center GPU share from Nvidia. AMD has struck massive partnerships with OpenAI and Meta in the past year with gigawatts of compute installations and hundreds of billions of dollars potentially hanging in the balance.

For a quick refresher, the Helios rack-scale accelerator will take the fight to Nvidia’s Vera Rubin NVL72 system when it arrives later this year. Helios joins together 72 next-generation Instinct MI455X GPUs with an aggregate of 31.1TB of HBM4 memory capacity across the system. Those GPUs offer as much as 1.4 exaFLOPS of FP8 compute and 2.9 exaFLOPS of FP4 for AI models using those OCP AI data types.

AMD is targeting 260 TB/s of scale-up bandwidth within the rack, on par with Nvidia’s Vera Rubin NVL72 rack-scale system, and 43 TB/s of scale-out bandwidth using UALink over Ethernet, or about twice that of Vera Rubin, although the performance of UALink over Ethernet in practice remains to be seen.

Microsoft and AMD also announced that Azure will add two new VM series built on AMD's upcoming sixth-gen Epyc Venice CPUs: the HDv2 series for "agentic AI and data pipelines," and the HXv2 for semiconductor design workflows. Microsoft will also leverage its existing deployment of AMD Pensando DPUs to integrate that hardware into its Azure Boost offerings to accelerate networking and storage processing operations.

Tom’s Hardware will be on the ground at AMD’s Advancing AI event this week, where we expect to learn more about AMD’s AI ambitions for the second half of this year and beyond. Stay tuned for our coverage from that event.

A report by The Conversation has claimed that the government can seize private land to make way for new transmission lines needed to meet the surging electricity demand of data centers. According to the July 16 report by Aaron Walayat, Assistant Professor of Law at the University of Dayton, power companies can use eminent domain — the legal authority that grants the government the power to take private property and convert it to public use in exchange for compensation — to implement the takeovers.

The AI boom has led to a surge in data centers in the U.S., with thousands already operational and several more planned or under construction. While the massive superclusters are necessary for the technological revolution AI has brought about, there's growing opposition to their construction over several concerns. Among them are land use, noise pollution, water usage, and the impact of immense electricity consumption, issues that have reportedly made 70% of Americans opposed to building data centers nearby.

In many instances, data centers draw the required electricity from the grid. As the industry enters the gigawatt era, utility companies are under pressure to increase supply to meet surging demand. This requires building new power infrastructure, such as transmission lines, which often have to cross private land. When this happens, the power companies try to buy the land. Should the owner refuse, the government can force a sale through the eminent domain law.

The law grants the government the power to take private land, regardless of the owner's consent, provided that the land is for public use and the owner receives just compensation. The government can also delegate the power to “private entities or common carriers,” such as utility companies. It is this legal authority that power companies can enact to implement the takeovers.

According to the report, the law does not automatically grant infallible authority. The company must prove that the infrastructure will be for public use. Several states also reserve the right to interpret eminent domain laws according to their own constitutions.

The report highlights another layer of legalities amid ongoing data center tensions. Opponents have successfully blocked 75 planned data center projects in the first quarter of 2026, including the 2,100-acre Digital Gateway project, eventually canceled over a newspaper-notice technicality. On the other hand, several other projects have gone ahead, often with community support. Meta recently announced plans to expand its Hyperion AI supercluster from 2 GW to 5 GW.

The head of the SK Group, which owns SK hynix, said in a press briefing that memory semiconductor prices are “abnormally high,” and that the industry must take steps to increase supply and help lower prices. The Chosun Daily writes that SK Group Chairman Chey Tae-won delivered this statement during a press briefing at an industry forum, where he also noted that the group is looking at building a memory chip plant in the U.S. to increase production.

“Memory prices are currently at an abnormally high level. While AI companies can absorb increased costs through investments, PC and smartphone manufacturers have no choice but to pass rising semiconductor costs onto product prices,” Chey told the media. “Since most customers for these products are individuals, there are limits to how much prices can be raised. To prevent 'chipflation'—where rising chip prices drive up finished product costs—supply must be expanded.”

Aside from expanding supply and easing memory prices for consumer products, extremely high memory prices brought about by limited supply could negatively impact the big three companies. That’s because the high margins could lead to the entry of new players or give smaller manufacturers the opportunity to grow their presence and challenge the established firms. We’re already seeing this, as some Chinese brands have ditched them for domestically produced CXMT and YMTC chips. While this could be driven by Beijing, even major multinational brands like Corsair and Lenovo have started sourcing from these companies just to get chips. Even Apple is asking permission from Washington to buy memory chips from CXMT. Aside from that, he also cited Elon Musk’s interest in building his own fab as a potential threat to existing memory semiconductor companies.

