Elon Musk has slammed reports that SpaceX is developing a handheld AI device as "Utterly False," in a post on X. It follows a report from the Wall Street Journal on Wednesday claiming that SpaceX had demonstrated an early prototype of a handheld device featuring xAI's artificial intelligence technologies and a proprietary operating system to a small group of investors ahead of the company's initial public offering.

The handheld device from SpaceX is reportedly thinner than Apple's iPhone, is based on a Qualcomm Snapdragon system-on-chip, and runs a proprietary operating system, according to the WSJ story. The main selling point of the device is its artificial intelligence technologies from xAI. However, the report does not elaborate on their nature or how they integrate with the operating system. The concept device remains in its infancy and may never become a commercial product. Furthermore, assuming that it is in its early stages of development, its final look and specifications would likely differ significantly from the prototype.

Earlier this year, Reuters reported that SpaceX was developing a smartphone, citing its sources, which reportedly stressed that SpaceX has had plans for a handset for years. At the time, Musk told Reuters that a hypothetical Starlink phone would be 'not out of the question at some point,' but admitted that it would be an AI-centric device that would be 'very different than current phones.' Nonetheless, in an X post, Musk denied his company was building 'a phone.' Based on the comment Musk made today, SpaceX is still not developing a smartphone-like device.

Meanwhile, this would not be the first time Musk has denied something that later turns out to be true. After Reuters reported Tesla had canceled its inexpensive Model 2 vehicle in April, 2024, Musk replied, 'Reuters is lying (again)' without elaborating. Tesla still has not launched its low-cost electric vehicle, but prioritized Robotaxi instead. Furthermore, the company has not touched upon Model 2 in its conference calls in detail in the years that followed, essentially confirming that the entry-level Model 2 in the form that was envisioned before 2024 has been cancelled.

Building a smartphone that connects to Starlink satellites and terrestrial 5G networks is theoretically possible: there are 3GPP Release-18 and 3GPP Release-19 5G-NTN specifications that are designed to achieve just that, and many modern devices already support texting using NTNs. For a mobile device that supports both 5G and NTN networks today, the biggest challenges are to build a modem and a front-end module that can hit the right balance between performance, reliability, area, and power consumption. Of course, the constellation itself has to offer enough 'backbone' bandwidth to support global communications.

While an always-connected handheld device with advanced AI capabilities would, from many points of view, reflect Musk's long-standing ambition to create an 'everything app,' it would automatically make SpaceX compete against companies like Apple, Google, Samsung, and many other multinational conglomerates. Tough competition is arguably not something that any CEO would like to admit, especially when the product in question is in its infancy and there is an IPO ahead.

Tesla has hired Gary Jiang, an Intel veteran who most recently was responsible for installing equipment and transferring Intel's leading-edge 18A technology process from development fab in Oregon to the company's high-volume fab in Arizona, as spotted by Electrek.co. The appointment marks the first publicly identified senior leadership hire for Elon Musk's semiconductor production project, Terafab, which demonstrates Tesla's effort to build an experienced semiconductor manufacturing organization from the ground up by hiring veterans from other companies.

Tesla poaches an Intel veteran

Gary Jiang joined Tesla in June 2026 after spending over 17 years at Intel, according to his LinkedIn profile. Interestingly, there is little to glean about his current role from his LinkedIn profile, aside from noting that he is a director at Tesla. His final position at Intel was as Factory Manager, where he oversaw the construction of the production facility, the installation of fabrication equipment, factory startup, product certification, preparation for high-volume manufacturing, and, ultimately, the transfer of Intel 18A technology from the development fab in Oregon to high-volume Fab 52 in Arizona.

Earlier in his Intel career, Jiang held multiple management positions at the company's Ocotillo campus in Chandler, Arizona, where he managed technician teams accountable for startup, ramp, yield, and output for 22nm, 14nm, and 10nm-class process technologies (which include Intel 10nm SuperFin and 10nm Enhanced SuperFin/ Intel 7) at Fab 32 and Fab 42.

Tesla has been looking for a Technical Program Manager (TPM) for semiconductor infrastructure, focused on end-to-end fab program delivery, since March, but without any success, as the job listing is still listed on the company's website. Therefore, Terafab — the joint initiative between Tesla, SpaceX, and xAI — still does not have a formal leader who is going to lead the whole project.

In his most recent role at Intel, Gary Jiang worked closely with supply chain, finance, and materials logistics for new factory planning for output, wafer cost, yield, and profit & loss, according to his LinkedIn profile. He also managed the billion-dollar capital equipment and startup of the fab. Hence, without any doubt, Jiang appears to have been one of the senior manufacturing leaders responsible for building and equipping Intel's new 18A-capable manufacturing facilities in Arizona (primarily Fab 52, and potentially Fab 62 as the campus expands). However, it would still be inaccurate to say he was the person responsible for building Fab 52 alone.

Jiang's skills roughly match what one would expect from a senior manufacturing executive helping commission a new leading-edge fab, so he will be instrumental in turning a newly constructed fab shell (or even cleanroom) into a production-ready semiconductor manufacturing facility.

However, he did not oversee the entire Fab 52/Fab 62 program and was not responsible for every stage of the project — from permitting and groundbreaking to construction, tool installation, and the ramp to high-volume manufacturing. Likewise, he is unlikely to lead the Terafab project as a whole. Nonetheless, given that Terafab is set to license Intel's 14A process technology, Gary Jiang is probably among the best candidates to equip a fab for an Intel manufacturing node.

One major caveat

In fact, one of the most confusing parts about Tesla's hiring people to work at Terafab is that Tesla itself will not own any high-volume semiconductor production facilities; SpaceX will, according to Elon Musk.

In the near term, Tesla plans to build a $3 billion semiconductor R&D center at its Texas campus. The facility will house a small pilot line capable of processing a few thousand wafers per month to develop and validate new manufacturing technologies before they are scaled for commercial production.

Once the pilot line shows signs of success, SpaceX is expected to construct a full-scale high-volume manufacturing fab. However, coordinating a joint project between Tesla and SpaceX will add complexity, as major decisions require approval from both companies' boards and must undergo conflict-of-interest reviews, which will likely slow execution.

That said, given that Gary Jiang was hired by Tesla, not SpaceX, his responsibilities could be to equip and ramp a development facility at its Gigafactory Texas campus rather than build, equip, and ramp a high-volume fab for SpaceX. In any case, we are speculating here, and nothing can really stop SpaceX from hiring Jiang at some point down the line.

The U.S. Department of Justice (DOJ) has filed a memorandum in response to NAACP’s lawsuit against Elon Musk’s xAI (now SpaceX) Colossus 2 data center. According to Wired, the federal government, through the DOJ, said that stopping the natural gas turbines needed to run the xAI data center “threatens American national, economic, and energy security by seeking to shut off the power supply for artificial-intelligence innovation that supports the Department of War’s military operations.” It further said that Grok is one of only four AI models the military and security agencies use to “support mission-critical operations across Secret and Top-Secret classified networks.”

Due to those purported national security interests, DOJ lawyers have joined xAI and the state of Missisippi in requesting that the lawsuit be dismissed.

The NAACP, through the Southern Environmental Law Center (SELC), sued xAI last year after an investigation discovered that Musk’s Colossus supercomputer facility used “illegal” generators to power its AI GPUs. The Memphis supercluster was launched in just 19 days — a major feat given that most projects of this scale usually take four years, according to Nvidia CEO Jensen Huang.

The site officially powered on in July 2024, but it wasn’t May 2025 that it became fully operational after it received 150MW of power from Memphis Light, Gas, and Water (MLGW) and the Tennessee Valley Authority (TVA). But instead of waiting months to get connected to the grid, xAI bridged the gap with mobile generators to get the electricity needed to run the hundreds of thousands of power-hungry GPUs.

It turns out that xAI failed to secure the permits needed to run the majority of these units. It’s been reported that the company applied for 15 portable turbines, but thermal images show at least 35 units on site. The company is allegedly using a loophole that allows it to run these units for 364 days without needing paperwork, but the Environmental Protect Agency (EPA) has since closed this loophole.

Aside from the offending generators in the first Colossus site, which is rented out to Anthropic, 27 natural gas turbines at Colossus 2 were cited as noncompliant in the initial lawsuit filed in April 2026.

However, as of May 2026, the SELC says that 57 turbines on the site are operating on site without a permit. According to the nonprofit, this resulted in a 111% increase in nitrogen oxide exhaust, an 83% jump in PM2.5 pollutants, and an 88% uptick in formaldehyde emissions since these generators were added. The lawsuit argues that these emissions endanger public health, and that continued use of the turbines “increases risks of asthma attacks and heart disease” in the surrounding communities.

In response to the DOJ's filing, the SELC issued the following statement: “With this filing, the Trump administration is launching an unprecedented attack on the public’s ability to defend themselves from illegal pollution. This is a blatant attempt to let well-connected corporations like xAI unlawfully pollute without any consequences, putting communities across the country at risk and threatening to open the door to large-scale pay-to-pollute corruption in the process,” SELC Litigation Director Kym Myer said. “The Department of Justice’s frivolous arguments fly in the face of decades of well-established legal precedent and we look forward to fighting them in court.”

The case is National Association for the Advancement of Colored People et al v. X.AI Corp. et al, filed in the United States District Court for the Northern District of Missourt.

Google, Microsoft, and Elon Musk's xAI agreed today to give the U.S. Commerce Department's Center for AI Standards and Innovation (CAISI) access to their AI models before public release, Bloomberg reports. OpenAI and Anthropic, which had existing evaluation partnerships with the center dating to 2024, renegotiated their deals to align with priorities in Trump's AI Action Plan, the agency said.

The agreements mean that every major U.S. frontier AI lab now participates in voluntary pre-release government evaluations. CAISI has completed more than 40 model assessments to date, including evaluations of unreleased state-of-the-art systems, according to the Commerce Department.

CAISI operates within NIST and was originally established in 2023 under Biden as the AI Safety Institute. The Trump administration renamed it last June, with Commerce Secretary Howard Lutnick calling the rebrand a move away from what he called regulation "used under the guise of national security." Despite the shift in rhetoric, the center's core function has remained largely the same: evaluating frontier models for cybersecurity, biosecurity, and chemical weapons risks.

"These expanded industry collaborations help us scale our work in the public interest at a critical moment," CAISI director Chris Fall said of the new agreements. Fall took over the center after Collin Burns, a former Anthropic and OpenAI researcher, was pushed out just four days into the job. The Washington Post reported last month that White House officials were concerned about Burns's Anthropic ties, given the administration's ongoing dispute with the company. Burns had relocated across the country and given up Anthropic equity to take the position.

The center still lacks permanent legal standing, and some lawmakers have introduced draft legislation to codify it, but nothing has passed. Trump's AI Action Plan, announced in July last year, directs CAISI to serve as part of an "AI evaluations ecosystem" and lead national security-related model assessments. It also instructs regulators to explore using evaluations when applying existing law to AI systems.

Anthropic's renegotiated deal with CAISI sits alongside a separate and hostile set of interactions with the federal government. The Pentagon designated Anthropic a supply chain risk in March after it refused to lower guardrails on autonomous weapons, though a federal judge later called that move "Orwellian." Both Defense Secretary Pete Hegseth and Trump have outlined a six-month phaseout period for government use of Anthropic's tools, and two active lawsuits remain unresolved.

The new CAISI agreements also come one day after reports that the Trump administration was considering a mandatory pre-release review process for AI models via executive order, with Anthropic's Mythos model cited as the catalyst. The voluntary agreements announced Tuesday, and any potential mandatory review framework, would run in parallel, though it remains unclear how they might interact.

Microsoft used to push its AI services towards its user base, especially with the launch of the Copilot+ PC, but it seems that even the company itself does not trust its creation. According to the Microsoft Copilot Terms of Use, which was updated in October last year, the AI large language model (LLM) is designed for entertainment use only, and users should not use it for important advice. While this may be a boilerplate disclaimer, it’s quite ironic given how hard the company wants people to use Copilot for business uses and has integrated it into Windows 11.

“Copilot is for entertainment purposes only. It can make mistakes, and it may not work as intended,” the document said. “Don’t rely on Copilot for important advice. Use Copilot at your own risk.” This isn’t limited to Copilot, too. Other AI LLMs have similar disclaimers. For example, xAI says “Artificial intelligence is rapidly evolving and is probabilistic in nature; therefore, it may sometimes: a) result in Output that contains “hallucinations,” b) be offensive, c) not accurately reflect real people, places or facts, or d) be objectionable, inappropriate, or otherwise not suitable for your intended purpose.”

These may sound common sense for people familiar with how LLMs work, but, unfortunately, some people treat AI output as gospel, even those who are supposed to know better. We’ve seen this with Amazon’s services, after some AWS outages were reportedly caused by an AI coding bot after engineers let it solve an issue without oversight. The Amazon website itself has also been hit with a few “high blast radius” incidents that were linked to “Gen-AI assisted changes,” resulting in senior engineers being called up in a meeting to resolve the matter.

While generative AI is a useful tool and can indeed increase productivity, it’s still just a tool that offers no accountability for any mistakes that it might make. Because of this, people who use it must be careful to always doubt its output and double-check its results. But even if you’re aware of the limitations of current AI technology, humans are susceptible to automation bias, wherein we tend to favor the results that machines produce and ignore data that might contradict that. AI could make this phenomenon more severe, especially as it can create results that look plausible or even true with a cursory glance.

Companies in general usually add disclaimers like these to their products and services to protect themselves from lawsuits. But as AI tech companies push their AI services as the ultimate productivity hack, they might be minimizing the risks attached to the use of AI tools just to get customers paying and recoup the billions they’ve invested in hardware and talent.

After criticizing leading chipmakers for slow capacity expansion and claiming his companies need 100 – 200 billion AI processors annually, Elon Musk last week unveiled TeraFab — a chipmaker aiming to produce logic chips, HBM4 memory, and advanced packaging under one roof. Backed by an initial ~$20 billion investment, the project targets manufacturing chips consuming 1 terawatt (1 TW) of power per year using leading-edge process technology within the next several years.

But an exhaustive analysis by Tom's Hardware Premium reveals so many factors working against TeraFab, an effort designed primarily to produce chips in-house, that it appears highly unrealistic — at best a step towards partial vertical integration for Tesla, SpaceX, and xAI.

Barriers to entry in the semiconductor industry are so high that launching a new player capable of manufacturing chips in high volumes on leading-edge process technologies is nearly impossible, from both a capital investment and an expertise point of view. All new foundries established in recent decades were either spun off from leading integrated device manufacturers (Intel Foundry, GlobalFoundries, Samsung Foundry), backed by governments (Rapidus, Tata Semiconductor, Hua Hong/HLMC, SMIC), or focused on niche markets (SkyWater, Ayar Labs, Lightmatter). And many of these new players — Intel Foundry, Rapidus, and Tata — have yet to prove that they can be competitive world-class contract semiconductor makers.

TeraFab does not plan to become a foundry; its only purpose is to serve the silicon needs of Elon Musk's companies, including Tesla, SpaceX, and xAI. Yet its need for capital (from $4 to $5 trillion, depending on how you count), equipment, expertise, and a skilled workforce are extremely vast. Meeting the aforementioned needs is quite literally impossible within a realistic timeframe. Here's why.

