Elon Musk has received the go-ahead from the Federal Trade Commission (FTC) to acquire Mesh Optical Technologies, an AI infrastructure startup that develops light-based networking hardware for data centers. Records published by the FTC on June 25 show that the regulatory body granted early termination of its antitrust review of the transaction, permitting Musk to procure Mesh. While the deal is yet to be finalized, with no official statement from either party, the government's green light indicates it’s all but done, as this was the last hurdle.

Interestingly, Mesh was founded by three former SpaceX employees who helped develop the Starlink optical communication links that keep thousands of satellites interconnected. So, why is Musk — who is simultaneously building the world's largest multibillion-dollar semiconductor manufacturing facility and an 11-million-square-foot orbital data center factory — seeking to own a company founded by his former employees? The answer appears to be optical interconnects, a critical technology that connects all three.

The connection problem: AI's latest bottleneck

As AI continues to grow in capability and user base, so do the enabling AI clusters, many of which now comprise tens to hundreds of thousands of processors. The hardest problem in scaling an AI cluster has evolved beyond making the chips faster to moving data between them. Training and inference tasks on frontier AI models are split across thousands of GPUs using parallel-computing techniques, requiring the processors to exchange enormous volumes of data every fraction of a second.

While per-chip compute capacity has raced ahead, the bandwidth linking those chips has not kept pace, a mismatch the industry refers to as the "I/O wall." The processors mostly communicate via copper interconnects, which currently dominate AI clusters. However, copper presents inherent limitations. As per-lane signaling climbs toward 200 gigabits per second (Gbps), attenuation, crosstalk, and the skin effect all worsen at higher frequencies, driving up power and corrupting the signal until passive copper becomes impractical beyond a meter or two.

To overcome these constraints, the industry is increasingly turning to optical networking, bringing the technology closer to the processor. Optical links use transceivers to convert a chip's electrical signals into light for transmission over fiber, then convert them back into electrical signals at the receiving end. They can carry far more data over much longer distances while consuming less power than equivalent high-speed copper connections, making them increasingly essential as AI clusters grow larger. Chipmakers and networking vendors are racing to deliver faster 800G and 1.6T optical transceivers while shortening electrical paths with co-packaged optics, which place the optical engine alongside the switch ASIC (application-specific integrated circuit).

This shift has transformed optical interconnects from a supporting technology into one of the industry's most strategically important AI infrastructure markets, attracting billions of dollars in investments and resulting in major partnerships for new and existing industry players. One such player is Mesh, the optical hardware startup that has drawn the interest of the world’s richest man.

A mesh solution to Musk’s ambition?

Elon Musk has been one of the most aggressive players in the AI industry. After co-founding OpenAI, he went on to launch a proprietary company, xAI, before turning his focus to building data centers. In less than two years, xAI deployed the Colossus supercomputer with over 200,000 Nvidia Hopper- and Blackwell-generation accelerators. Colossus 2, with a long-term target of 1 million GPUs, is already operational. For Musk, however, buying the chips was not enough. Why not build them, too?

Characteristic of the world's richest man’s preference for complete vertical integration, SpaceX — in collaboration with Tesla and xAI — is now building Terafab, a vertically integrated, multi-billion-dollar semiconductor manufacturing facility aimed at producing chips capable of delivering an unprecedented over 1 terawatt of AI compute capacity annually. Located in Austin, Texas, the colossal facility aims to consolidate every stage of chip production under one roof, handling everything from logic and memory fabrication to advanced packaging and testing. An ambitious project that we've also analyzed for its feasibility.

The facility's output will serve to meet the chip needs of the broader AI industry, as well as those of Musk’s xAI, self-driving vehicles, Optimus humanoid robots, and SpaceX's orbital AI data center plans. Musk says 80% of Terafab's total compute output is ultimately destined for Earth orbit to support SpaceX's orbital data centers.

“But there aren't any data centers floating around in space,” observers may point out. Introducing Gigasat, Musk's 11-million-square-foot fix for that reality. Gigasat is yet another massive facility under construction, this time for manufacturing everything needed for SpaceX’s AI1 satellite, the company's most likely world-first orbital data center with 150 kW of compute.