This isn’t a problem at the moment as there is more than enough demand to go around for incumbent memory and storage chip manufacturers and new entrants. In fact, memory companies can use this as a defense against price-fixing lawsuits, with the massive AI demand and the higher margins it offers standing as a plausible reason why memory chip companies focused on HBM production and reduced DRAM output. But before the AI boom, the memory industry was actually suffering from one of the worst memory downturns to hit the industry for over a decade. So, if and when the memory shortage is over and the memory industry resumes its boom-and-bust cycle, more manufacturers jockeying to find customers in a tight market would make it that much harder for these companies to survive and thrive.

Researchers from Northwestern University in Illinois have built a drone that rotates so fast that it is cloaked by motion blur. The new Phantom Twist drone spins at 25 times per second, which, thanks to a quirk of human vision, makes it appear like a faint smudge in the air. The video below shows it off to great effect, but I hope they can do something about the whiny noise.

The first demonstration of the Phantom Twist took place as part of a Robotics: Science and Systems 2026 presentation on Thursday. “Most efforts to hide drones focus on making them look like their surroundings,” said Northwestern’s Michael Rubenstein, who led the work. “Instead, we asked whether we could design the drone itself around the way humans perceive motion. This idea of low visibility through persistent motion is something few people have explored.”

The stationary photo of the Phantom Twist reveals a rather unique drone design, far removed from the now traditional quadcopter. It has just one motor and one propeller, and while the propeller spins in one direction, and the rest of the drone spins in the opposite direction, explains the Northwestern University blog.

AI was used to refine the design

The researchers tasked a computational model to generate roughly 20,000 drone configurations capable of stable flight. Subsequently, they applied AI and other optimization algorithms to repeatedly rearrange the drones’ major components, including a motor, propeller, circuit board, counterweight and batteries.

After narrowing down the candidates to around 500, the engineers superimposed simulated drones in flight over a hundred real-world backgrounds. A perception model then whittled down the designs based on their visibility scores. According to Rubenstein this entire process was automated, and only “when we were confident that a drone met all our criteria, we built it.” The demonstrated Phantom Twist drone is apparently “10 times less visually perceptible than a conventional quadcopter” according to the researcher’s visibility metrics.

Yes, the team knows that the Phantom Twist drone is rather noisy, and the visual impact of its wires and support rods could be reduced. Thus, work is planned to utilize more transparent materials and / or quieter propulsion systems in upcoming revisions.

Samsung has cut jobs across its U.S. consumer electronics business as it prepares to move Samsung Electronics America's headquarters from New Jersey to Texas by the end of the year, with a WARN notice listing 739 affected roles at its Englewood Cliffs offices and roughly 100 workers let go at its Plano, Texas, site, according to a Reuters report.

Samsung told the outlet on Sunday that a majority of the affected New Jersey employees received relocation offers, while others were let go. SEA runs U.S. sales and marketing for Samsung's smartphones, TVs, displays, and home appliances, and doesn't include the company's semiconductor operations. The Plano figure, which covers workers in the mobile division among others, came from one laid-off employee who spoke to Reuters anonymously.

SEA employs about 1,200 people in New Jersey, according to U.S. Rep. Josh Gottheimer's office, which said the Englewood Cliffs campus opened last September, replacing the Ridgefield Park headquarters Samsung had occupied since 1992. The WARN notice covers over 60% of that workforce, roughly 10 months after the site opened. Internal documents reviewed by Reuters show employees were notified of an “enterprise-wide reduction in force” on June 30.

Samsung said in its statement that the relocation is intended to foster "stronger collaboration and optimize the organization," and that there's currently no broad global restructuring underway in its consumer products business. A current SEA employee told Reuters that workers are nonetheless worried about further cuts and a possible consolidation of the appliance, home entertainment, and mobile divisions as the company concentrates resources on chips.

Samsung's chip division booked a preliminary $58.5 billion in second-quarter operating profit, a 19-fold year-on-year jump that passed Nvidia's most recent quarter and put Samsung on track to out-earn its entire 40-year semiconductor history in 2026. Its mobile division is expected to post its first-ever quarterly loss over the same period, squeezed by competition from Apple and by the memory price increases its own chip unit is charging. Conventional DRAM contract prices rose 58% to 63% in Q2, and smartphone makers are raising handset prices to offset persistently high memory costs.

The consumer arm's relocation puts it alongside Samsung's existing chip footprint in Texas, which includes two fabs in Austin, the delayed Taylor fab, and a mobile hub in Plano. Samsung employed 11,770 people in the U.S. at the end of 2025, including its chip division. Samsung's IT services affiliate, Samsung SDS America, warned in a separate June notice that 179 roles could be cut in Ridgefield Park, New Jersey.

Ukraine, which is fighting a years-long war of attrition against Russian aggression, has seen a lot of combat innovations, especially in drone warfare. But even as modern technologies like Starlink-powered long-range drones and jam-proof fiber optic drones enter the battlefield, troops on both sides are also relying on their own ingenuity and deploying “field upgrades” on their equipment. Just like during the Second World War, when some American tankers welded thick metal plates to increase the armor on their vehicles to counter the feared German Panzers, the Defence Blog reports that Russians are crudely installing $2 magnetic compasses on some Molniya drones to combat electronic warfare.