A question of capital

Money is the most obvious challenge for Elon Musk's chip venture. To build 1 TW of AI silicon per year, Elon Musk's TeraFab will need to process the equivalent of 22.4 million Rubin Ultra GPU wafers per year, 2.716 million Vera CPU wafers per year, and 15.824 million HBM4E wafers per year, according to estimates from premier semiconductor analysis firm Bernstein. To do so, TeraFab will need from 142 to 358 fabs, the report claims.

Bernstein's calculations are based on a top-down conversion of compute demand into semiconductor manufacturing requirements. They start with Musk's goal of 1 TW of annual AI compute and translate it into the number of AI racks needed, using assumptions about Nvidia's rack power (e.g., 120 kW to 600 kW), GPUs per rack, and system architectures similar to Nvidia’s Blackwell and Rubin platforms.

They then convert those systems into chip, wafer, and fab demand using fixed assumptions for die sizes (e.g., ~825 mm² GPUs, ~800 mm² CPUs), HBM configurations, and yields. This is where Bernstein's analysis gets a bit rough: The firm assumes the capacity of a modern fab is 50,000 Wafer Starts Per Month (which is too high for a leading-edge logic fab, and too low for a DRAM fab) and that it costs $35 billion to build (which is not enough for a 50K WSPM logic fab, but may be too high for a DRAM fab). These assumptions increase the estimated costs of the whole project; while the ballpark of several trillion seems to be correct, $5 trillion may be too high.

A modern leading-edge logic fab (or rather, phase of a fab) typically has a production capacity of around 20,000 wafer starts per month (WSPM), so this facility completes roughly 240,000 wafers per year, assuming stable operation and no major yield or downtime losses. Assuming that CPUs and GPUs are made using the same node, then to produce these logic processors (on 25.116 million wafers per year), one would need 105 leading-edge wafer fabs, provided a roughly 100% yield and no downtime.

For context, TSMC shipped 15.023 million 300-mm-equivalent wafers in 2025 — the foundry's biggest year ever — which includes millions of outdated 200mm wafers and 300mm wafers processed on legacy process technologies.

A modern leading-edge 2nm-class logic fab costs roughly $25 billion to $35 billion. And while the economies of scale drive costs down, we are still talking about a $30 billion ballpark per fab, which means about $3.15 trillion for logic fabs alone, assuming there are near 100% yields and no production disruptions.

Keeping in mind that TeraFab will be a new kid on the block, it's unrealistic to expect its yields will be close to 100%; assuming instead an 80% yield, more capacity will be needed, bringing the logic fab number to 126 and the total capacity investment to $3.78 trillion. To put the number into context, TSMC currently operates between 35 and 50 300mm fab modules that have been constructed across several decades.

Memory fabs are cheaper than logic fabs and have higher capacities due to the nature of the DRAM market, but we are still talking about tens of billions of dollars per fab. A modern DRAM fab usually has production capacity between 100,000 and 200,000 WSPM, which means that with a mid-point capacity of 150,000 WSPM, one will need around 9 fabs to produce 15.824 million HBM4E wafers.

That being said, for HBM memory, effective capacity is significantly constrained by yield, stacking, and packaging, not just wafer starts. As a result, while a DRAM fab may process hundreds of thousands of wafers per month, only a fraction of that output can be converted into high-end HBM, which is why memory may become a bottleneck. In any case, with a 70% yield for HBM, we are still talking about at least 12 fabs each costing at least $20 billion, or $240 billion in total.

Bear in mind that the figure covers front-end wafer capacity only. To put this enormous cost into context, the Big Three DRAM makers (Samsung, SK hynix, and Micron) currently operate three dozen DRAM fab modules built since the early 2000s.

2.5D and 3D packaging facilities are fairly expensive, though at $2 - $3.5 billion per phase, clearly cheaper than logic fabs. Yet keeping in mind that TeraFab will need tens, or maybe even hundreds, of advanced packaging facilities to integrate AI processors and assemble HBM stacks, the company will need to invest hundreds of billions in advanced chip packaging.

In total, it looks like TeraFab must invest well north of $4 trillion to meet Elon Musk's goal of producing AI chips that would consume 1 terawatt of power per year, not including land acquisition, development of process technologies, software development, and ecosystem buildout. Yet Bernstein's calculations are even more aggressive as analysts from the company believe that investments in TeraFab will be around $5 trillion.

Raising $5 trillion would be extraordinarily difficult. For context, even the largest companies like Nvidia, Apple, and Alphabet have market capitalizations of $4.34 trillion, $3.71 trillion, and $3.5 trillion at the time of writing, so Musk would need to mobilize capital exceeding the value of the world's most valuable corporations. It's hard to imagine private fundraising, a consortium, or even a sovereign deal of this scale.

Perhaps the only way for Elon Musk to fund this initiative is to apply at once for multi-government backing, sovereign wealth funds, and hyperscalers, as well as seeking support from capital markets. While an application for government support, per se, does not hurt, it is extremely unlikely that he will get such funding. After all, the onshoring trend in the semiconductor industry limits to one the number of governments likely willing to invest in TeraFab. Yet even the U.S. government will find it hard to invest $5 trillion, given the country's annual budget of around $7 trillion.

Equipment and raw materials supply chain

At a scale of $5 trillion in the semiconductor industry within a foreseeable timeframe, constraints would extend well beyond capital and would include limited availability of equipment, construction materials, raw materials, and a sufficiently large, skilled workforce to build, operate, and maintain TeraFab's futuristic fabs.

A modern 3nm-class logic fab with a capacity of 20,000–30,000 WSPM requires 80 – 100 DUV + EUV lithography scanners, hundreds of baking and developing tools, hundreds of etching tools, hundreds of deposition tools, and over 100 metrology and inspection tools, as well as hundreds of other tools that process wafers.

In addition, a fab uses thousands of various utility subsystems (pumps, generators, chemical delivery systems, specialty gas delivery systems, etc.) that make things happen. Exact fab configurations are unknown, and many tools are clustered with multiple process chambers, so the tool count understates the number of actual processing modules. DRAM fabs use fewer tools because it is easier to make memory than logic. Nonetheless, we are still talking about thousands of litho, deposition, metrology, and inspection tools per fab.

Since leading-edge logic process technologies are particularly lithography-intensive (even though some EUV multi-patterning sequences may be substituted by machines like Applied Materials' Centura Sculpta), let's assume that a TeraFab logic fab will need 100 DUV + EUV lithography systems, which means that 126 of these fabs will need 12,600 lithography tools for logic. For context, ASML shipped 48 EUV and 131 ArFi DUV scanners in 2025 (for a total of 179 fabs), up from 44 EUV and 129 immersion DUV machines (173 in total) a year before.

As a result, it will take ASML 70 years at its current production rate to equip TeraFab with lithography scanners for logic production alone (not counting scanners for DRAM manufacturing), and that's only if it exclusively supplies them to Musk's company.

ASML has been gradually increasing its output of EUV and ArFi DUV scanners for some time, as these machines are exclusively used for advanced nodes by companies like TSMC. That being said, ASML's production capacity depends not only on its own production capacity, but also on the production capacity of its suppliers, as the company integrates tens or even hundreds of thousands of components into every scanner. Increasing TSMC's output meaningfully is hard, but it is even harder to increase the output of all its suppliers. Getting 12,600 lithography tools in a short amount of time is quite literally impossible.

Also, keep in mind that ASML currently employs 44,000 people. If it needs to assemble 70X more litho systems (assuming it gets enough components), it will probably need to increase its headcount to match the scale of a Foxconn or Walmart, rather than a semiconductor tool company.

The same applies to other suppliers of wafer fabrication equipment: They can produce a limited number of tools, and their suppliers can make them a limited number of components. Therefore, getting millions of process chambers in a few years is simply impossible.

Finally, getting enough raw materials of perfect purity required for leading-edge chip production for a venture that is larger than Intel, Samsung, and TSMC combined (and we're not even talking about memory here) will be problematic, as they will have to expand their supply chains as well. Still, it should be easier than scaling production of lithography tools.

Time is not on their side

Now that we've mentioned capital and supply-chain challenges for a semiconductor venture of Terafab's scale, it's time to talk about the thing that has driven multiple chipmakers out of business in recent decades: leading-edge process technologies.

Developing a modern, leading-edge fabrication technology requires billions of dollars and a lot of time, and while Elon Musk tends to raise enough money for his projects, time is something money cannot buy.

Development of a new leading-edge process technology is a 5+ year effort that begins with pathfinding, materials research, and transistor architecture exploration. Once the transistor architecture is figured out, engineers run countless simulations to model key physical effects such as doping profiles, strain engineering, and leakage behavior to tune these characteristics in a bid to achieve their design specifications for the whole node. This is arguably the only step in process technology development that can be licensed. For example, Rapidus licensed a 2nm GAA transistor design from IBM. Imec and CEA-Leti can also license some of their transistor-related technologies, but that's about it.

Once the transistor concept is finalized (or licensed), the real work begins: Engineers must construct hundreds of tightly interdependent process steps across FEOL, MOL, and BEOL modules. These steps cover everything from transistor formation to interconnect and metallization, as well as require atomic-scale precision in deposition, etch, lithography, and annealing. Every stage involves hundreds, or even thousands, of tunable parameters, all of which must be optimized for yield, performance, power efficiency, defectivity, and long-term reliability — a process that depends heavily on accumulated expertise, rather than licensable IP.

After individual steps are stabilized, integration becomes the primary challenge. Engineers combine modules — such as gate stacks and source/drain structures — into a logical process flow and tune sequencing and thermal budgets to avoid contamination and material degradation. On paper, this sounds like an easy task, but this integration phase effectively defines the node; it must be done in a development fab, and it cannot be outsourced or licensed.

Once device performance and yield targets are achieved in the development fab, the process must be made usable for chip designers through PDKs, SPICE models, and validated standard-cell libraries. All of these take hundreds or thousands of engineers and a lot of time.

Finally, the process technology must be transferred to a high-volume manufacturing fab, which introduces another layer of complexity. Achieving stable, high yields in a production environment is a long, iterative process that requires experienced engineering teams and continuous refinement. Even with abundant funding, this stage cannot be accelerated easily.

As a result, the key question remains whether a company that starts its semiconductor efforts from scratch can realistically complete this entire cycle — from concept to mass production — within five years. Rapidus will demonstrate whether it is possible in 2027, when it intends to start trial production of chips on its 2nm-class fabrication process.

It will be the 2030s before TeraFab can actually output chips using its own manufacturing technologies. Theoretically, if Tesla, SpaceX, and xAI have limitless demand for AI processors, TeraFab could potentially license process technologies from Tesla's foundry partners, though it remains to be seen whether production nodes can indeed be licensed and integrated in a reasonable amount of time into an existing fab.

The workforce shortfall

To meet Elon Musk's 1 TW of compute per year goal, TeraFab must operate over 150 fabs (well, fab modules, or phases) and plenty of advanced packaging facilities. These fabs and facilities must be built by people, and as TSMC discovered with its Fab 21 phase 1 in Arizona, qualified construction workers are hard to find. Finding people who will run these fabs is even harder.

A leading-edge fab employs between 4,000 and 7,000 construction workers on site at peak, depending on the scale. When we talk about the full build cycle, there are usually 10,000 or more workers involved; for example, TSMC expects 40,000 construction jobs to be created as a result of its expansion in Arizona. To complete Fab 21 phase 1, the company had to send 500 additional workers from Taiwan, perhaps because local workers were unfamiliar with TSMC's procedures.

The 150+ fabs required by TeraFab will necessitate hundreds of thousands of construction workers, which will inevitably create a labor bottleneck, especially for highly specialized workers required for cleanroom and sub-fab systems.

Once the fabs are built, they will need employees with very specific skills. All leading-edge fabs are highly automated manufacturing facilities, but they still employ thousands of people to manage, serve, and maintain them. TSMC's 20,000 WSPM Fab 21 phase 1 currently employs around 3,000 people (which includes plenty of management roles that will not be required for subsequent phases), whereas Intel's 40,000 WSPM Fab 52 has created 3,000 high-tech manufacturing jobs and 3,000 tool technician jobs in the area, along with thousands of indirect jobs.

Even assuming that next-generation advanced fab modules will require just 1,500 employees, Elon Musk's venture will need over 300,000 highly skilled people. To put the number into context, TSMC had 83,825 full-time employees serving in various capacities as of December 31, 2024. Where TeraFab can find 300,000, and whether this can be done at all, is hard to fathom.

Reality check

Elon Musk's TeraFab aims to produce AI logic chips and HBM memory, consuming 1 TW of power per year, which requires trillions of dollars and hundreds of fabs, which is far beyond current industry capacity in terms of capital, supply-chain capabilities, and skilled workforce availability.

Beyond capital, equipment constraints are severe, as ~100 fabs would require 12,600 lithography tools, while ASML shipped only 179 scanners in 2025, and there is no way it can scale up production within a reasonable timeframe.

Process technology development remains a 5+ year effort that involves hundreds of tightly integrated steps, extensive simulations, and yield optimization that cannot be licensed or accelerated easily, even with access to partners like IBM or imec.

Finally, TeraFab would require hundreds of thousands of construction workers and over 300,000 skilled employees.

Given all the capital and supply-chain limits, the project in its current form looks quite unrealistic at full scale. Yet if this is an element of partial vertical integration that will be used to make some of the chips that Tesla, SpaceX, and xAI require in-house, then why not? Perhaps Musk's real goal is far less ambitious: success for his other ventures, rather than wholesale transformation of the entire global semiconductor market.

Although the $20 billion funds injected in Elon Musk's TeraFab project — which is supposed to build logic and memory chips as well as package them under one roof — is barely enough to build a 7nm-class logic fab, Elon Musk's eventual ambitions include producing millions or billions of AI chips that consume 1 terawatt (1 TW) of power per year. This ambition by far exceeds today's industry capacity, and if Musk pursues it, he will need $5 trillion, according to Bernstein, a premier semiconductor analysis firm (via @Jukan05). Interestingly, the order of the sum is similar to what Sam Altman was seeking for his failed fab network a couple of years ago.

To build 1 TW of AI silicon per year, Elon Musk's TeraFab would need to process 22.4 million Rubin Ultra GPU wafers, 2.716 million Vera CPU wafers, and 15.824 million HBM4E wafers annually using from 142 to 358 fabs, according to Bernstein.

The firm gets these figures, which it describes as "a very rough back-of-the-envelope wafer capacity calculation," by using a top-down approach, translating rack-level power demand into required semiconductor manufacturing capacity. Using power consumptions of rack-scale systems (120 kW for Rubin to 600 kW for Rubin Ultra), analysts convert system volumes into chip counts and then into wafer demand using their die sizes, such as ~825 mm² for GPU dies, ~800 mm² for CPU dies, the number of HBM stacks, and yields.