At first glance, everything seems in place for the next generation of Ultra-capable AI infrastructure. However, there is one critical missing piece in this stack, one that we've established earlier. Hundreds of gigawatts of extremely powerful silicon are not particularly useful if the data can't move between the dies fast enough in AI clusters, whether on the ground or in space. The industry-prevalent copper hits a wall long before you reach the scale Musk is chasing. Hence, the need for the missing piece: optical interconnects.

This brings us to Mesh, a manufacturer of precisely that missing piece. Mesh Optical Technologies is a US optical communications startup that develops high-speed optical interconnect hardware — optical transceivers that convert a chip's electrical signals into light for high-speed transmission over fiber — for AI data centers and space communications.

Its flagship product, the Alpha C1, supports 800G and 1.6T data rates and reportedly draws about a third of the power of competing modules, using a flip-chip die-bonding process the company says makes the optical engine repeatable at the volume — potentially millions of links — that AI clusters demand.

These are the characteristics needed to seamlessly interconnect the next-generation terrestrial AI supercomputers and, potentially, future space-based computing platforms, which Terafab aims to deliver. An added benefit is the space-related experience of the three Mesh founders, who happen to be ex-SpaceX employees who helped build the laser-based inter-satellite links that connect Starlink's constellation.

Again, in typical Musk fashion, rather than simply buying the hardware, he is moving to acquire the entire company, gaining full control of its R&D and supply chain. Should the deal — which is all but done — go through, Musk will own the full stack of critical infrastructure needed to power the future of the AI industry.

Smart money is flowing to optical interconnects

The SpaceX ecosystem is just one of many entities that recognize the immense technical and economic importance of optical networking in AI. AI chipmakers are actively investing in the optical supply chain to secure manufacturing capacity and prevent hardware bottlenecks.

Nvidia alone has committed a reported $4 billion across component makers Coherent and Lumentum to lock up supply. Elsewhere, several hyperscalers, including Microsoft, Meta, and OpenAI, have teamed up with hardware giants Broadcom, AMD, and Nvidia to establish an Optical Compute Interconnect (OCI) Multi-Source Agreement (MSA) group, with the goal of developing protocol-agnostic scale-up interconnection technology for AI clusters.

To counter chipmakers' dominance, entities such as Japan's NTT established the $500 million IOWN (Innovative Optical and Wireless Network) Fund. This fund explicitly targets the creation of an open photonic ecosystem to accelerate the global transition from copper to light-based AI clusters.

Then there are the smart-money moves by investors, as well as the rising balance sheets of companies. Lumentum stock reportedly soared 339% in 2025 and delivered an additional 135.4% return in the first five months of 2026 alone, while Fabrinet, Cisco, and Coherent all recorded significant revenue surges attributable to optical hardware sales, meaning that Musk's move to acquire Mesh is extremely prescient, given Terafab's ambition.

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 Chinese government quietly approved the Space Computing Industry Innovation Center in early June, which aims to bring together rocket and satellite manufacturers, semiconductor fabs, and AI tech companies to build a space computing network. According to the Beijing government, this aims to “connect the entire industrial chain of space computing and boost the development of the satellite Internet of Things (IoT) sector. Research firm SemiAnalysis said on X that China made this move a week before Elon Musk announced his AI1 satellite, which will run AI workloads while orbiting in space.

The center is set to be officially launched later this month, and it will focus on six major research areas: highly reliable, heat-resistant space-native computing chips, high-performance hyper-interconnected space computing payloads, space computing satellite platforms and standard systems, space-based large models under constrained power conditions, integrated space-ground cloud-based measurement and control networking, and space computing power service-oriented and tokenized operations. These are designed to build an orbiting AI data center that will not rely on Earth-bound energy sources and will avoid the bottlenecks that many ground-based data center developments face today.

While Beijing made this quiet announcement earlier than Musk’s AI1 reveal, we should note that Elon already had technical details available during the interview. In fact, the world’s first trillionaire has been talking about compute in space since November last year and filed for a one-million-satellite Orbital Data Center System with the FCC in February 2026. Jeff Bezos is also getting into the game with the 51,600-satellite Project Sunrise set to operate in a sun-synchronous orbit.