The Molniya is a cheap, medium-range drone that relies on the open-source ExpressLRS radio control system, which lets operators control it from a potential maximum range of up to 62 miles (100 kilometers). Some versions also used Starlink connections to combat the jamming efforts of Ukrainian electronic warfare units, but Ukraine’s requirement to register Starlink units in the country before they can be used has supposedly made this strategy unviable.

The drone can also be equipped with a fiber optic cable, which renders it impervious to any sort of electronic interference. Although this limits the potential range of the Molniya to just 18 miles (30 kilometers), it allows the drone to be more effective as it reduces the chances that the operator will lose control of the unit.

GPS jamming is also a common occurrence in Ukrainian airspace. This means that drone operators could potentially get lost on their way to their targets. Because of this, many frontline Russian units have started crudely installing magnetic compasses on their drones. This navigation tool is typically installed just below the FPV camera, allowing the operator to glance down to confirm their bearings. When combined with other telemetry data, the drone pilot could locate their relative location even without GPS.

One leaked image shows that the compass is just attached to the panel near the FPV camera with five Philips screws, suggesting that this isn’t a factory modification. Some people are mocking the development of this “add-on,” saying that it’s a step backward for the supposedly “advanced” Russian arsenal. But no matter how rough and primitive it looks, the fact that it helps the Russians achieve their missions means that it’s still a cheap, effective, and rather deadly modification.

A micro-drone designed to locate and eradicate mosquitoes has passed an important milestone. Tornyol Systems shared a video where the eponymous autonomous drone chalked up its first live air-to-air kill. For some reason (perhaps demonstration visibility), the 40g (1.4 ounce) drone’s first confirmed kill on video features a moth. Tornyol boldly claims that the demo shows a significant stride has been made “towards completely eradicating mosquitoes.”

Tornyol Systems co-founder Alex Toussaint shared the above Tweet, congratulating the engineering team that has worked alongside him on the project. On the company website, there is a mosquito-hostile manifesto laid out, which provides insight into the company’s primary drone development goal.

“Mosquitoes are one of humanity's oldest and worst enemies. They kill more than 700,000 people each year — more than all current wars,” according to the firm’s mission statement. “More than 700 million people contract a mosquito-borne disease each year. They impact many countries, including the West, with thousands of cases of West Nile Virus in the US alone.” It aims to use technology, including this “small, inexpensive, and yet very fast” micro-drone, to “completely eradicate mosquitoes from areas where humans live.”

The underlying motivation of Tornyol Systems first became apparent back at Hackaday Supercon 2024 when Toussaint shared a presentation about How to Detect (and Kill) Mosquitoes With Off-the-Shelf Electronics.

Tornyol's technology

Since those initial presentations, it looks like the tech has been significantly refined and miniaturized. Currently, the platform is dependent on the LeSonar2 phased array sonar base station with 380 smartphone microphones and an Artix-7 FPGA to map the world in 3D.

This feeds the drone enough information to measure 0.1 mm movements and identify mosquitoes through their unique wingbeat signature. The micro-drone is sent commands by a PC, which also leverages “car park assist sensors, and some clever DSP” to seek and kill mozzies up to 8m (~26 feet) away.

We’ve previously reported on ground-based AI-enhanced mosquito zappers, but this is the first time we’ve seen an air-to-air solution. Tornyol says that it is rolling out deployment on embedded hardware “in the next few weeks.” I guess that's removing the need for a PC.

U.S. residents interested in purchasing an autonomous Tornyol drone and base station are being asked to stump up a refundable $100 deposit. Then, there are two payment plans available. Choose a $50-a-month subscription or an “own it forever” $1,100 one-time fee.

ASML is reportedly considering increasing prices of its existing Low-NA EUV lithography tools. The potential price hikes have already upset TSMC, ASML's largest client, reports The Information. But can ASML rapidly raise the prices of the tools it sells to chipmakers? Price adjustments are not going to happen overnight, said Roger Dassen, CFO of ASML, said during the company's quarterly earnings call. But there is one thing to remember about ASML's price hikes: They are going to affect the entire semiconductor industry.

"When it comes to Low-NA [EUV tools] pricing, of course, you know that we keep on increasing the productivity of the Low-NA tool, [which] gives us a pretty strong runway for potential price improvements going forward," said Roger Dassen, chief financial officer of ASML, during the company's quarterly earnings call. […] Given the long order lead times that we have, that does not translate into pricing effects tomorrow."

ASML just reported record results for its second quarter of 2026, with total net sales of €9.326 billion ($10.67 billion) and net income of €2.918 billion ($3.338 billion). The company now expects net sales between €43 billion ($49.2 billion) and €45 billion ($51.5 billion) in 2026, a range that comfortably exceeds its own guidance and the expectations of industry analysts.

More complex, more expensive

ASML has generally increased the average selling price of EUV scanners over successive generations as they increase in complexity and productivity, so the idea of higher pricing of EUV tools is not particularly new. The company calls the concept of 'value-based pricing' and gradually increases its average selling price (ASP) based on the value that its tools provide to its operators.