But Bernstein seems to overstate the typical capacity of logic fabs (50,000 wafer starts per month, WSPM, instead of 20,000 WSPM), understates the capacity of DRAM fabs (50,000 WSPM instead of 100,000 – 200,000 WSPM), and assumes prices per fab at $35 million, which likely inflates total estimates even if the multi-trillion-dollar magnitude is generally correct.

Trillions for fabs and packaging facilities

Based on what we know about the modern semiconductor industry, a modern leading-edge logic fab typically delivers around 20,000 WSPM, or roughly 240,000 wafers per year. To produce 25.116 million logic wafers annually, TeraFab would require about 105 fabs at perfect yields, or 126 fabs at 80% yields. A 2nm-class capable fab costs from $25 billion to $35 billion (~$30 billion midpoint), so logic capacity alone would require around $3.15 trillion, assuming a 100% yield and $3.78 billion at 80% yield.

For context, TSMC shipped 15.023 million 300-mm-equivalent wafers in 2025, which includes 200-mm wafers and 300-mm wafers made on outdated process technologies. Also important, TSMC currently operates about 50 300-mm fab modules built over two decades.

Large-scale high-bandwidth memory (HBM) production is also crucially important for achieving Elon Musk's goals for TeraFab. Modern DRAM fabs — run by Micron, Samsung, and SK hynix — typically offer 100,000 to 200,000 WSPM (so, let us take 150,000 WSPM as the midpoint). Producing 15.824 million HBM4E wafers would require about 9 fabs at 100% yield, or ~12 fabs at 70% yield. Each of these fabs costs at least $20 billion, or roughly $240 billion for front-end memory capacity alone. However, HBM output is constrained by stacking and packaging capabilities and yields, not only by the output of DRAM devices. For comparison, the three major DRAM makers currently operate only ~30 fab modules built since the early 2000s.

Advanced packaging facilities used for 2.5D and 3D integration, as well as HBM assembly, cost around $2 billion to $3.5 billion per phase, and TeraFab would require tens or even hundreds of such facilities to assemble AI processors and HBM stacks, which means hundreds of billions of dollars in additional investment.

Altogether, TeraFab would require well north of $4 trillion, which generally aligns with Bernstein's $5 trillion estimate, excluding land, process R&D, software, and ecosystem development.

Constraints beyond money

Raising $5 trillion would be extraordinarily difficult. For context, companies like Nvidia, Apple, and Alphabet have market capitalizations of $4.34 trillion, $3.71 trillion, and $3.5 trillion, respectively, so Musk would need to mobilize capital exceeding the value of the world's most valuable corporations. It is hard to imagine a private fundraising round, consortium, or even sovereign funding of this magnitude. For example, even if the U.S. would like to fund Musk's semiconductor venture, it could not do this that easily, as its budget for this year is about $7 trillion.

The only conceivable path would involve multi-government backing, sovereign wealth funds, hyperscalers, and capital markets acting in concert. However, we doubt this is possible at all. Furthermore, at a scale of $5 trillion deployed within a foreseeable timeframe, constraints would extend beyond capital and would include limited availability of wafer fabrication equipment, construction materials, and a sufficiently large and skilled workforce to build, operate, and maintain such fabs.

Then again, does Musk really plan to build a foundry that would leave behind TSMC, Samsung, and Intel combined just to make enough chips for Tesla, SpaceX, and xAI? Well, this is an open question.

SpaceX has officially announced its acquisition of xAI, allowing the two companies to vertically integrate their operations and help Elon Musk achieve his dream of artificial intelligence in space. According to the company’s announcement, space is the only logical solution to scaling AI data centers, as we do not have enough resources on Earth to power these systems.

“Current advances in AI are dependent on large terrestrial data centers, which require immense amounts of power and cooling. Global electricity demand for AI simply cannot be met with terrestrial solutions, even in the near term, without imposing hardship on communities and the environment,” the company said in its statement. “In the long term, space-based AI is obviously the only way to scale. To harness even a millionth of our Sun’s energy would require over a million times more energy than our civilization currently uses!”

The company has already begun taking the first steps to achieving this dream with its latest FCC filing mentioning plans to launch a million satellites into orbit. These orbital data centers would directly harness the power of the sun without interference from the Earth’s atmosphere or rotation, allowing it to run more efficiently compared to terrestrial infrastructure.

This isn’t a small project, either. Musk says that “launching a million tons per year of satellites generating 100 kW of compute power per ton would add 100 gigawatts of AI compute capacity annually, with no ongoing operational or maintenance needs.” He even mentioned launching up to 1TW/year, which would make this orbital data center the most powerful one operated by an AI tech company.

Although launching satellites into space is quite an expensive and resource-intensive endeavor, Musk claims that the efficiency of these data centers would make them “the lowest cost way to generate AI compute.” This is made possible by SpaceX’s advancements with the reusable Starship rocket, which will also be launching the newer, much bigger V3 Starlink satellites this year. He also mentioned his plans of using the platform to build a manufacturing base on the moon and use it to launch up to 1,000TW/year into deep space and help humanity become a Kardashev Type II civilization.

Despite Musk’s massive financial resources, his dream still faces some challenges, which is why Nvidia CEO Jensen Huang doubts whether this project will work. For one, electronics like advanced AI chips are susceptible to cosmic radiation, corrupting data and frying circuits. There’s also the question of cooling, as the usual solutions that work on Earth’s surface aren’t applicable in space, instead relying on the vacuum of space to serve as an "infinite heatsink." And last, but not least, putting so many satellites in orbit around the Earth risks a Kessler Syndrome event, which would throw enough space junk in orbit to make launching anything — from satellites to crewed deep-space missions — an utter impossibility for the next couple of hundred years.

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

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

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

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

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

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

The U.S. Environmental Protection Agency just released a ruling that removes any permitting exemptions for all types of gas turbine generators. According to The Guardian, this move comes as some Memphis residents have been fighting a legal battle against Elon Musk and his xAI startup for its allegedly illegal use of generators to deliver power to his multiple AI data centers.

While any deployment of a gas turbine generator generally requires a permit from the authorities, the county where Musk’s first AI supercluster is located has a loophole that allows their operation without a permit, provided they are moved within 364 days. xAI took advantage of this technicality, allowing it to set up and run an AI data center with 100,000 Nvidia H200 GPUs in just 19 days — something that Nvidia CEO Jensen Huang says usually takes four years. It achieved this by using multiple portable power generators to power the site while waiting for the 150MW substation to finish construction.

Other AI hyperscalers have also started using this technique while waiting for a connection to the grid, with OpenAI planning to use gas turbines at its first Stargate site for additional power. However, EPA’s latest rule now requires air permits for gas turbines, even portables ones that are deployed on a temporary basis. More than that, companies that want to use them now have to abide by the requirements set in the federal Clean Air Act.

This is going to be a blow to companies that are rushing to build AI data centers and bring them online, even without available grid power. The U.S. electricity grid is being stretched thin by the sudden influx of power demand from millions of power-hungry AI GPUs, and many hyperscalers are now turning to on-site generators for electrical power while waiting for their site to get a grid connection. While the EPA’s ruling is not a ban on portable or even permanent turbines, it increases the regulatory burden for those who want to deploy them. Furthermore, they can no longer rely on legal loopholes in local ordinances, meaning they’ll have to follow the higher standard that the federal government has set when it comes to permits and air quality.

Elon Musk has revealed that xAI has purchased a third building in its Memphis, Tennessee, site near the Colossus 2 data center to expand its training capacity. The billionaire said on X that the structure will be named 'MACROHARDRR', an extension of his 'Macrohard' project, wherein Musk intends to build software completely from the ground up by solely using AI agents. This additional purchase will supposedly push xAI’s overall training compute capacity to a staggering 2 gigawatts.

The new building is just the beginning of the expected expenses, though. After all, xAI still must acquire the GPUs, power sources, and more for the site to be productive. And even though Musk is one of the richest people on earth, he’s still working to raise tens of billions of dollars to help fund it, as xAI is burning through over a billion dollars a month, as it races to build the most advanced AI on the planet.

Even though Musk is still working on the funding for the project, Nvidia has already reportedly signed a deal to deliver the needed GPUs for the site, helping him reach his goal of acquiring 50 million H100-equivalent GPUs in the next five years. Musk's ultimate goal, though, is to have more AI compute than everyone else combined, challenging Microsoft and other AI titans. Aside from acquiring the chips for the AI data center, the billionaire also needs to find a way to power it. It’s already been confirmed that Musk bought an overseas power plant and is shipping it to the U.S. to power Colossus 2. xAI has already set its sights on installing a gas turbine facility, which is set to supply 460MW from natural gas, helping the firm achieve its lofty compute capacity goals.

Although xAI is a relatively new entrant to the AI race, it quickly caught up with other, more established players like OpenAI due to the significant resources being poured into the project by Elon Musk. Nvidia CEO Jensen Huang even called his first Colossus project a “superhuman” effort, especially after the facility began operation after just 19 days — a feat that usually takes four years. But with other players in the AI game also spending billions of dollars on their own projects, it would be interesting (or terrifying, depending on your perspective) to see where all this expenditure will ultimately lead in the future.

Elon Musk took to social media to claim that xAI, his pet AI project, will have more computing power than everyone else in the world combined in less than five years. The billionaire said this in response to an X post that says xAI is the “best team to join if you want to innovate,” specifically mentioning efficiency (intelligence per watt/mass) and scale (total harnessed energy/matter). The post Musk was responding to also shared the assessment of semiconductor industry research group SemiAnalysis, saying that xAI painting “Macrohard” on the roof of its Colossus 2 data center shows how serious it is in challenging Microsoft’s dominance.

Macrohard is Musk’s tongue-in-cheek name for his project to build a software company from the ground up solely using AI. Even though Elon ostensibly coined the name as a joke, SemiAnalysis says that the progress behind it is very real, with the Tennessee site showing progress towards more than 400MW of computing power, with the billionaire ordering a complete powerplant from overseas to help hit his target of 2GW of computing power in just one location. More than that, xAI is raising up to $20 billion to purchase more Nvidia GPUs to further expand Colossus 2.

xAI has been pushing hard to expand its AI computing capabilities, with its first supercluster with 100,000 H200 Blackwell GPUs getting set up in just 19 days — a process which Nvidia founder and chief executive Jensen Huang says usually takes four years. It has also been growing rapidly, with Musk stating that xAI aims to target 50 million “H100-equivalent” AI GPUs in five years, with 230,000 GPUs already operational and training Grok. This could be a plausible target, especially as the billionaire is seemingly willing and able to burn through more than a billion dollars a month to see his project through.

However, we’re unsure whether Musk’s claim that xAI will have more computing power than everyone else combined will prove accurate. After all, even though he might have an enormous amount of resources behind him, other companies and institutions are also spending boatloads of money to stay ahead of the competition. For example, OpenAI has a large data center in Texas with a 300MW capacity — and it’s projected to reach gigawatt-scale by mid-2026. In addition, China is investing in its AI industry, with the government considering allocating $70 billion to domestic chip fabrication alone.

It’s conceivable for xAI to become the largest AI company in the world within the next five years. Still, it’s unlikely that it will be able to top the combined AI computing power of Amazon, Microsoft, Google, OpenAI, Oracle, and other tech giants that are also spending billions of dollars on the industry. Even if Musk throws everything he has into xAI, including his own resources, labor force, and industrial might, he likely lacks the resources, labor force, and industrial capacity to beat all of the industry combined.

In addition to hardware costs, power generation and delivery and cooling requirements will be among the main constraints for massive AI data centers in the coming years. X, xAI, SpaceX, and Tesla CEO Elon Musk argues that over the next four to five years, running large-scale AI systems in orbit could become far more economical than doing the same work on Earth.

That's primarily due to 'free' solar power and relatively easy cooling. Jensen Huang agrees about the challenges ahead of gigawatt or terawatt-class AI data centers, but says that space data centers are a dream for now.

Terawatt-class AI datacenter is impossible on Earth

"My estimate is that the cost of electricity, the cost effectiveness of AI and space will be overwhelmingly better than AI on the ground so far, long before you exhaust potential energy sources on Earth," said Musk at the U.S.-Saudi investment forum. "I think even perhaps in the four- or five-year timeframe, the lowest cost way to do AI compute will be with solar-powered AI satellites. I would say not more than five years from now."

Jensen Huang, chief executive of Nvidia, notes that the compute and communication equipment inside today's Nvidia GB300 racks is extremely small compared to the total mass, because nearly the entire structure — roughly 1.95 tons out of 2 tons — is essentially a cooling system.

Musk emphasized that as compute clusters grow, the combined requirements for electrical supply and cooling escalate to the point where terrestrial infrastructure struggles to keep up. He claims that targeting continuous output in the range of 200 GW – 300 GW annually would require massive and costly power plants, as a typical nuclear power plant produces around 1 GW of continuous power output. Meanwhile, the U.S. generates around 490 GW of continuous power output these days (note that Musk says 'per year,' but what he means is continous power output at a given time), so using the lion's share of it on AI is impossible. Anything approaching a terawatt of steady AI-related demand is unattainable within Earth-based grids, according to Musk.

" There is no way you are building power plants at that level: if you take it up to say, a [1 TW of continuous power], impossible," said Musk. You have to do that in space. There is just no way to do a terawatt [of continuous power on] Earth. In space, you have got continuous solar, you actually do not need batteries because it is always sunny in space  and the solar panels actually become cheaper because you do not need glass or framing and the cooling is just radiative."

While Musk may be right about issues with generating enough power for AI on Earth and the fact that space could be a better fit for massive AI compute deployments, many challenges remain with putting AI clusters into space, which is why Jensen Huang calls it a dream for now.

"That's the dream," Huang exclaimed.

Remains a 'dream' in space too

On paper, space is a good place for both generating power and cooling down electronics as temperatures can be as low as -270°C in the shadow. But there are many caveats. For example, they can reach +120°C in direct sunlight. However, when it comes to earth orbit, temperature swings are less extreme: –65°C to +125°C on Low Earth Orbit (LEO), –100°C to +120°C on Medium Earth Orbit (MEO), –20°C to +80°C on Geostationary Orbit (GEO), and –10°C to +70°C on High Earth Orbit (HEO).

LEO and MEO are not suitable for 'flying data centers' due to unstable illumination pattern, substantial thermal cycling, crossing of radiation belts, and regular eclipses. GEO is more feasible as there is always sunny (well, there are annual eclipses too, but they are short) and it is not too radioactive.

Even in GEO, building large AI data centers faces severe obstacles: megawatt-class GPU clusters would require enormous radiator wings to reject heat solely through infrared emission (as only radiative emission is possible, as Musk noted). This translates into tens of thousands of square meters of deployable structures per multi-gigawatt system, far beyond anything flown to date. Launching that mass would demand thousands of Starship-class flights, which is unrealistic within Musk's four-to-five-year window, and which is extremely expensive.

Also, high-performance AI accelerators such as Blackwell or Rubin as well as accompanying hardware still cannot survive GEO radiation without heavy shielding or complete rad-hard redesigns, which would slash clock speeds and/or require entirely new process technologies that are optimized for resilience rather than for performance. This will reduce feasibility of AI data centers on GEO.