Early ASML Twinscan NXE systems were commonly discussed in the roughly €100 million–€120 million ($115 million-$137 million) range, while later production models such as the NXE:3400C and NXE:3600D moved toward roughly €140 million–€170 million ($160 million-$195 million). The latest NXE:3800E is climbing even further. High-NA EXE systems represent another major step up, at more than €350 million ($400 million) per machine according to industry reports.

Meanwhile, both productivity and performance of ASML's Twinscan NXE systems have been steadily increasing: While the NXE:3400C and NXE:3600D can process 160 – 170 wafers per hour (WPH) and feature a matched machine overlay (MMO) of ≤ 1.1nm, the NXE:3800E and NXE:3800F increase productivity to 220 WPH and 260 WPH, respectively, while increasing MMO to 0.9nm. With the NXE:4200G and NXE:4200H (which will likely feature an all-new light source), we are looking at productivity beyond 300 WPH and MMO of ≤0.8nm - ≤0.7nm.

ASML makes no secret that more advanced EUV lithography tools carry a higher average selling price than their predecessors.

"You should also recognize that the tool mix that we are going to ship next year is a different tool mix from the tool mix that we shipped this year," Dassen said. "When it comes to EUV in particular, the tool mix that we are going to ship next year will be EXE: 3800E and EXE:3800F [tools], while this year it is a combination of EXE:3600D and EXE:3800E [systems]."

ASML's value-based pricing logic is essentially: if an upgraded Low-NA machine processes more wafers and therefore generates more economic value for a fab, ASML can capture part of that additional value through a higher system price. This is what ASML has been doing for some time. That said, the important nuance in Dassen's statement is that he appears to be discussing further price increases for Low-NA EUV tools.

ASML's EUV dominance

ASML is the only supplier of EUV lithography systems on the planet; every single chipmaker must comeq to the company to get one of these scanners, so the company sells these machines years in advance. This year, the company expects to have the capacity to build 65 EUV tools and intends to increase capacity by 30% next year to around 84 – 85 EUV systems. In 2028, ASML plans to produce 110 EUV scanners.

"For 2027, we are now close to being fully covered with orders for Low-NA EUV, and we are planning to increase our Low-NA EUV capacity by around 30%," Dassen said. "Looking ahead to 2028, we have already received a significant number of Low-NA EUV orders. Strong demand forecasts from our customers have led us to investigate a further 30% capacity increase for that year."

Given the strong demand for ASML's tools, the company is in a position to increase prices, at least according to the rules of the free market. As always, comments on pricing are made in opaque corporate language, so one can make a lot out of it, or nothing at all.

"Clearly, the environment that we live in today, with the [substantial] value that our products bring to customer, of course, gives us flexibility on pricing, more so than what you would have seen in the past," Dassen said. "Of course, we are executing on that as well."

However, it is not that easy for ASML to just hike prices overnight, and there are important things to note. When ASML gets an order, and that order enters the reported backlog (which totaled €38.8 billion as of late Q4 2025), it already carries a sales value (with a possibility of inflation adjustments), so there is necessarily an agreed or otherwise contractually determined price basis attached to it.

This makes Dassen's comments particularly interesting. If ASML is already close to fully booked for 2027 Low-NA EUV capacity and has received a significant number of Low-NA orders for 2028, then much of that capacity is already represented by customer orders with associated sales values. As a result, unless ASML can subsequently renegotiate the fixed price before delivery, it cannot hike prices on tools that it intends to ship in 2027 and part of 2028. This essentially means that ASML could only book new orders that it intends to ship in 2028 (presumably for the second half of the year) and onwards at higher prices. However, ASML has not confirmed this directly.

ASML's next-generation NXE:4200G is on track to arrive in 2029. That scanner is poised to be more expensive than the NXE:3800E/NXE:3800F anyway, so the company's Low-NA EUV ASPs will inevitably get a bump that year. Of course, ASML is set to continue shipping its EXE:3800-series after 2029. The big question is whether the company will adjust the prices of these units that will ship starting in 2028 or not.

We do not know the answer to this question, unfortunately. But during the conference call, ASML's financial chief implied that the company would like to get 'rewards' from its customers not only for improved productivity, but also for other improvements of its tools, which means that ASML will stick to its value-based pricing model, but may adjust the pricing for the EXE:3800 model.

"We have always been able to show customers not just productivity upgrades, but also the value from better imaging, the value of better overlay, etc.," Dassen said. [But] you got this very strong correlation between throughput improvements and ASP. That is just the way things panned out, which, put in another way, customers were paying for the productivity upgrade, and the value that we gave them for free was the value associated with, let us say, overlay improvement, imaging quality […]. In the current environment, with the value that we bring, we are also having conversations with customers on how we get rewarded for that additional value."

TSMC's backlash against the price hikes

As ASML will likely not be able to hike prices on EUV tools that have already been pre-ordered and which are set to be delivered in the course of the next two years, existing chipmakers like TSMC will not feel the effects of the price increase at least over the next 24 months. Of course, if ASML adjusts prices of its EXE:3800-series scanners due to ship in 2028 – 2028, nobody is going to be happy. And yet, TSMC seems to be so upset that the sentiment has made it to the press. There are several reasons behind the reported backlash, but the major one seems to be strategic.

For years, TSMC has said that ASML's High-NA EUV lithography tools were too expensive, and that the company's engineers can continue innovating using Low-NA EUV systems. Also, adopting all-new High-NA EUV scanners will not be an easy undertaking in general, as the transition also requires new photoresists, photomasks, pellicles, metrology equipment, design rules, computational lithography flows, and numerous other supporting technologies and process innovations.

TSMC's mid-term expansion strategy, as well as a leading-edge roadmap through 2030, has been built around extracting more performance and resolution from conventional Low-NA EUV scanners using techniques such as improved masks, computational lithography, and multi-patterning where necessary. If ASML proceeds with its price hikes for Low-NA EUV tools, this can seriously hit one of the key economic foundations of TSMC's strategy.

TSMC's intention to avoid the usage of High-NA EUV scanners until at least 10A-class (1nm) process technology has an important economic advantage. A High-NA EUV system costs more than €350 million, whereas Low-NA systems are substantially cheaper. As a result, TSMC can choose its own trade-off between additional process steps and buying much more expensive lithography equipment. In contrast, Intel, which is set to adopt High-NA EUV lithography for its 14A fabrication process, has much less flexibility: If a particular process technology is designed around High-NA EUV layers, these extremely expensive scanners, along with other ingredients, become part of the manufacturing flow and affect wafer pricing.

One of the reasons TSMC would like to extend usage of Low-NA EUV systems is that it already has the world's largest install base of these tools, as well as mature processes, established flows, and plenty of innovations that enable it to stay ahead of the competition. However, if ASML systematically ties productivity improvements of each new Low-NA EUV generation to higher prices, the cost advantage of staying with Low-NA EUV gradually narrows, which greatly undermines TSMC's strategy.

There are several other reasons for TSMC to be particularly unhappy with ASML. Firstly, TSMC needs an enormous number of tools for its upcoming fabs in Taiwan, the U.S., and Japan. All of the company's leading-edge process technologies through 2029 rely on Low-NA EUV production tools. Even a relatively modest percentage increase applied to dozens of Low-NA EUV scanners can add billions of dollars to the company's capital expenditure (CapEx).

Secondly, ASML is essentially arguing that it deserves a share of its customers' improved economics. From TSMC's perspective, potential price adjustments are very different from charging more because a new scanner is materially more expensive to manufacture. ASML is effectively saying: 'Your fabs are more profitable, and our tools are more productive, therefore we want a piece of that.'

Thirdly, pricing negotiated today determines pricing and TSMC's economy for years to come. Now that ASML is sold out for 2027 and a significant part of 2028, it is negotiating terms for the second half of 2028 and onwards. Accepting a substantial increase now will inevitably carry through to dozens or hundreds of future Low-NA EUV systems, which directly affects TSMC's economics.

TSMC's A14 (1.4nm-class) fabrication process has made rapid progress in the last three months and is well ahead of N2 at the same stage of development, according to the company's update provided at its earnings call this week. The technology also faces strong customer interest and engagement across both smartphone and AI/HPC applications.

"A14 technology development is on track and progressing well. Internal product-like vehicle demonstrated close to 90% device performance and close to 90% 256Mb SRAM yield," said C.C. Wei, chief executive of TSMC, during the earnings call with analysts and investors.

A14 — which is expected to enter mass production in 2H 2028 — is making rapid progress in terms of performance and yield improvements. This April, the company disclosed that the production node achieved >85% target transistor performance and >80% 256Mb SRAM yield. Roughly three months later, both figures are approaching 90%, which suggests a gain of around 5% in device performance and nearly 10% in SRAM yield.

For comparison, TSMC's N2 demonstrated more than 80% of its target device performance and over 50% yield on a 256Mb SRAM test chip in April 2023. By April 2024, the process had advanced to more than 90% of its target device performance and over 80% SRAM yield. While development trajectories are not directly comparable, the figures suggest that A14 is maturing considerably faster than N2 did at a similar stage of development.

The very rapid progress of A14 compared to the relatively slow maturation of N2 at similar stages of development can probably be attributed, at least in part, to TSMC's growing experience with gate-all-around (GAA) nanosheet transistors. Back in 2023, the company barely had enough experience with the production of gate-all-around (GAA) nanosheet transistors, as N2 is its first process technology to adopt such a structure. By contrast, A14 relies on TSMC's 2nd Generation of GAA devices, so it can probably benefit from the transistor-design improvements, process refinements, and manufacturing expertise accumulated during the development and ramp of N2.

It appears TSMC has likely eliminated many of the yield limiters with A14 and N2, though keep in mind that a high 256Mb SRAM yield merely indicates low enough defect density and good process uniformity across a highly repetitive test structure, but it is not directly representative of functional or parametric yield of a commercial processor.