On top of that, high-bandwidth connectivity with earth, autonomous servicing, debris avoidance, and robotics maintencance all remain in their infancy given the scale of the proposed projects. Which is perhaps why Huang calls it all a 'dream' for now.

Elon Musk has proposed a public coding contest between xAI’s Grok 5 and former OpenAI research lead Andrej Karpathy, comparing it to the 1997 showdown between Garry Kasparov and IBM’s Deep Blue. Karpathy declined, saying he’d rather collaborate with Grok than compete against it.

The challenge came in response to a clip from Karpathy’s recent interview on the Dwarkesh Podcast, where he argued that AGI is likely still a decade away and described Grok 5 as trailing GPT-4 by several months.

Musk, who has said Grok 5 has a 10% and rising chance of reaching AGI, took that as an invitation: “Are you down for an AI coding contest,” he posted on X, tagging Karpathy directly.

Karpathy replied that his contribution would “trend to ~zero” in such a matchup, and emphasized that he sees current models more as collaborators than adversaries.

The idea of a formal model-versus-human coding contest is not far-fetched. DeepMind earlier this year said Gemini 2.5 solved 10 of 12 problems under ICPC World Finals conditions, placing it at a gold-medal level. Both OpenAI and DeepMind have now achieved perfect 12/12 scores on the same benchmark using GPT-4 and GPT-5. These problems are drawn from university-level algorithm competitions, judged for both correctness and runtime performance, and run within strict resource and time constraints.

Earlier this year, a Polish programmer beat OpenAI’s custom model in a 10-hour head-to-head final at the AtCoder World Tour Finals, prompting speculation that it may be the last human win at the highest level. That contest was tightly controlled and fully transparent.

If Musk wants Grok 5 to be taken seriously in that class, he’ll need to subject it to the same conditions. The Deep Blue comparison only works if the match is measurable. That means fixed-length contests using a public problem set, identical access to tooling and compute, and no external inference or human assistance. The results would need to be scored independently and published in full.

Karpathy’s decision not to participate reflects a broader shift in the way machine learning practitioners talk about performance. Rather than staging head-to-head contests, many are now focused on how well models can accelerate human output. But competitive programming still offers a clear and well-defined benchmark. And so far, Grok has yet to post a score.

If xAI wants to demonstrate parity or superiority, a formal run on ICPC-grade tasks would be the obvious place to start.

Asset management firm BlackRock has joined forces with Microsoft, Nvidia, and xAI to purchase Aligned Data Centers (Aligned) through an investment consortium called the Artificial Intelligence Infrastructure Partnership (AIP). Aligned is one of the largest data center companies in the world, having a total of planned and operational capacity of 5GW spread across 50 campuses. According to the company, the data center operator is present globally, including sites in Northern Virginia, Chicago, Dallas, Ohio, Phoenix, Salt Lake City, Sao Paulo (Brazil), Queretaro (Mexico), and Santiago (Chile).

Aside from AIP, other partners in the deal also included MGX, an AI investment firm owned and funded by the Abu Dhabi government in the United Arab Emirates (UAE), BlackRock subsidiary Global Infrastructure Partners (GIP), the Kuwait Investment Authority, and Temasek, a state-owned investment company owned by the Singapore Minister for Finance.

These investments show how state-owned funds are interested in diving into artificial intelligence, either through their own infrastructure, like the Stargate AI project in the UAE, or through monetary contributions to projects like these. Other companies, like Cisco Systems, GE Vernova, and Next Era Energy, are also part of the partnership as technology and power providers.

But what’s more interesting is how some of the biggest names in AI joined together to form AIP. BlackRock, GIP, Microsoft, and MGX founded the partnership in September 2024, with Nvidia and xAI joining later. Even though it took more than a year for the group to make its first purchase, it’s a massive one — the deal is expected to cost $40 billion, making it one of the most expensive data center purchases in history.

This combination of capital and technical expertise would help Aligned stay ahead in the global AI race. “Partnering with the Consortium will accelerate our mission to deliver the infrastructure powering tomorrow’s digital economy,” Aligned Data Centers CEO Andrew Schaap said. “With AIP, MGX, and GIP’s global reach, extensive resources, and deep expertise across AI, energy, and finance, we are poised to scale faster, innovate further, and redefine what’s possible in sustainable data center infrastructure.” AIP expects to complete the deal by the first half of 2026, provided it passes regulatory oversight.

Elon Musk’s AI venture, xAI, is preparing to raise roughly $20 billion through a combination of equity and debt, with Nvidia playing a central role in the funding structure and chip supply. According to Bloomberg, the funds are being split between $7-$8 billion in new equity and as much as $12 billion in debt, routed through a special-purpose vehicle that will purchase Nvidia GPUs and lease them out to xAI. Nvidia itself is believed to be investing up to $2 billion into the equity portion, according to Bloomberg sources.

The arrangement effectively finances purchases of Nvidia’s own hardware, while ensuring that xAI secures priority access to GPUs during a period of tightening supply. It also gives Nvidia a foothold in one of the most aggressive AI training deployments in the United States. The chips are earmarked for Colossus 2, xAI’s 100 MW Memphis site, which came online earlier this year. Musk has plans to double the site’s GPU count to 200,000.

In September, Musk publicly denied rumors that claimed xAI was raising $10 billion at a $200 billion valuation, taking to X to say, “Fake news. xAI is not raising any capital right now.” The deal now described by Bloomberg is larger, more complex, and more tightly bound to Nvidia. Reuters confirmed that the chips would be used at xAI’s Memphis site, while others believe this to be a continuation of the AI industry's circular financing.

Following rumblings that Microsoft was considering adding Grok 4 to its Azure AI Foundry, the company has confirmed that the model is now available to use for its customers, following a private preview. It comes following a period of testing, which may be related to previous instances of erratic behaviour on Grok's part. Now, Microsoft has just announced that the model is available to everyone.

Grok 4 is, as described by its constituents, a "frontier intelligence" model, which means it excels at stuff like logic, scientific problem-solving, coding, advanced math, etc., and not so much at creative writing. Both OpenAI and Google are ahead in visual comprehension as well; Grok 4's multi-modal capabilities are lackluster compared to the competition. Most businesses, though, don't really care about that — all they want is options, wrapped around Microsoft's security promises.

For instance, a company might use GPT-4 for basic tasks but prefer Grok for reasoning-heavy analysis. Grok 4 being available under the Azure umbrella makes it easy to deploy and build AI agents for specific workloads. This aligns with Microsoft's efforts to essentially build an "AI supermarket" where models from every vendor are available. The only other place Grok 4 can be found (apart from xAI directly) is Oracle; Amazon is currently missing Grok 4 from its AWS Bedrock service.

Microsoft has priced Grok 4 at $5.5 per million input tokens and $27.5 per million output tokens. There are three different flavors available, too: Grok 4 Fast Reasoning for complicated analytical tasks; Grok 4 Fast Non-Reasoning for simpler jobs like summarizations; and Grok Code Fast 1 for developer workflows. All of these are supposed to be speedy, if you couldn't tell by the "fast" in their names, but they cater to different crowds. These will be available worldwide, seemingly with no restrictions, as part of Microsoft's "Global Standard" deployment category.

Musk's AI model is not without controversy. Notably, earlier this year, xAI had to delete comments by the bot after it started praising Hitler and referring to itself as 'MechaHitler.' While resolved, these issues will no doubt have played some part in the decision to exercise caution with Grok 4's rollout on Azure AI Foundry.

Grok, xAI's flagship chatbot, has become an officially sanctioned part of United States government operations. xAI and the United States inked a deal on Wednesday that guarantees the use of Grok across the federal government for $0.42 per agency, per the government's press release.

The newly reskinned "Grok for Government", based on the Grok 4 and Grok 4 Fast models, is available for all federal agencies effective immediately. The deal lasts for 18 months, terminating in March 2027, and is the longest AI contract yet signed by the government.

Elon Musk's xAI also offers step-up models focused on higher-security classifications for unknown dollar amounts on a per-agency basis. At all levels of access to Grok for Government, xAI has pledged a "dedicated engineering team" and agency training programs to support governmental Grok and encourage its most efficient use at all times.

Grok joins Meta, OpenAI, Google, and Anthropic in the list of major AI companies that have been contracted by the U.S. government in the last week. The Government Services Administration (GSA) has been moving quickly with its "OneGov" initiative, based on securing a wide field of AI agents, chatbots, and tools for government workers to boost productivity and efficiency.

"Widespread access to advanced AI models is essential to building the efficient, accountable government that taxpayers deserve," said Federal Acquisition Service Commissioner Josh Gruenbaum in the GSA press release. All other new OneGov LLM deals were also signed for incredibly low price tags, with Meta undercutting the bunch with a zero-dollar deal with the U.S. government.

This partnership with Grok may come as a surprise for many reasons. Elon Musk, the personality and CEO behind xAI, had a very well-documented and highly contentious falling out with President Donald Trump. The pair began the year as fierce political allies, with Musk heading up Trump's "DOGE" office for governmental cuts, including those that effectively killed the Biden-era CHIPS and Science Act. But that relationship fractured as Musk stepped down from the government and resulting in a public fallout on social media.

That xAI is signing deals with the U.S. government under the watch of both Trump and Musk signals to some that the rift between the two may be mending. Musk had begun to walk back his more abrasive takes on the president in recent months, and the pair have been seen together in public.

GSA's acquisition of Grok as an option for federal workers has attracted its own share of controversy. Civil rights groups signed petitions requesting the Trump administration bar Grok from government, citing the occasional ideological bent of Grok's responses as a violation of Trump's own AI Action Plan. That plan requires AI used by government to be "neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas."

Grok's X account famously began calling itself "MechaHitler" for a few days in July, accompanied by a slew of other posts praising the German Nazi Party during World War 2 and calling for anti-Semitic action. xAI has since "addressed these responses" and assures users they will not happen again.

xAI is moving faster than anyone expected on its power strategy. According to a new report from SemiAnalysis, Elon Musk’s AI startup already has 460MW of natural gas generation either installed or under construction, split between its Memphis campus and a new site across the border in Southaven, Mississippi.

The numbers check out against state filings and local reporting. Shelby County granted xAI a permit in July for 15 stationary gas turbines at its Paul R. Lowry Road facility in Memphis after months of wrangling with environmental groups, which alleged that dozens of turbines had been running without proper approval. In Mississippi, regulators issued a 12-month authorization to operate gas turbines at 2875 Stanton Road, a property xAI acquired from Duke Energy this summer, while the company builds out a permanent plant.

Equipment lists emerging from legal disclosures align with SemiAnalysis’s reporting of 12 SMT-130 turbines, rated at roughly 16MW apiece, on the Memphis side, and seven Titan 350 units in Southaven, each capable of more than 35MW. Together, that brings xAI’s on-site capacity close to half a gigawatt, or roughly the output of a midsize utility plant, stood up in less than a year.

Artificial intelligence has been the flavor of the month ever since ChatGPT ushered in a new big bang for Silicon Valley. The big players involved in this race now are all gunning for ludicrously dense GPU clusters that can scale AI operations rapidly. One of those players is Elon Musk, who owns X and its namesake xAI, which seems determined to one-up Sam Altman and OpenAI, which Musk, interestingly enough, co-founded. Last month, the titular investor said that xAI will use 50 million H100-equivalent GPUs in the coming five years, and yesterday he reiterated that goal.

It's important to note how Musk specifically says "H100 equivalent-AI compute" because he mentions his version of 50 million GPUs offering that kind of power is better (read: more efficient). Right now, Nvidia's Blackwell-based B200 GPUs are the most efficient AI accelerators on the market, but Musk's wording could suggest xAI will move away from Nvidia in the future. This could mean a pivot to AMD, or xAI developing its own accelerators with a partner like Broadcom, which already designs custom ASICs for others.

Yesterday's reply is a simple doubling down on this goal, but more interestingly, Musk teases AI compute at a magnitude not conceived yet by saying, "Eventually, billions." That implies xAI will one day have enough power to match billions of H100 GPUs. Of course, this sounds like a bit of a disconnect from reality, considering how AI is already perpetuating environmental risks and these large datacenters are constantly affecting local populations. There are also thermal and electrical requirements to consider.

Across the pond, Musk's foil, Sam Altman, has highlighted his own mission of over a million H100 GPUs by year-end, along with a vision of bringing 100 million GPUs online by some point. That's going to require as much money as the entire GDP of the UK. In comparison, at the moment, xAI is operating with around 200,000 H200 GPUs, significantly short of the 10 million per year (based on 50 over five) that Musk wants.

Meta is another competitor in this league, and its head honcho, Mark Zuckerberg, shares similar visions. He's building a "Hyperion" data center that's almost as big as Manhattan and will consume up to 5GW of power, which is nearly equal to NYC's base electrical load. In terms of actual compute, Zuckerberg has also promised to break the million AI GPUs barrier by the end of this year, but, more importantly, the company is on the cusp of developing homegrown chips that would end its reliance on foreign manufacturers.

Leading AI companies have been bragging about the number of GPUs they use or plan to use in the future. Just yesterday, OpenAI announced plans to build infrastructure to power two million GPUs, but now Elon Musk has revealed even more colossal plans: the equivalent of 50 million H100 GPUs to be deployed for AI use over the next five years. But while the number of H100 equivalents looks massive, the actual number of GPUs to be deployed may not be quite as great. Unlike the power they will consume.

50 ExaFLOPS for AI training

"The xAI goal is 50 million in units of H100 equivalent-AI compute (but much better power-efficiency) online within 5 years," Elon Musk wrote in an X post.

One Nvidia H100 GPU can deliver around 1,000 FP16/BF16 TFLOPS for AI training (these are currently the most popular formats for AI training), so 50 million of such AI accelerators will have to deliver 50 FP16/BF16 ExaFLOPS for AI training by 2030. Based on the current performance improvement trends, this is more than achievable over the next five years.

Only 650,000 Feynman Ultra GPUs

Assuming that Nvidia (and others) will continue to scale BF16/FP16 training performance of its GPUs at a pace slightly slower than with the Hopper and Blackwell generations, 50 BF16/FP16 ExaFLOPS will be achievable using 1.3 million GPUs in 2028 or 650,000 in 2029, based on our speculative guesses.

If xAI has enough money to spend on Nvidia hardware, it is even possible that the goal of getting to 50 ExaFLOPS for AI training will be achieved even earlier.

Elon Musk's xAI is already among the fastest companies to deploy the latest AI GPU accelerators to boost its training capability. The company already runs its Colossus 1 supercluster that uses 200,000 H100 and H200 accelerators based on the Hopper architecture, as well as 30,000 GB200 units based on the Blackwell architecture. In addition, the company aims to build its Colossus 2 cluster consisting of 550,000 GB200 and GB300 nodes (each of such nodes has two GPUs, so the cluster will feature over a million GPUs) with the first nodes set to come online in the coming weeks, according to Musk.

Steady performance increases

Nvidia (and other companies) recently switched to a yearly cadence of new AI accelerators release and Nvidia's schedule now resembles Intel's Tick-Tock model from back in the day though in this case we are talking about an architecture -> optimization approach using a single production node (e.g., Blackwell -> Blackwell Ultra, Rubin -> Rubin Ultra) rather than switching to a new process technology for a known architecture.

Such an approach ensures significant performance increases every year, which in turn ensures dramatic longer-term performance gains. For example, Nvidia claims its Blackwell B200 delivers 20,000 times higher inference performance than the 2016 Pascal P100, offering around 20,000 FP4 TFLOPS versus the P100’s 19 FP16 TFLOPS. Though not a direct comparison, the metric is relevant for inference tasks. Blackwell is also 42,500 times more energy efficient than Pascal when measured by joules per generated token.

Indeed, Nvidia and others are not slowing down with performance advancements. The Blackwell Ultra architecture (B300-series) offers a 50% higher FP4 performance (15 FPLOPS) compared to the original Blackwell GPUs (10 FPLOPS) for AI inference, as well as two times higher performance for BF16 and TF32 formats for AI training, yet at the cost of lower INT8, FP32, and FP64 performance. For reference, BF16 and FP16 are typical formats used for AI training (though FP8 seems to be evaluated as well), so it is reasonable to expect Nvidia to boost performance in these formats with its next-generation Rubin, Rubin Ultra, Feynman, and Feynman Ultra GPUs.

Depending on how we count, Nvidia increased FP16/BF16 performance by 3.2 times with H100 (compared to A100), then by 2.4 times with B200 (compared to H100), and then by 2.2 times with B300 (compared to B200). Actual training performance of course depends not only on pure math performance of new GPUs, but also on memory bandwidth, model size, parallelism (software optimizations and interconnect performance), and usage of FP32 for accumulations. Yet, it is safe to say that Nvidia can double the training performance (with FP16/BF16 formats) of its GPUs with each new generation.

Assuming that Nvidia can achieve the aforementioned performance increases with its four subsequent generations of AI accelerators based on the Rubin and Feynman architectures, it is easy to count that around 650,000 Feynman Ultra GPUs will be needed to get to around 50 BF16/FP16 ExaFLOPS sometime in 2029.

Gargantuan power consumption

But while Elon Musk's xAI and probably other AI leaders will probably get their 50 BF16/FP16 ExaFLOPS for AI training over the next four or five years, the big question is how much power will such a supercluster consume? And, how many nuclear power plants will be needed to feed one?

One H100 AI accelerator consumes 700W, so 50 million of these processors will consume 35 gigawatts (GW), which is equal to the typical power generated by 35 nuclear power plants, making it unrealistic to power such a massive data center today. Even a cluster of Rubin Ultra will require around 9.37 GW, which is comparable to the power consumption of French Guiana. Assuming that the Feynman architecture doubles performance per watt for BF16/FP16 compared to the Rubin architecture (keep in mind that we are speculating), a 50 ExaFLOPS cluster will still need 4.685 GW, which is well beyond 1.4 GW – 1.96 GW required for xAI's Colossus 2 data center with around a million AI accelerators.

Elon Musk's next xAI data centers are expected to house millions of AI chips and consume so much power that Elon Musk has reportedly bought a power plant overseas and intends to ship it to the U.S., according to Dylan Patel from SemiAnalysis, who outlined xAI's recent progress in a podcast. Interestingly, Musk confirmed the statement in a subsequent tweet.

Elon Musk's current xAI Colossus AI supercomputer is already one of the world's most powerful and power-hungry machines on the planet, housing some 200,000 Nvidia Hopper GPUs and consuming around an astounding 300 MW of power, and xAI has faced significant headwinds in supplying it with enough power.

The challenges only become more intense as the company moves forward — Musk faces a monumental challenge with powering his next AI data center, one that is predicted to house one million AI GPUs, thus potentially consuming the same amount of power as 1.9 million households. Here's how the data center could consume that much power, and how Musk plans to deliver it.

Beyond Colossus

Elon Musk's xAI has assembled vast computing resources and a team of talented researchers to advance the company's Grok AI models, Patel said. However, even bigger challenges lay ahead.

It is no secret that Elon Musk has already run into trouble powering his existing xAI data center. Currently, the company's main data center, Colossus, which houses 200,000 Nvidia Hopper GPUs, is located near Memphis, Tennessee. To power this machine, xAI installed 35 gas turbines that can produce 420 MW of power, as well as deploying Tesla Megapack systems to smooth out power draw. However, things are going to get much more serious going forward.

Beyond the Colossus buildout, xAI is rapidly acquiring and developing new facilities. The company has purchased a factory in Memphis that is being converted into additional data center space, big enough to power around 125,000 eight-way GPU servers, along with all supporting hardware, including networking, storage, and cooling.

A million Nvidia Blackwell GPUs will consume between 1,000 MW (1 GW) and 1,400 MW (1.4 GW), depending on the accelerator models (B200, GB200, B300, GB300) used and their configuration.

However, the GPUs are not the only load on the power system; you must also account for the power consumption of CPUs, DDR5 memory, storage, networking gear, cooling, air conditioning, power supply inefficiency, and other factors such as lighting. In large AI clusters, a useful approximation is that overhead adds another 30% to 50% on top of the AI GPU power draw, a figure typically expressed as PUE (power usage effectiveness).

That said, depending on which Blackwell accelerators xAI plans to use, a million-GPU data center will consume between 1,400 MW and 1,960 MW (given a PUE of 1.4). What can possibly power a data center with a million high-performance GPUs for AI training and inference is a big question, as this undertaking is comparable to powering the potential equivalent of 1.9 million homes.

A power plant?

A large-scale solar power plant alone is not viable for a 24/7 compute load of this magnitude, as one would need several gigawatts of panels, plus massive battery storage, which is prohibitively expensive and land-intensive.

The most practical and commonly used option is building multiple natural gas combined-cycle gas turbine (CCGT) plants, each capable of producing 0.5 MW – 1,500 MW. This approach is relatively fast to deploy (several years), scalable in phases, and easier to integrate with existing electrical grids. Perhaps, this is what xAI plans to import to the U.S.

Alternatives like nuclear reactors could technically meet the load with fewer units (each can produce around 1,000 MW) with no direct carbon emissions, but nuclear plants take much longer to design, permit, and build (up to 10 years). It is unlikely that Musk has managed to buy a nuclear power plant overseas, with plans to ship it to the U.S.

In practice, any organization attempting a 1.4 – 1.96 Gigawatt deployment — like xAI — will effectively become a major industrial energy buyer. For now, xAI's Colossus produces power onsite and purchases power from the grid; therefore, it is likely that the company's next data center will follow suit and combine a dedicated onsite plant with grid interconnections.

Apparently, because acquiring a power plant in the U.S. can take too long, xAI is reportedly buying a plant overseas and shipping it in, something that highlights how AI development now hinges not only on compute hardware and software but also on securing massive energy supplies quickly.

There's no other way

The first phase of Elon Musk’s xAI Memphis Supercluster has just reached full operational capacity as the on-site substation goes online and connects to the main power grid. According to the Greater Memphis Chamber, the site will receive 150 MW from Memphis Light, Gas, and Water (MLGW) and the Tennessee Valley Authority (TVA). Aside from that, the xAI Colossus supercomputer also boasts another 150 MW of Megapack Batteries that will serve as backup, allowing it to stay powered in case of outages or during times of increased demand.

Musk first turned on his AI cluster in July last year, which housed 100,000 Nvidia H100 GPUs on a single fabric. The xAI supercomputer was set up at such an impressive speed, as it only took the company 19 days to make it operational — something Nvidia CEO Jensen Huang said usually takes four years. However, this speed meant it had to cut some corners, such as launching without getting power from the grid, so the site used a plethora of natural gas turbine generators for its electricity needs. Initial reports say 14 generators, outputting 2.5 MW apiece, were parked on its premises, but some residents have recently complained that over 35 turbines have been spotted in the vicinity.

This development means that the site’s first phase of development can now run completely on power from the TVA, which sources about 60% of its capacity from renewable sources like hydroelectric, solar, wind, and nuclear. Because of this, xAI will now demobilize about half of the generators it temporarily used to power Colossus — the other half will have to remain, though, to deliver the electrical needs of the second phase of the Memphis Supercluster.

We do not expect this to stay on for long, though, as a second substation, which will deliver another 150 MW, is expected to come online in the Fall of this year. This means that Colossus will have a total capacity of 300 MW — enough to power 300,000 homes. This is a massive power requirement, and there were previously some concerns about whether the TVA has enough capacity to accommodate it. The power provider has reassured various stakeholders that it can deliver that demand without affecting the supply to everyone else.

The road to powering a million GPUs

Shipping graphics cards with just 8GB of VRAM is tantamount to "bringing a butter knife to a gunfight," opines the Grok AI, built-into Twitter/X. The AI agent was commenting on a thread about the recent RTX 5060 Ti 8GB performance analysis – one which showed this model may be up to 10% slower than the 16GB variant in popular games. Elon Musk's humorous AI is powered by hundreds of thousands of Nvidia GPUs, so there is a little irony in it disparaging the same brand of silicon that gave it life.

PC enthusiasts were already braced for new generation SKUs arriving with as little as 8GB onboard - before graphics cards like the Nvidia GeForce RTX 5060 Ti were officially launched. Nevertheless, seeing these fears materialize was still painful. And this feeling of disappointment now looks set to continue, with the drip-drip of analysis of GPU commentators sharing benchmarks and 'told you so' tales. We must also add in to the unhappy mix the certainty that newer titles will only be pushing VRAM demands higher.

Grok - this 'humorous AI' isn't joking, or hallucinating

Responding to PunmasterStp on X, Grok highlighted that "Modern AAA games are chomping through VRAM faster than a kid with a bag of candy—especially at 1440p or 4K with all those juicy high-res textures and ray tracing bells and whistles." It went on to contrast the RTX 5060 Ti 8GB and 16GB variants. Users of the former will see it prematurely age with "stutters, texture pop-ins, and even crashes in heavy hitters like Hogwarts Legacy and Space Marine 2," Grok said. Meanwhile, the latter model, with 16GB, was said to be comfortably "cruising" in some of the same titles.

Grok continued with its unvarnished RTX 5060 Ti 8GB takedown by stating that "If you’re planning to game for the next few years without constantly tweaking settings down to potato mode, 8GB just ain’t gonna cut it." Potato mode seems a bit harsh, but the message is clear to those eyeing their budget and new/used 8GB graphics cards – save up more or adjust your expectations and preferences.

AMD is also expected to launch 8GB '60 card(s) shortly

The Colossus Supercomputer may be a colossal problem for the world's richest man.

Last July Elon Musk fired up the Colossus, a supercomputer that uses 100,000 Nvidia H100 GPUs on a single fabric. The site could draw only 7 MW at its launch, however, enough to power just 4% of its GPUs. To solve this, Musk deployed several massive mobile generators to deliver the site’s electrical demands.

This was supposed to be a temporary solution. At the same time, the Memphis facility awaited approval from the Tennessee Valley Authority (TVA) for at least 50 MW of supply and the completion of its 150 MW substation (which was supposed to be finished by 4Q24).

It’s already 2Q25, and Ars Technica reports that residents say the site is still using over 30 gas turbines, which “release harmful pollution that is tied to asthma, respiratory illnesses, and certain types of cancers.”

Currently, xAI has an ongoing application with the Memphis authorities for 15 turbines. But when the Southern Environmental Law Center (SELC) partnered with South Wings to photograph the site with thermal imaging cameras, they discovered over 30 hotspots, indicating the number of generators operating on the site.

These power sources operate without permits because a legal loophole allows generators to be used for 364 days without one. But with July 2025 coming quickly, xAI needs to have its applications approved. Otherwise, it risks slowing down its operations (or shutting down completely) without access to the electricity it needs for all its GPUs.

Thermal Imaging Shows At Least 35 Portable Methane Gas Turbines

Last year it turned out that Elon Musk’s xAI had to install additional ‘portable’ generators near its facility adjacent to Memphis, Tennessee, to power the Colossus supercomputer with over 100,000 Nvidia H100 GPUs as local power grid could not support the load. The Southern Environmental Law Center contends the generators are "illegal," yet they can keep running, reports The Guardian.

xAI’s Colossus supercomputer consumes about 150 MW of power when equipped with 100,000 H100 GPUs. More when it is upgraded to 200,000 processors. However, last July the site near Memphis, Tennessee, could only provide 8 MW of power to the facility, so Musk’s company had to install gas fueled turbines to power the colossal supercomputer.

However, the company had only sought permits for 15 units, but aerial images now show double that number in operation. Apparently, 20 gas generators are now running without the required approvals, raising alarm from environmental groups and nearby residents, according to The Guardian who refer to KeShaun Pearson, director of the advocacy group Memphis Community Against Pollution.
Pearson stated to the Shelby count board of commissioners (Memphis, Tennessee) that xAI is using 35 methane gas burning turbines while only having a permit for 15.

The company seems to be using a legal loophole that allows turbines to run without permits if they are not fixed in one location for more than 364 days. In January, xAI filed for approval of 15 turbines, but the other 20 were not included and still have no legal clearance. However, they can keep running for months before xAI gets formal approval, if local authorities actually approve them.

The Southern Environmental Law Center revealed images and calculations showing that the turbines can produce 420 MW, an amount sufficient to power a city. According to senior attorney Amanda Garcia, xAI has essentially created an unregulated power facility in a residential area, posing significant health and environmental risks.

The law center stated in a letter that these generators are a major pollution source and breach federal air quality rules, including emissions of hazardous and cancer-causing substances. They demanded that the local health agency issue an emergency halt to the operations and fine the company $25,000 for every day it continues to run them without proper authorization.

That's a rather laughable fine, frankly. The 100,000 H100 GPUs in the xAI Colossus would cost about $2.5 billion on their own — never mind the rest of the data center infrastructure and hardware. $25,000 per day would amount to just $9.1 million per year. Providing 150MW of electricity, 24/7, on the other hand, even at a price of $0.05 per kWh (we're not sure what xAI pays to run the portable generators) would be about $180,000 per day.

Small wonder that "a company representative scheduled to address the county board did not attend the meeting," according to The Guardian.

The developer who created seminal and essential Windows components like Task Manager, Zip Folders, and erm... Windows Pinball has shared a game he made in xAI's Grok 3 today. Breakout, a ball-bouncing brick-breaking arcade title initially released by Atari in 1976, was the target of Dave W Plummer's latest adventure in code. However, this time, the development was done at incredible speed using just a few choice 'prompts' and the power of Grok 3.

Spoiler alert: Plummer did indeed share the code on GitHub. You can review, download, and tinker with it yourself. He only needed to make a few prompt revisions to get Grok3-Breakout up and running in an acceptable form.

He asked Grok 3, "How about a colored version of Breakout?" Checking the results, he revised the prompt to "Make the player move automatically under computer control, and make the ball go 10% faster each time it bounces off the paddle." Plummer says the final revision was "Good, but the ball can get stuck in a vertical bounce. How did the original game handle that? Do the same! And make the player aim for remaining bricks."

If you download the raw HTML file and try it in a Javascript-enabled browser, you will notice that the game 'plays itself.' This seems to be because Plummer followed up with a couple of re-prompts: "a) Make the player automatic, [and] b) Make the player aim at the remaining bricks." So, if you want to play this version of 'Grok3-Breakout', you should use the earlier set of prompts in Grok 3, as reproduced in the above paragraph. Those feeling more adventurous could add gaming frills like power-ups, multiple balls, and/or laser cannons.