Nonetheless, the close to 90% device performance and close to 90% 256Mb SRAM yield about 2.5 years away from expected mass production start put TSMC's A14 progress well ahead of N2. Such progress can potentially enable TSMC to start high-volume manufacturing (HVM) using A14 ahead of schedule, provided that customer designs are ready, or initiate HVM with better-than-usual functional and parametric yields.

Speaking of customer design readiness, Wei indicated that clients strive to tape-out their A14 designs ahead of schedule, which is a good sign. It is also interesting to note that despite the fact that A14 lacks Super Power Rail backside power delivery (A12 will gain SPR in 2H 2019), it is set to be adopted not only by client processors, but also by AI/HPC applications.

"We are observing a strong level of customer interest and engagement on both smartphone and HPC/AI applications, and customer new tap-out activity is ongoing and ahead of schedule," Wei said.

A14 is TSMC's next-generation process technology that combines the company's 2nd Generation GAA nanosheet transistors with a new standard-cell architecture to improve performance, power efficiency, and transistor density. Compared with N2, TSMC expects A14 to deliver a 10% – 15% performance uplift at the same power and transistor count, or reduce power consumption by 25%–30% at the same frequency and complexity. The node is also projected to increase transistor density by around 20% for mixed designs and by 23% for logic.

Federal agents arrested 21-year-old Zyaire Dontaevious Zamarion Wilkins of North Lauderdale, Florida, on Tuesday and charged him with conspiracy to obtain information by computer for private financial gain, according to a 15-page criminal complaint first reported by WPLG Local 10. The FBI alleges Wilkins helped run an operation that embedded malware in eight video games, infected around 8,000 devices, and stole at least $220,000 from roughly 80 cryptocurrency wallets between May 2024 and February 2026. Investigators put a name to the scheme by following stolen Bitcoin to more than 150 gift cards, most of them spent on Uber Eats.

The complaint identifies the distribution channel only as a "popular digital distribution software company," but the games it lists, including BlockBlasters, Dashverse, Lunara, and PirateFi, match the titles named when the FBI began seeking victims of infected Steam games in March. The case is being prosecuted in Seattle federal court, near Valve's headquarters in Bellevue, Washington, and Wilkins' arrest is the first publicly reported in the investigation.

Wilkins allegedly financed and marketed the malware rather than writing it, with Local 10 reporting that agents had already searched the home of the unidentified developer who built the programs, and that Signal chats seized there tied Wilkins, operating under the handle Sibel.eth, to a $10,000 purchase of a remote access trojan and to discussions about tricking victims into approving transactions that emptied their wallets. That developer isn't named in the complaint and doesn't appear to have been charged.

The conspirators promoted the games on Discord, Telegram, X, and LinkedIn, and used bots to find users with large cryptocurrency holdings and message them directly, according to the complaint. Roughly 80 wallets were drained from 8,000 infections, a hit rate of about 1% consistent with that targeted approach. ZachXBT and vx-underground estimated BlockBlasters alone took more than $150,000 from between 261 and 478 victims, including $32,000 in donated cancer treatment funds taken from a Twitch streamer in September 2025.

Payments from the scheme's Bitcoin wallet went to Bitrefill, a gift card service, where the 150-plus cards were purchased, agents said. A subpoena to Uber matched the cards to an account with deliveries at Wilkins' family home and his addresses at the University of West Florida. When agents searched the North Lauderdale house a week before the arrest, they seized several devices and three cryptocurrency wallet seed phrases, one belonging to a Monero wallet. Wilkins' transaction history showed $382,000 in cryptocurrency sent or received, per the complaint.

Wilkins faces up to 10 years in prison if convicted and was scheduled to appear in Fort Lauderdale federal court on July 15. Valve, whose storefront has seen a steady run of malware incidents over the past two years, including the Chemia Early Access game that shipped with three malware strains, hadn't responded to Local 10's request for comment as of publication.

Australian Prime Minister Anthony Albanese said that the government is working to establish the “Australian Standards for A.I.,” which would stand as a national framework that AI companies must follow if they want to operate in the country. As backlash against the negative effects of data centers is making it harder for AI hyperscalers to build and expand infrastructure in the U.S. and Europe, the New York Times reports that many of these firms are now eyeing the Land Down Under for its vast lands and abundant renewable energy sources. But even before they start setting up shop in the country, Canberra wants to get ahead and ensure that these developments do not cause any problems for the general public.

“Every country on earth is grappling with these challenges right now. Australia will be the first country in the world to bring these issues into a single, national framework,” the prime minister said in his speech. One part of this policy will enforce a “legal obligation” for data centers to produce the same amount of power that they consume, ensuring that their presence does not put an unnecessary burden on the power grid that would result in increased utility prices for the average citizen. It also wants to ensure that these projects be as water efficient as possible, especially given that Australia is the driest populated continent on Earth, according to the International Groundwater Resources Assessment Center (IGRAC).