Breakout can be a fun game, and there have been many entertaining versions over the years (a personal favorite is 1987's Bolo by Meinolf Schneider for Atari ST mono mode). However, some social media commenters asked Plummer to try other titles (like Tempest or Pinball), and others decided to show off their AI-generated games/prompts.

Earlier this week, xAI made an early preview of the Grok 3 AI model available to the masses. Yesterday, I heard from xAI owner Elon Musk about his intention to create an AI gaming studio to develop 'AI games' with photo-realistic graphics. Other simple bubble and block games were shared at the time. However, how we move from recreating arcade classics with Grok 3 to photo-real masterpieces remains something of a head-scratcher. Perhaps we will have to wait for the model leaving beta and feature update rollouts that were teased during the official launch day announcement.

During a broadcast showcasing the newly-launched Grok 3, Elon Musk and xAI announced their intentions to begin building a new games studio. While details remain light for now, the developers showcased that Grok 3 can generate a facsimile of Tetris using Python.

"We're launching an AI gaming studio at xAI. If you're interested in joining us in building AI games, please join xAI," Musk said during the live stream.

No further details were given to the public regarding the nature of the studio itself, its intentions, the direction that xAI might take in what it seeks to do, or what it might develop. A social media post from an xAI employee showcased that the fledgling team currently consists of nine members — including Musk himself.

In another example posted by a user, Grok 3 can generate a version of the basic 2D title Bubble Trouble, featuring physics, collisions, and a basic 2D UI. However, the user's prompt revealed that Grok wasn't quite able to generate the retro-inspired sound effects that they had described.

Musk further chimed in to mention that xAI's new game studio was working on a way to "integrate photo-realistic graphics" into AI games.

Musk also claims that Grok can also improve the graphics resolutions of titles simply by prompting the LLM. However, it's not explained if this has to be done in a game that's currently being created by Grok itself or if it might debut as a discrete technology, like Nvidia's DLSS Super Resolution or AMD's FidelityFX Super Resolution.

However, whether or not the AI-grounded studio can create "a dynamically generated video game" that is more complex than the simple 2D titles that Grok 3 can seemingly produce remains to be seen, especially considering Musk's goal of implementing photo-realistic graphics.

While the xAI games studio is clearly in its early stages, Grok 3, xAI's latest LLM, has showcased impressive benchmarks. Both the Grok-3 and Grok-3 mini models can outperform the likes of GPT-4o, Gemini-2 Pro, DeepSeek-V3, and Claude 3.5 Sonnet across various categories.

The LLM was trained using 100,000 Nvidia H100 GPUs at the Colossus Supercluster in Memphis, utilizing parallelized training for development.

While details about Musk's new initiative remain muddy, some mainstream gaming studios are beginning to adopt mainstream AI into development workflows. Gaming giant Capcom shared in January that it's using generative AI to come up with "unique ideas" for background assets in collaboration with Google Cloud. The developer is said to be utilizing tools such as Gemini Pro, Gemini Flash, Imagen, and Vertex AI.

Right now, Grok 3 is enjoying the limelight, so it may be a while until we hear from whatever xAI is cooking up. Musk intends to seemingly build a fully-fledged games studio, instead of tools for developers to utilize and implement into their workflows.

The billionaire's intent to scale up the Colossus supercluster to over a million GPUs might speed up the development of newer AI models in the future, aiding the efforts of the fledgling studio.

Elon Musk just launched Grok 3, the latest version of xAI’s LLM that was trained at the Colossus Supercluster in Memphis, Tennessee using 100,000 Nvidia H100 GPUs. He had previously said, about a week ago, that its full release was imminent and claimed that it would outperform its rivals. Today he launched the AI model via a live stream on X (formerly Twitter) showcasing impressive performance benchmark results.

Musk began the presentation by saying “The mission of xAI and Grok is to understand the universe,” and explaining that he wants to answer questions like, “What’s going on? Where are the aliens? What is the meaning of life? How does the universe end? How did it start?” He added, “Of course, that’s to be a maximally truth-seeking AI even if that truth is sometimes at odds with what is politically correct.”

After speaking about his goals with AI, Musk proclaimed that Grok 3 is an order of magnitude more capable than Grok 2, and that it was trained in a very short period. This was likely possible because of the massive number of GPUs xAI used for parallelized training, which also took just 19 days to set up — a record time especially since Nvidia's CEO Jensen Huang said that that usually takes four years.

Grok 3 isn’t just a single LLM though — instead, it’s a family of several models, with the first ones launched being Grok 3 and Grok 3 mini. xAI also showed off Grok 3 Reasoning and Grok 3 mini Reasoning, which are similar to OpenAI 03-mini and DeepSeek R1 models and will solve problems through a step-by-step logical process.

Benchmarks shown by the xAI team reveal Grok-3 and Grok-3 mini models outperforming its competition, including Gemini-2 Pro, DeepSeek-V3, Claude 3.5 Sonnet, and GPT-4o, in several tests, including Math (AIME), Science (GPQA), and Coding (LCB). The reasoning models, which are accessible via the Grok app, also outperform the competition using the same benchmarks. Aside from this, the Grok app will have a new feature called DeepSearch, which scours the internet when questioned to then distill all the information into a single answer.

Other experts have been given access to Grok 3 in advance and were able to test these claims. For example, former Tesla Director of AI and OpenAI founder Andrej Karpathy shared his test results on X, saying that Grok 3 + Thinking feels similar to OpenAI’s o1-pro model while being a bit better than DeepSeek-R1 and Gemini 2.0 Flash Thinking. This is actually quite a feat, especially since OpenAI and Google have had a massive head start over xAI.

Grok 3 will be available to X Premium+ subscribers first. However, those who want to access more advanced features will need to sign up for SuperGrok, which is rumored to cost around $30 a month or $300 annually.

Elon Musk has previously confirmed that Grok 3 will arrive soon. This AI model was trained at the Colossus Supercluster with its 100,000 GPUs. However, the billionaire narrowed down the timescale at the World Government Summit in Dubai, suggesting Grok 3 will be ready in two or three weeks, reports Reuters. Musk also claimed that Grok 3 will be more powerful than any AI out there, including the groundbreaking DeepSeek AI.

“Grok 3 has very powerful reasoning capabilities, so in the tests that we’ve done thus far, Grok is outperforming anything that’s been released, that we’re aware of, so that’s a good sign,” said Musk during a video call where he was addressing the delegates at the international summit. He also said that it’s in the final stages of development and that it will be released in a week or two, reports the source.

When Grok 3 finally arrives, we can compare it with other existing AI LLMs and see how effective Musk’s investment in the Colossus Supercluster is. After all, he had to beg for these GPUs while having dinner with Nvidia CEO Jensen Huang, so we’re interested if his begging and billions paid off.

This is the second time that a member of the Trump administration has appeared at an international summit involving world leaders, with VP JD Vance proclaiming at the Paris AI Action Summit that the most powerful AI chips will be built in America. While it’s unclear if Musk appeared on screen as a representative of the White House or not, he is one of the heads of the Subcommittee on Delivering on Government Efficiency, also known as DOGE, which is tasked with streamlining the government and slashing federal spending.

In related news, Musk has made a $100-billion offer for OpenAI as the ChatGPT maker is seeking to transition from being a non-profit to a for-profit organization. The latter is quick to reject this bid, though, especially as Musk has previously sued it to block the move. OpenAI CEO Sam Altman has said that this move was Musk’s attempt to destabilize the company; the company also said that this offer clashes with Musk’s previous lawsuit.

While Musk is leading a takeover attempt of OpenAI, his own AI company, xAI, is currently pushing for more funding. It has recently secured $6 billion, doubling its total capital raised and putting its valuation at $50 billion. This amount should be enough to secure 100,000 Nvidia GPUs, which is precisely what he plans for the Colossus Supercluster in Memphis.

Elon Musk and a group of investors have offered $97.4 billion to acquire OpenAI's nonprofit arm that controls the company, reports the Wall Street Journal. Sam Altman, chief executive and a co-founder of Open AI, and the board of directors, have rejected the bid, stating the company's governance prevents outside control. Musk argues the nonprofit should remain committed to open-source safety-focused AI, while OpenAI is moving toward a for-profit model.

"It is time for OpenAI to return to the open-source, safety-focused force for good it once was," a statement by Elon Musk published by his lawyer Marc Toberoff reads. "We will make sure that happens."

The acquisition proposal is supported by Musk's AI company, xAI, and multiple investment companies, including Atreides Management, Baron Capital, 8VC, Valor Equity Partners, Vy Capital, and 8VC. In addition, the proposal is backed by Hollywood executive Ari Emanuel, CEO of Endeavor, and Joe Lonsdale, a co-founder of Palantir.

Musk's attorney, Marc Toberoff, submitted the offer to OpenAI's board, aiming to buy all nonprofit assets. This proposal challenges Sam Altman's plans to transition OpenAI into a traditional for-profit company while ensuring the nonprofit retains equity. OpenAI is raising billions to fund its AI infrastructure and a for-profit entity could raise more money and therefore speed up the progress of AI development in general and artificial general intelligence (AGI) in particular.

OpenAI is currently seeking to secure up to $40 billion in new funding in an investment round led by SoftBank, which is negotiating to contribute between $15 billion and $25 billion. If the company succeeds, this could increase OpenAI's valuation to $300 billion. The funding round follows a $6.6 billion funding round in October that valued OpenAI at $157 billion.

That said, OpenAI has confirmed it has no intention of accepting Musk's bid. In a message to employees, Altman emphasized that OpenAI's structure prevents any single person from taking control. He called Musk's offer a way to destabilize the rapidly developing company.

Musk, who co-founded OpenAI with Altman in 2015, left the company in 2019. Since then, OpenAI has raised substantial funding from Microsoft and other investors. Musk has taken legal action against OpenAI, claiming it abandoned its original mission by working closely with Microsoft. He has also questioned how OpenAI values its nonprofit entity as it transitions to for-profit status. In January, his lawyer contacted attorneys general in California and Delaware, urging them to oversee the valuation process to ensure fairness.

OpenAI insists Musk's legal claims are unfounded and that the nonprofit will be fairly compensated in the restructuring. The company also released documents showing Musk had previously supported a for-profit transition but withdrew when he could not gain control.

A part of OpenAI's vision is Stargate, an initiative to invest $500 billion in AI infrastructure in the U.S. over the next four years. Musk has publicly criticized Stargate, arguing that its supporters lack the necessary funding: shortly after the announcement, he called Altman a 'swindler' and claimed the project was overpromising. Altman denied Musk's allegations, but the dispute has added to the uncertainty surrounding OpenAI's transition.

HPE has secured a deal exceeding $1 billion to build AI servers for the X social platform previously known as Twitter, Bloomberg reports. The company beat out prominent AI server suppliers Dell and Supermicro for the contract. Since the deal is confidential, HPE did not comment to Bloomberg.

The deal was reportedly finalized in late 2024, following a competitive bidding process involving Dell and Supermicro. HPE, historically behind Dell and Supermicro in AI server sales, has ample reason to consider this agreement a significant endorsement of its technology. While $1 billion is a lot of money even by AI industry standards, this is a big win for HPE and a significant loss for Dell and the troubled Supermicro.

Bloomberg does not provide details about the servers HPE will supply to the microblogging platform, but most big companies rely on Nvidia AI GPUs. In general, GPUs are believed to account for half of an AI server's cost, so assuming we are dealing with GPU-based servers, a $1 billion deal would imply $500 million worth of GPUs.

Assuming that HPE and X will get a Blackwell GPU at $50,000 a unit (the actual cost of these GPUs varies), X will get a cluster of AI servers with 10,000 Blackwell GPUs. Such a cluster would offer rather formidable performance: up to 90 FP4 ExaFLOPS and 45 INT8/FP8 ExaTOPS. It remains to be seen how X plans to use this computational superpower. Of course, we are speculating as we do not know what is inside those machines.

Twitter used to buy custom servers from companies like MiTAC (Tyan), Supermicro, and Wiwynn. After Elon Musk acquired Twitter in 2022, the company ceased to buy servers from MiTAC (Tyan) and Wiwynn but retained its business deals with Supermicro. Interestingly, Musk's AI venture xAI contracted Dell to build its Colossus supercomputer for artificial intelligence with 100,000 Nvidia H100 GPUs.

Elon Musk has announced that xAI's Grok 3 large language model (LLM) has been pretrained, and took 10X more compute power than Grok 2. He did not reveal many details, but based on timing, the Grok 3 LLM was pre-trained on the Colossus supercluster, which contains some 100,000 Nvidia H100 GPUs.

"Grok 3 is coming soon," Elon Musk wrote in an X post. "Pretraining is now complete with 10X more compute than Grok 2."

Given the timing and context, this confirms previous reports that xAI's Colossus supercomputer, which boasts around 100,000 Nvidia H100 GPUs, was specifically built to accelerate large-scale AI projects. The mention of tenfold 'more compute than Grok 2' further supports the idea that Grok 3's pretraining leveraged this immense computational infrastructure. For obvious reasons, Grok 3 used data generated by users of X.

Specific details about the computational infrastructure used to train Grok 2 have not been widely disclosed, but we can figure out that it used a considerably less powerful cluster than Grok 3. Still, Grok 2 was pretrained on powerful, though not yet groundbreaking, computational resources.

Companies like xAI need systems like Colossus to keep up with competitors like OpenAI, Google DeepMind, and Anthropic. The ability to pretrain faster and at greater scale allows for quicker deployment of cutting-edge models, such as LLMs like Grok 3 or GPT-4 that contain hundreds of billions of parameters. Training these models involves trillions of floating-point operations. This is why Colossus will be expanded to 200,000 H100 and H200 GPUs in the coming months so that Grok Next will be pre-trained on an even more colossal system.

It is noteworthy that xAI plans to deploy a supercomputer powered by over a million GPUs over time. That version of Colossus will be used to train LLMs that will likely contain trillions of parameters and will be far more accurate than Grok 3 or GPT-4o. However, in addition to a greater number of parameters, newer models may feature more advanced reasoning, which brings them closer to artificial general intelligence, which is the ultimate goal for companies like xAI and OpenAI.

Elon Musk's xAI has raised $6 billion in its latest funding round, pushing its total capital raised to $12 billion at a valuation of $50 billion, reports TechCrunch. The round, which involved 97 investors, followed a doubling of xAI's valuation within six months and solidified the company's position as a growing competitor in the AI sector.

Major investors in this funding round include Nvidia, AMD, Andreessen Horowitz, Blackrock, Fidelity, Kingdom Holdings, and Sequoia Capital, among others. Only previous investors who backed Musk's earlier ventures, such as the Twitter acquisition, could participate. The minimum investment per participant was $77,593, but the identities of most investors remain undisclosed. The company plans to raise additional funds next year to sustain its growth as it seeks to challenge larger rivals in the generative AI market.

xAI has already built the Colossus supercomputer, which is powered by 100,000 Nvidia H100 GPUs. In the coming months, it is expected to expand to 200,000 Nvidia GPUs, and eventually, Elon Musk hopes to build a supercomputer with one million GPUs.