Aside from concern for data centers’ use of natural resources, the Australian government also wants to ensure that the intellectual property rights of its people are protected. Albanese said that Australian creators, including writers, musicians, artists, and news reporters, should “retain control of the price and value of their work” when used for AI training. “Anything less is theft. No country has got this right yet,” says the prime minister.

While some business industry groups expressed their support for the government’s goal, they were also a bit cautious, saying that overregulation could mean that Australia will miss out on the opportunities the AI data centers will bring for the company. University of New South Wales in Sydney professor Toby Walsh, who specializes in Artificial Intelligence, also told The Times that the PM is on the right track, as it addresses the concerns that most Australians have regarding AI and the infrastructure behind it. However, these are just planned policies, and the regulation behind them must still be worked out. “The devil will be in the details exactly what they do,” says Prof. Walsh.

Beijing-based Moonshot AI has released Kimi K3, a 2.8 trillion parameter model that the company describes in its technical blog as the world's first open 3T-class system and the largest open-weight AI model to date. Moonshot said K3 still sits behind Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol on overall performance, but it outperformed every other model in the company's evaluation suite, including Claude Opus 4.8 and GPT 5.5, across coding and agentic benchmarks. The model has a 1 million token context window, native vision, and activates just 16 of its 896 experts per token, roughly 1.8% of the pool. Full weights are due by July 27.

Arena ranked K3 first in its Frontend Code evaluation at 1,679 points, ahead of Fable 5, in blind developer testing. API pricing is $0.30 per million cache-hit input tokens, $3 per million on cache misses, and $15 per million output tokens. Kimi K2 launched a year ago at $0.60 per million input tokens, so uncached K3 input costs five times as much.

Moonshot claims roughly a 2.5x improvement in scaling efficiency over Kimi K2, attributed to two architectural changes: Kimi Delta Attention, a hybrid linear attention scheme, and Attention Residuals, which change how information moves between layers. Quantization-aware training starts at the supervised fine-tuning stage, using MXFP4 weights and MXFP8 activations, a combination Moonshot says it chose for broad hardware compatibility. Bank of America analysts led by Alex Liu said in a note cited by CNBC that K3 shows large-scale pre-training plus architectural work can still deliver step-change gains for flagship Chinese models despite compute constraints.

Moonshot's kernel optimization benchmark ran on Nvidia's H200, and what the blog identifies only as a "GPGPU from an alternative vendor," which the company didn't name. MiniTriton, a Triton-like compiler K3 built from scratch, is charted against Triton on an Nvidia L20, the cut-down Ada-based card sold into China under U.S. export rules. Moonshot recommends serving K3 on supernodes of 64 or more accelerators, keeping expert-parallel traffic inside one high-bandwidth domain. The blog doesn't say where the H200 hardware is; Congress passed a bill in January to close the offshore cloud rental loophole that gave Chinese firms remote access to restricted accelerators.

In one case study, K3 spent a single 48-hour autonomous run designing a simulated inference chip for a nano model built on its own architecture, using open-source EDA tools and the Nangate 45nm library. The design closed timing at 100 MHz within 4mm squared, packed 1.46 million standard cells and an INT4 MAC array, and sustained more than 8,700 tokens per second of simulated decode.

At the moment, every published K3 number is a claim made by Moonshot-reported or drawn from API access and can’t be verified until the weights are made public on July 27. Anthropic accused Moonshot in February of using 3.4 million Claude exchanges to train its models through distillation, and K3 now benchmarks within a few points of the models named in that complaint.

As the semiconductor industry increasingly relies on ASML's EUV lithography tools both for logic and memory production, ASML is considering increasing prices of these systems as they deliver greater productivity and better value for its customers. However, with 2027 production nearly sold out and substantial 2028 orders already booked, meaningful increases may primarily affect systems delivered from late 2028 onward. Nonetheless, the idea has already angered TSMC, ASML's largest client, reports The Information.

"When it comes to Low-NA [EUV tools] pricing, of course, you know that we keep on increasing the productivity of the Low-NA tool, [which] gives us a pretty strong runway for potential price improvements going forward," said Roger Dassen, chief financial officer of ASML, during the company's quarterly earnings call. "Given the long order lead times that we have, that... doesn't translate into pricing effects tomorrow."

Value-based pricing set to persist… in a new way

ASML has long followed what it calls value-based pricing and gradually increased its quotes based on output, patterning costs, power consumption, and other benefits its new tools offer to clients.

Usually, this involved increasing prices once in a while. For example, if early Twinscan NXE Low-NA EUV systems cost roughly €100 million – €120 million ($115 million–$137 million), the more advanced are priced starting at €170 million ($195 million. It is still well below rumored quotes for High-NA EXE scanners that exceed €350 million ($400 million). At the same time, Low-NA productivity has risen from 160–170 wafers per hour (WPH) and ≤1.1nm matched-machine overlay to 220 WPH/260 WPH with NXE:3800E/NXE:3800F at 0.9nm. Future NXE:4200G/NXE:4200H systems are expected to exceed 300 WPH and improve overlay to ≤0.8nm–≤0.7nm.