For reference, $6 billion is roughly enough to procure a supercomputer with servers containing 100,000 Nvidia GPUs at $30,000 per processor. Typically, GPUs account for around half of the cost of a supercomputer cluster. With more powerful supercomputers, xAI will be able to train more sophisticated large language models to gain an edge over OpenAI's ChatGPT and Google's Gemini.

Musk has positioned xAI to compete directly with industry leaders like OpenAI. He has accused OpenAI and its partner Microsoft of engaging in anti-competitive practices that hinder funding for alternative companies. Also, Musk claims that xAI benefits from data drawn from X. At the same time, X's recent privacy policy change allows xAI to train its models using user-generated content on the X platform. Beyond X, xAI leverages data from other Musk enterprises like Tesla and SpaceX to enhance its AI models.

xAI has rapidly advanced its technology with its flagship AI model, Grok, which powers several tools on X, including a chatbot accessible to X Premium users and image generator Flux. Unlike politically correct OpenAI, Grok can answer provocative questions while maintaining some boundaries on sensitive topics.

Grok currently supports customer service for SpaceX's Starlink internet service, and xAI is exploring potential collaborations with Tesla for R&D purposes. However, some Tesla shareholders have expressed concerns, accusing Musk of reallocating resources from Tesla to xAI and viewing the two companies as competitors.

xAI generates approximately $100 million annually, significantly trailing competitors like Anthropic and OpenAI, which are targeting billions in revenue. They also received significantly more funding: Anthropic and OpenAI have secured billions in funding. The report claims that AI venture capital activity reached $31 billion in Q3 2024, and as xAI accelerates its development, it aims to carve out a larger share of this booming market.

Elon Musk has taken legal action to prevent OpenAI from transitioning into a for-profit entity, accusing the organization of abandoning its founding mission. Now, OpenAI is firing back, claiming that Musk is trying to damage the company to set it back and benefit his xAI, a competing AI venture. Also, according to the New York Times, OpenAI claims that Musk wanted to transform OpenAI into a for-profit company and gain financially.

"I believe pretty strongly that Elon will do the right thing and that it would be profoundly un-American to use political power to the degree that Elon would hurt competitors and advantage his own businesses," Altman told the New York Times.

Sam Altman. Greg Brockman, Ilya Sutskever, and Elon Musk co-founded OpenAI in 2015 to advance artificial intelligence responsibly for the benefit of the public. Musk then left the organization in 2018, claiming that OpenAI founders Sam Altman and Greg Brockman violated the nonprofit’s principles by forming a multibillion-dollar partnership with Microsoft. Musk alleges this prioritizes profits over public good, which breaches the company's original purpose.

After launching its highly successful ChatGPT in late 2022 and partnering with various commercial companies, OpenAI became a for-profit organization.

In March, Musk filed a lawsuit in federal court accusing OpenAI of violating its founding principles. He has since expanded his claims and requested a preliminary injunction to halt OpenAI's planned restructuring, separating its nonprofit governance from its operational decisions. Musk also amended his lawsuit in November to include Microsoft as a defendant, accusing the company of steering OpenAI's priorities toward commercialization.

Since Musk now runs a competitive entity, OpenAI called his claims self-serving and disruptive to its development plans. One of OpenAI's arguments is that Musk had previously advocated for the organization to adopt a for-profit structure.

According to NYT, the company presented evidence that in 2017, Jared Birchall, head of Musk's family office, registered a company intended to be a for-profit version of OpenAI. Documents and text messages revealed that Musk sought a 50% to 60% equity stake in the proposed entity, which he claimed was necessary to finance an $80 billion Mars colonization project.

Additionally, OpenAI alleges that Musk sought personal financial gain through equity compensation in the new structure. These claims were supported by messages from Shivon Zilis, an early board member and liaison to Musk. Zilis is also the mother of at least three of Musk's children. She did not comment to the NYT.

Despite the escalating conflict, Sam Altman expressed sadness over the fallout with Musk and stated that he trusts Musk to act ethically now that he has political influence as an advisor to the Trump administration.

When Elon Musk announced plans to expand xAI's Colossus AI supercomputer from 100,000 GPUs today to 1 million GPUs in the future, the plan seemed overwhelming. But xAI will not be alone in having such a gargantuan supercomputer. Broadcom predicts that three of its clients, among hyperscalers, will deploy AI supercomputers with one million XPUs in fiscal 2027.

"As you know, we currently have three hyperscale customers who have developed their own multi-generational AI XPU roadmap to be deployed at varying rates over the next three years," said Hock Tan, President and CEO of Broadcom, at the company's Q4 2024 earnings call. "In 2027, we believe each of them plans to deploy 1,000,000 XPU clusters across a single fabric."

In addition to serving three major hyperscaler customers, Broadcom disclosed during the call that it had landed orders from two more 'hyperscalers and is in advanced development for their own next-generation AI XPUs.' It was rumored that ByteDance and OpenAI teamed up with Broadcom to develop their AI chips. Broadcom, of course, does not mention names.

Broadcom develops chips for AI, general-purpose data processing, or custom data center hardware — for multiple big-name companies, including Google and Meta. Broadcom and its customers identify the workload demands, such as AI training, inference, or data processing. Then, the company and its partners define the specifications of their chips and develop key aspects of their main differentiators, such as the architecture of processing units, leveraging Broadcom's expertise in silicon design. Broadcom then implements this architecture in silicon and equips it with platform-specific IP, caches, inter-chip interconnects, and interfaces. Broadcom-designed high-performance XPUs are then manufactured by TSMC.

Broadcom may sell its XPUs or custom ASICs directly to customers with long-term supply agreements depending on the contract with a particular customer. In addition, Broadcom may assist in developing products, charging for collaborative engineering and/or research and development efforts.

Broadcom's XPU business is a key ingredient of its strategy to capitalize on the growing demand for AI and cloud infrastructure, making it a critical player in the AI hardware ecosystem. The company believes that in 2027, the serviceable addressable market (SAM) for AI XPU and networking will be between $60 and $90 billion, and the firm is positioned to command a leading share of this market. It is unclear how the company counts SAM, though, as this year, Nvidia alone will earn about $100 billion selling its GPUs, DPUs, and networking hardware to the AI market.

Supermicro CEO Charles Liang announced on X (formerly Twitter) that he will set up shop near xAI’s Colossus AI supercomputer to help Musk achieve his dream of one million GPUs. Liang said that he plans to establish “local operations/production, validation, service and support.” Although it has been previously reported that Elon Musk shifted $6 billion worth of AI server orders from Supermicro to its competitors because of its financial problems, it seems xAI still plans to order from the beleaguered server supplier.

While Nvidia is likely the biggest winner behind Musk’s planned one million GPUs, which will most probably be powered by a mix of Hopper and Blackwell GPUs, it will also require all the accouterments needed to run a complete supercomputer. This includes servers and liquid-cooling solutions, which Supermicro is known for building. And with such a massive order, it does make sense for Liang and his company to put up local operations to support Musk’s ‘Gigafactory of Compute’.

This move will benefit xAI’s operations, as Supermicro would be able to deliver its products and services to such a massive AI server with much greater efficiency. But aside from that, it would also mean that the company would invest millions, if not billions, of dollars in Memphis and the surrounding area to get its operations up and running in support of the Colossus AI supercomputer.

An investment of this scale would greatly benefit the city and the state of Tennessee, bringing in a lot of money in investments and likely creating several thousand new jobs, both directly and indirectly. However, some of the surrounding communities are also concerned about the negative impact of having such a power-hungry industry setting up shop in their area.

Some experts estimate that the 100,000 H100 GPUs Musk currently has in his Memphis facility need 155MW to run. While it already has approval from the Tennessee Valley Authority (TVA) to receive 150MW, some community members are concerned that this huge demand will cause energy prices to spike for residential consumers and that the quick solution to this problem would be to build a new gas plant.

After all, if Musk brings an additional 900,000 GPUs online in the coming years, then his AI facility would likely require over 1.5GW of power. And if other corporations, like Supermicro, set up shop locally, they would need power, too, thus putting a lot of strain on the local electricity supply grid.

The massive power requirements of AI training have pushed several tech companies to invest in nuclear power. Amazon, Google, Microsoft, and Oracle have already signed contracts to develop small modular reactors or restart old large nuclear plants. Even Meta is getting in on the action, having recently released a request for proposals. However, it seems that Elon hasn’t made any plans to invest in nuclear power just yet.

This development is good news for the labor force in Memphis and the state of Tennessee, especially as it has the potential to bring in new jobs and investments in the state and city. However, it must also consider the impact of such a power-hungry industry entering the area.

Elon Musk's AI company, xAI, is set to expand its Colossus supercomputer to over one million GPUs, reports the Financial Times. Thus, the expanded Colossus machine will be one of the most powerful supercomputers in the world. However, it will require significant investments, supply, and infrastructure availability.

Colossus, which is used to train the large language model behind Grok, already operates over 100,000 H100 processors from Nvidia and is set to double the number of GPUs shortly to become the largest supercomputer in a single building. The plan to increase the number of GPUs is underway, though this one is going to take a sizeable amount of time and effort. To accomplish the mission, xAI is working with Nvidia, Dell, and Supermicro. Furthermore, Memphis, Tennessee, where Colossus is located, has reportedly established a dedicated xAI operations team to aid the endeavor.

It is unclear whether xAI plans to use current-generation Hopper or next-generation Blackwell GPUs during the expansion. The Blackwell platform is expected to scale better than Hopper, so it makes more sense to use the upcoming technology instead of the current one. But in any case, getting the 800,000 – 900,000 AI GPUs is hard, as demand for Nvidia's products is overwhelming. Another challenge is to make 1,000,000 GPUs work in concert with maximum efficiency and, again, Blackwell would make more sense here.

The financial requirements of this expansion are colossal, of course. Acquiring GPUs — costing tens of thousands of dollars each — alongside infrastructure for power and cooling, could push investment into the tens of billions. xAI has raised $11 billion this year and recently secured another $5 billion. Currently, the company is valued at $45 billion.

Unlike rivals such as OpenAI, which partners with Microsoft for computing power, and Anthropic, supported by Amazon, xAI is independently building its supercomputing capacity. This strategy puts the company in a high-stakes race to secure advanced AI hardware, but given the scale of xAI's investments, this actually puts Musk's company ahead of its rivals.

Despite its rapid progress, xAI has faced criticism for allegedly bypassing planning permissions and the project's strain on the regional power grid. To address concerns the company has emphasized grid stability measures, including deploying Tesla's megapack technology to manage power demands.

While xAI's focus on hardware has earned acclaim, its commercial offerings remain limited. Grok reportedly lags behind leading models like ChatGPT and Google's Gemini in both sophistication and user base. However, investors view Colossus as a foundational achievement that demonstrates xAI's ability to rapidly deploy cutting-edge technology.

Pat Gelsinger, the CEO of embattled tech company Intel, praised the xAI team after visiting Elon Musk’s Memphis Supercluster. He posted on X (formerly Twitter) that xAI uses Intel Xeon processors for its AI head node — the dedicated server that manages the entire 100,000-GPU-strong cluster — and that it’s “incredible what’s been built in such a short amount of time!”

After losing over $1.6 billion in its data center and foundry businesses, Intel is struggling. It has also missed the AI bandwagon, especially as its Gaudi 3 AI accelerator still suffers from issues.

On the flip side, Elon Musk has so far spent around $10 billion on AI training hardware this year, allowing his team to set up 100,000 Nvidia H200 GPUs in just 19 days — a process that usually takes four years, according to Nvidia CEO Jensen Huang. Although Nvidia powers Elon’s AI processors, he still needs a CPU to direct the AI cluster’s massive processing power. Pat Gelsinger’s post confirms that the xAI team chose Intel Xeon processors to drive their AI head node. Team Blue launched its latest 128-core flagship CPU called Granite Rapids in September 2024, but Gelsinger did not confirm which model the team uses.

The Intel CEO also praised Michael Dell, the founder and CEO of Dell Technologies and the current provider of xAI’s head node servers — at least those pictured here. Musk previously purchased systems from Supermicro, but it's unclear if the Dell servers are used in place of other Supermicro offerings — it is entirely possible that the company chose to use Dell head nodes while continuing to outfit the rest of the data center with Supermicro gear. The use of Dell head nodes will further rumors that Supermicro's legal issues have pushed xAI to switch suppliers, but that might not be the case.

Aside from exploring xAI’s massive AI cluster, we don’t know if Pat Gelsinger had other reasons for visiting the Memphis data center. Musk plans to double the GPUs on the site to 200,000 shortly, and he even mentioned plans to go as high as 300,000, although it seems that it will be a later phase of expansion. All these additional GPUs will still reportedly be Nvidia AI accelerators and could be Blackwell GPUs, so it’s unlikely that Pat is selling Elon some of Intel’s Gaudi 3 chips.

However, such a massive GPU purchase means that xAI would also need many CPUs, so he might try staying in Musk’s good graces to sell more Xeon chips. After all, even though xAI isn’t buying Intel’s AI chips yet, it still would do the company well if it could move its data center CPUs, helping it recover from its financial troubles revealed in August this year.

xAI, Elon Musk’s AI startup, has shifted all AI server orders from troubled Supermicro to Dell, reports UDN.com. Dell, already among the largest makers of servers, reportedly benefits from this decision, just like its suppliers, Inventec and Wistron. In contrast, losing a multi-billion business to rivals amid a potential NASDAQ delisting could be another devastating blow for Supermicro.

Dell and Supermicro used to supply Musk’s companies, including xAI and Tesla. Musk even appeared publicly with Supermicro’s CEO, Charles Liang, who revealed that xAI had made substantial purchases of Supermicro’s liquid-cooled AI servers. However, after the U.S. Department of Justice began to probe Supermicro for accounting manipulations and alleged export violations to China and Russia and its stock plummeted 35% in one day, UDN says Musk’s companies decided to shift orders away from the troubled company.

Among the largest AI server suppliers, Dell is well-positioned to absorb orders. If true, Wistron, which produces motherboards for Dell’s AI servers and does some assembly tasks, would be one of the biggest beneficiaries of this shift. In fact, Wistron is already expanding its production capacity to meet surging AI demand, particularly in its three Hsinchu facilities in Taiwan and its Mexican operations. Wistron is optimistic about the growing demand for AI servers and aims for triple-digit annual growth.

Inventec, another major supplier for Dell, would also reap the rewards from an order realignment. Inventec has long been involved in AI server production and is one of Dell’s top three global server assembly partners. This year, the company primarily supplied machines based on Nvidia’s Hopper processors. UDN claims the company will be ready to mass-produce Nvidia Blackwell-based machines (powered by B200 and B200A GPUs) in the first quarter of 2025. The company reportedly has spare manufacturing capacity in Mexico, so it will likely be able to produce more AI servers for companies formerly served by Supermicro.

Supermicro’s issues stem from delayed financial filings, putting the company at risk of being delisted from NASDAQ. To avoid delisting, Supermicro needed to submit a plan by November 16 explaining the delay and specifying when the required 10-K annual report would be filed. Since the 16th fell on a Saturday, the company’s final deadline is November 18.

If Supermicro is delisted, it could face serious financial repercussions, including a sharp decline in stock value and the immediate repayment of $1.725 billion in convertible notes, which could severely harm the company.

Elon Musk’s ‘Gigafactory of Compute,’ the xAI Colossus, received approval from the Tennessee Valley Authority in early November to receive 150MW from the state’s power grid. This increases the site’s initial supply of 8MW by almost twenty times, triggering concerns from local stakeholders about how this much power demand from xAI would impact supply reliability and power prices across the Tennessee Valley. Furthermore, Power Grid International reports that Elon plans to double the site’s computing capacity, doubling the facility’s energy requirements.

xAI spent a Herculean effort to put up this supercomputer, which took the company only 19 days to set up (versus the four years it usually takes, according to Nvidia CEO Jensen Huang). However, the site only had 8MW available at the time of its opening in July. So, Musk used massive portable power generators to meet the company’s needs. While Memphis Light, Gas & Water (MLGW) upgraded the existing substation to 50MW over the summer, running all 100,000 GPUs concurrently on the site is insufficient.

Experts estimate that Musk needs 155MW to run 100,000 GPUs, so his 150-MW request for the xAI site is conservative. Nevertheless, some people are concerned about the impact of such a demand on the state’s power supply.

“We are alarmed that the TVA Board rubberstamped xAI’s request for power without studying the impact it will have on local communities,” says Southern Environmental Law Center senior attorney Amanda Garcia. “Last year, TVA questioned power reliability and proposed a new dirty gas plant in South Memphis, and today, Board members expressed concern about the impact large industrial energy users have on power bills across the Tennessee Valley. TVA should be prioritizing families over data centers like xAI.”

Power Grid Internation reports that MLGW, the distribution company that delivers power to the xAI supercomputer, gave assurances to the Memphis City Council that xAI’s power demands “would not strain the grid or impact reliability for local customers.” Its CEO, Doug McGowen, claims that the additional 150MW it will deliver to the company is still within the utility’s peak load forecast and that it could buy more power from the TVA if needed.

This approval will give Elon Musk the power to drive his massive AI supercomputer. Experts say that data centers will need gigawatts of energy to train future AI models, something that the local power grid will likely be unable to handle without massive upgrades. That’s why many companies, including Amazon, Google, Microsoft, and Oracle, are investing in nuclear power to meet their future needs. However, this will take five years or more to deploy, so data centers must make do with the existing infrastructure until these research and development efforts bear fruit.

Tesla and xAI, Elon Musk's companies, will bring online $10 billion worth of training compute capacity by the end of this year, as observed by Sawyer Merritt, a co-founder of TwinBirch and a Tesla investor. And yet, it probably means that both companies will be somewhat behind schedule set by Elon Musk.

Elon Musk and his companies have recently been actively making announcements about AI supercomputers, so indeed, we are talking about huge investments.

In July, xAI began AI training using the Memphis Supercluster, which is set to integrate 100,000 liquid-cooled H100 GPUs. This system requires a gargantuan amount of power, drawing at least 150 MW, as the 100,000 H100 GPUs alone account for around 70 MW. The system's total cost is unknown, though GPUs alone would cost around $2 billion (if bought at $20,000 per unit), and typically, AI GPUs account for half of the cost of the whole system.

In late August, Tesla unveiled its Cortex AI cluster, equipped with an impressive 50,000 Nvidia H100 GPUs and 20,000 of Tesla's own Dojo AI wafer-sized chips. The Dojo cluster is projected to train Tesla's full self-drive (FSD) capability, so this machine is strategically vital for the company. 

As for costs, we are talking about two billion on the H100-based machine and about at least a billion on the Dojo supercomputer. That billion could be underestimated as Dojo machines are entirely custom-designed. For example, each Dojo D1 cabinet consumes more than 200 kW (to put it into context, each B200 NVL72 cabinet is expected to consume 120 kW) and therefore requires a fully custom cooling distribution unit (CDU) and power supply, which dramatically increases its cost.

Finally, in early September, xAI began operating its Colossus supercomputer, which already integrates 100,000 H100 GPUs and is expected to add 50,000 H100 and 50,000 H200 GPUs in the coming months. This giant AI supercomputer also costs billions.

XAI and Tesla probably announced spending of well over $10 billion on AI hardware this year. Of course, it will take some time before all those AI servers are installed and come online, so we can only guess the total cost of functioning AI hardware the two companies installed in 2024.

But the most ironic thing about these enormous spending is that they seem to be somewhat behind the ambitious plan that Elon Musk outlined this April, when he said that Tesla alone would spend $10 billion on AI hardware this year. 

"Tesla will spend around $10 billion this year on combined training and inference AI, the latter being primarily in car," Musk wrote in an X post. "Any company not spending at this level, and doing so efficiently, cannot compete."

While Tesla's Cortex AI cluster is probably a costly endeavor that will likely get more expensive over time should the company decide to install more Dojo or more Nvidia-based machines, we doubt that it is that costs significantly more than, say, $5 billion. As for the costs of AI inference hardware in cars, we cannot imagine that AI compute hardware in the vehicles set to be produced by Tesla this year costs $5 billion.

One of the challenges with building high-end AI data centers is connecting servers and making tens of thousands of GPUs work in concert and without problems, making network interconnections as important as GPUs. To build xAI's Colossus supercomputer, which now has 100,000 of Nvidia's Hopper processors and will expand to 200,000 H100 and H200 GPUs in the coming months, the company chose Nvidia's Spectrum-X Ethernet.

Nvidia's Spectrum-X platform includes the Spectrum SN5600 Ethernet switch, which enables port speeds up to 800 Gb/s and is built on the Spectrum-4 switch ASIC. The network platform works with Nvidia's BlueField-3 SuperNICs to deliver exceptional speed and efficiency when transferring massive data flows required for AI training. With Spectrum-X, Colossus achieves consistently high data throughput (95%) and virtually eliminates network latency issues and packet loss, allowing seamless operation at an unprecedented scale.

The green company says that traditional Ethernet would struggle to handle such a scale, often experiencing heavy congestion and low data throughput. By contrast, Spectrum-X's adaptive routing, congestion control, and performance isolation technologies tackle these issues, ensuring a stable, high-performance environment.

"AI is becoming mission-critical and requires increased performance, security, scalability and cost-efficiency," said Gilad Shainer, senior vice president of networking at Nvidia. "The Nvidia Spectrum-X Ethernet networking platform is designed to provide innovators such as xAI with faster processing, analysis and execution of AI workloads, and in turn accelerates the development, deployment and time to market of AI solutions."

Even with 100,000 Hopper GPUs, xAI's Colossus is one of the world's most powerful supercomputers for AI training. Yet, it was constructed in just 122 days, and its rapid deployment contrasts sharply with typical timelines for such massive systems, which often span months or even years. This efficiency extended to its operational setup, where training commenced 19 days after the first hardware was delivered and installed.

It remains to be seen how long it will take xAI to install 100,000 more Hopper GPUs, though it is safe to say that for a while, this will be the world's most powerful AI supercomputer, at least before Microsoft and Oracle deploy their Blackwell-based PCs.

"Colossus is the most powerful training system in the world," said Elon Musk on X. "Nice work by xAI team, NVIDIA and our many partners/suppliers."

Billionaire Elon Musk has taken to Twitter / X to boast that his remarkable xAI data center is set to double its firepower “soon.” He was commenting on the recent video exposé of his xAI Colossus AI supercomputer. In the highlighted video, TechTuber ServeTheHome was stunned when he saw the gleaming rows of Supermicro servers packed with 100,000 state-of-the-art Nvidia enterprise GPUs.

So, the xAI Colossus AI supercomputer is on course “Soon to become a 200k H100/H200 training cluster in a single building.” Its 100,000 GPU incarnation, which only just started AI training about two weeks ago, was already notable. While we think “soon” might indeed be soon in this case. However, Musk’s prior tech timing slippages (e.g., Tesla's full self-driving, Hyperloop delays, SolarCity struggles) mean we should be generally cautious about his forward-looking boasts.

The xAI Colossus has already been dubbed an engineering marvel. Importantly, praise for the supercomputer’s prowess isn’t limited to the usual Musk toadies. Nvidia CEO Jensen Huang also described this supercomputer project as a “superhuman” feat that had “never been done before.” xAI engineers must have worked very hard and long hours to set up the xAI Colossus AI supercomputer in 19 days. Typically, projects of this scale and complexity can take up to four years to get running, indicated Huang.

What will the 200,000 H100/H200 GPUs be used for? This very considerable computing resource will probably not be tasked with making scientific breakthroughs for the benefit of mankind. Instead, the 200,000 power-hungry GPUs are likely destined to train AI models and chatbots like Grok 3, ramping up the potency of its machine learning distilled ‘anti-woke’ retorts.

This isn’t the hardware endgame for xAI Collosus hardware expansion, far from it. Musk previously touted a Colossus packing 300,000 Nvidia H200 GPUs throbbing within.

At the current pace of upgrades, we could even see Musk Tweeting about reaching this 300,000 goal before 2024 is out. Perhaps, if anything delays ‘Grok 300,000,’ it could be factors outside of Musk’s control, like GPU supplies. We have also previously reported that on-site power generation had to be beefed up to cope even with stage 1 of xAI's Colossus, so that’s another hurdle - alongside complex liquid cooling and networking hardware.

Elon Musk's new expensive project, the xAI Colossus AI supercomputer, has been detailed for the first time. YouTuber ServeTheHome was granted access to the Supermicro servers within the 100,000 GPU beast, showing off several facets of the supercomputer. Musk's xAI Colossus supercluster has been online for almost two months, after a 122-day assembly.

What's Inside a 100,000 GPU Cluster

Patrick from ServeTheHome takes a camera around several parts of the server, providing a birds-eye view of its operations. The finer details of the supercomputer, like its power draw and pump sizes, could not be revealed under a non-disclosure agreement, and xAI blurred and censored parts of the video before its release. The most important things, like the Supermicro GPU servers, were left mostly intact in the footage above.

The GPU servers are Nvidia HGX H100s, a server solution containing eight H100 GPUs each. The HGX H100 platform is packaged inside Supermicro's 4U Universal GPU Liquid Cooled system, providing easy hot-swappable liquid cooling to each GPU. These servers are loaded inside racks which hold eight servers each, making 64 GPUs per rack. 1U manifolds are sandwiched between each HGX H100, providing the liquid cooling the servers need. At the bottom of each rack is another Supermicro 4U unit, this time with a redundant pump system and rack monitoring system.

These racks are paired in groups of eight, making 512 GPUs per array. Each server has four redundant power supplies, with the rear of the GPU racks revealing 3-phase power supplies, Ethernet switches, and a rack-sized manifold providing all of the liquid cooling. There are over 1,500 GPU racks within the Colossus cluster, or close to 200 arrays of racks. According to Nvidia CEO Jensen Huang, the GPUs for these 200 arrays were fully installed in only three weeks.

Because of the high-bandwidth requirements of an AI supercluster constantly training models, xAI went beyond overkill for its networking interconnectivity. Each graphics card has a dedicated NIC (network interface controller) at 400GbE, with an extra 400Gb NIC per server. This means that each HGX H100 server has 3.6 Terabit per second ethernet. And yes, the entire cluster runs on Ethernet, rather than InfiniBand or other exotic connections which are standard in the supercomputing space.

Of course, a supercomputer based on training AI models like the Grok 3 chatbot needs more than just GPUs to function. Details on the storage and CPU computer servers in Colossus are more restricted. From what we can see in Patrick's video and blog post, these servers are also mostly in Supermicro chassis. Waves of NVMe-forward 1U servers with some kind of x86 platform CPU inside hold either storage and CPU compute, also with rear-entry liquid cooling.

Outside, some heavily bundled banks of Tesla Megapack batteries are seen. The start-and-stop nature of the array with its milliseconds of latency between banks was too much for the power grid or Musk's diesel generators to handle, so some amount of Tesla Megapacks (holding up to 3.9 MWh each) are used as an energy buffer between the power grid and the supercomputer.

Colossus's Use, and Musk's Supercomputer Stable

The xAI Colossus supercomputer is currently, according to Nvidia, the largest AI supercomputer in the world. While many of the world's leading supercomputers are research bays usable by many contractors or academics for studying weather patterns, disease, or other difficult compute tasks, Colossus is solely responsible for training X's (formerly Twitter) various AI models. Primarily Grok 3, Elon's "anti-woke" chatbot only available to X Premium subscribers. ServeTheHome was also told that Colossus is training AI models "of the future"; models whose uses and abilities are supposedly beyond the powers of today's flagship AI.

Colossus's first phase of construction is complete and the cluster is fully online, but it's not all done. The Memphis supercomputer will soon be upgraded to double its GPU capacity, with 50,000 more H100 GPUs and 50,000 next-gen H200 GPUs. This will also more than double its power consumption, which is already too much for Musk's 14 diesel generators added to the site in July to handle. It also falls below Musk's promise of 300,000 H200s inside Colossus, though that may become phase 3 of upgrades.

The 50,000 GPU Cortex supercomputer in the "Giga Texas" Tesla plant is also under a Musk company. Cortex is devoted to training Tesla's self-driving AI tech through camera feed and image detection alone, as well as Tesla's autonomous robots and other AI projects. Tesla will also soon see the construction of the Dojo supercomputer in Buffalo, New York, a $500 million project coming soon. With industry speculators like Baidu CEO Robin Le predicting that 99% of AI companies will crumble when the bubble pops, it remains to be seen if Musk's record-breaking AI spending will backfire or pay off.

Elon Musk and the team behind xAI have achieved an engineering marvel, setting up a supercluster of 100,000 H200 Blackwell GPUs in a whopping 19 days. Nvidia CEO Jensen Huang told the story of Elon Musk's incredible installation prowess with members of the Tesla Owners Silicon Valley on X.

Huang describes Musk's 19-day escapade with awe and respect, calling the effort "superhuman". The team at xAI purportedly went from the "concept" phase to full-ready compatibility with Nvidia's "gear" in less than three weeks. This includes running xAI's first AI training run on the newly built supercluster as well.

From start to finish, the process involved building the massive X factory where the GPUs would reside and equipping the entire factory with liquid cooling and power to make all 200,000 GPUs operational. That's not to mention all of the coordination between Nvidia's and Elon Musk's engineering teams to get all of the hardware and infrastructure shipped and installed precisely and in a coordinated manner.

For perspective, Huang states that it takes an average data center four years to do what Elon Musk and his team were able to do in 19 days. Three years of that time alone would be dedicated to planning, while the last year would be used to ship the equipment, install it, and get it all working.

Huang also goes into detail describing how complex the networking is on Nvidia's hardware. He explains that networking Nvidia's gear isn't like networking traditional data center servers. "The number of wires that goes in one node...the back of a computer is all wires."

Elon Musk's integration of 100,000 H200 GPUs has "never been done before" (according to Jensen Huang) and probably won't be duplicated again by another company, at least not for a very long time.