"Clearly, the environment that we live in today, with the value that our products bring to customer — it's substantial— of course, gives us flexibility on pricing, more so than what you would have seen in the past," Dassen said. "Of course, we are executing on that as well."

However, later during the call, Dassen emphasized that ASML intends to maintain its value-based approach even in the current environment of high demand and limited supply in the semiconductor world. Yet, he stressed that from now on, ASML might want to charge for things beyond just productivity.

"We have always been able to show customers not just productivity upgrades, but also the value from better imaging, the value of better overlay, etc.," Dassen said. "[But] you got this very strong correlation between throughput improvements and ASP. That is just the way things panned out," he said, suggesting that ASML shares value with its clients.

No price hikes in the short term

ASML will be unable to hike prices of Low-NA EUV systems for another couple of years. Since orders that are in ASML's backlog already carry a sales value, subject to inflation adjustments, prices for much of the 2027 and early 2028 output may already be contractually determined. Unless existing contracts can be renegotiated, higher pricing could therefore primarily apply to 2028 shipments and beyond, or for new orders that somehow get squeezed in in 2027. The NXE:4200G, due in 2029, should naturally lift average selling prices anyway as it gets major performance improvements.

TSMC is upset

For TSMC, however, the issue is strategic. The foundry's leading-edge roadmap through 2030 relies on extending Low-NA EUV with better masks, computational lithography, and multipatterning. Until then, TSMC's strategy has always been avoiding High-NA EUV until at least its 10A-class (1 nm-class) technology. If ASML hikes prices of its future Low-NA EUV lithography systems, it will likely affect all of TSMC's plans for the next several years.

TSMC already operates the world's largest Low-NA EUV fleet and needs many more scanners for fabs in Taiwan, the U.S., and Japan as it executes its global expansion strategy. Consequently, even modest increases beyond TSMC's projections could add billions to capital spending, reduce the economic advantage of postponing High-NA, and ultimately raise its manufacturing costs. Moreover, accepting higher prices now could establish the baseline for dozens or hundreds of future systems, which will allow ASML to capture a larger share of the economic value created by increasingly productive lithography equipment.

Can this force TSMC to transition to High-NA EUV tools earlier than planned? Moving to High-NA EUV requires not only €350-million-plus scanners but also new resists, masks, pellicles, metrology, design rules, and computational lithography flows, which are likely not ready at TSMC.

A security researcher has published a method for lifting the client certificate off a Shark robot vacuum and using it to run root commands on other Shark vacuums across the same Amazon Web Services region, exposing live camera feeds, stored home maps, and Wi-Fi credentials held in plaintext. The researcher, who publishes under the handle tokay0, published the technique on Monday and says he first reported it to SharkNinja on March 1. As of the time of writing, the flaw is still unpatched, requiring a fix that sits entirely on SharkNinja's side of the cloud rather than on the robot.

The problem is an over-permissive AWS IoT policy. The certificate that a Shark vacuum uses to authenticate to Amazon's cloud broker was never restricted to the device carrying it, so a certificate pulled from one unit can subscribe to fleet-wide traffic and publish commands addressed to any device the broker serves. Those commands travel in an ordinary field called Exec_Command inside the per-device state document AWS keeps in the cloud, and a management daemon on the vacuum passes anything under 1,000 bytes from it to a shell.

The researcher tested the technique only on units he bought himself, including a cross-model reverse shell on an AV1102ARUS Shark IQ Robot Vacuum XL, which he then used to pull a live feed off that robot's onboard camera. Watching a single AWS region for 24 hours, he counted 1,517,605 unique Shark serial numbers, of which 673,816, or 44%, replied to a command probe. Those are devices observed responding, not devices he tested or compromised. Certificates are pinned to their AWS region, so a key lifted in one region only reaches devices in that region.

tokay0 says SharkNinja acknowledged his report on March 12, told him on April 27 that it was under review, and on July 3 promised a completion date by July 10 that never arrived. He also says the company downplayed the severity and questioned whether a CVE was warranted, despite a published disclosure policy that commits SharkNinja to "provide regular updates until the reported vulnerability is resolved." The company had posted nothing on the flaw as of July 16.

Remediation in this scenario doesn’t require a firmware update. Per Amazon, a non-compliant IoT policy is fixed by pushing a scoped version inside the operator's own AWS account until SharkNinja rescopes the policy or reissues the certificates.

SharkNinja's timeline is pretty similar to a vulnerability we saw with DJI Romo vacuums back in February, whereby an authorization flaw exposed roughly 6,700 vacuums, handing out camera feeds, audio, and floor plans; DJI patched it within weeks, and the researcher who found it later collected a $30,000 bounty. Cloud-side failures, where a backend fails to scope device access, have driven a run of robot-vacuum breaches and fueled interest in fully offline designs that keep mapping and camera data off any vendor cloud.