What makes China’s announcement different, though, is that it’s making multiple companies work together to build a system. On the other hand, SpaceX and Blue Origin are going at it alone — the two companies are competitors, and it doesn’t look like they’re cooperating with each other to develop the technologies required for space-based AI compute. It even seems that the former is intent on vertical integration, with its new 11-million-square-foot (around 190 to 200 football fields) Gigasat factory and Musk’s TeraFab megaproject.

We’re unsure which technique will be more effective in the long run: having one or two companies pour massive resources into this megaproject (with its success or failure being solely borne by those firms) or making several smaller companies and institutions work together to build a Space Intelligent Computing Research Institute, with the output presumably available for use by Chinese firms. But one thing is certain: Beijing is taking space-based data centers seriously enough to pour resources into them — a significant move for a nation with ample excess electricity and available infrastructure to build power-hungry data centers.

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.

Elon Musk has laid out the first detailed design of SpaceX's AI1 satellite in a 30-minute video posted to the company's X account, the opening generation of an orbital craft SpaceX wants to build by the million to run AI workloads off Earth's power grid. Carrying a 150 kW peak compute payload across a 70-meter deployed wingspan, the spacecraft uses an interchangeable hardware design that lets different chipmakers supply the processors. The timing of this announcement is no accident, coming just three days before SpaceX’s IPO, which is set to price on June 11th and trade on June 12th at a target valuation near $1.75 trillion.

In his announcement, Musk pegged the satellite’s compute payload at roughly the draw of a single Nvidia GB300 rack, which pulls around 140 kW on the ground. One AI1, in those terms, is about one rack in orbit. In terms of its overall specs, SpaceX disclosed an average compute payload of 120 kW, a peak of 150 kW, and a density of 70 kW per ton, with the craft operating at roughly 600 km.

A satellite with these specs comes with some serious space requirements, and its 70-meter deployed wingspan edges past the 68.4-meter span of a Boeing 747-8. As for the interchangeable compute, that leaves the platform open to whichever vendor ships the most competitive AI silicon, rather than locking it to a single supplier.

This interchangeability is no doubt important to Musk, not least because SpaceX can’t yet guarantee its own supply of chips. The company is currently building Terafab, a chip fab that’s running as a joint venture with Tesla, while its S-1 IPO filing warns it can’t currently secure enough chips.

That aside, the elephant in the room is cooling: a rack on Earth sheds heat into moving air and circulating water, neither of which exists in a vacuum, where the only viable route is radiating it away as infrared. AI1 features up to 110 m² of deployable liquid radiators, as well as redundant pumping loops and integrated micrometeroid shielding. By comparison, the International Space Station’s ETACS rejects roughly 70 kW of heat — around half of what’s needed to cool a 140 kW GB300 rack — across 422 m² of radiator at a cost of up to $500 million, according to SemiAnalysis.

Musk has previously waved off potential thermal critiques, telling SpaceNews back in March that it's "safe to say SpaceX knows how to do heat rejection in space" and pointing to the company's fleet of more than 10,000 Starlink satellites.

SpaceX filed with the FCC in January to launch up to a million orbital data center satellites and has already signed compute deals, including a $920 million-per-month agreement with Google. The model has prominent skeptics: OpenAI's Sam Altman called orbital data centers "ridiculous" earlier this year.

SpaceX has announced a new 11-million-square-foot Gigasat factory in Bastrop, Texas, dedicated to building the infrastructure needed to achieve the company's orbital data center goal. In an internally conducted interview posted on X on June 8, SpaceX CEO Elon Musk confirmed that the facility is expected to begin producing complete "AI satellites" by 2027. The company is targeting 1 gigawatt (GW) of orbital AI compute capacity by the end of next year, with plans to scale that figure by an order of magnitude annually thereafter.

During the interview, Musk unveiled the build and specs of the proposed AI1 satellite, the company’s take on orbital data centers. The satellite will span roughly 70 meters (229.66 feet), with a massive solar array — generating power at a density of 250 W/m² — making up the bulk of the structure. AI1 will also feature vertically oriented, double-sided radiators for cooling, with the 150-kilowatt (kW) peak compute payload positioned right in the middle of the structure.

The massive 11-million-square-foot Gigasat facility — more than ten times larger than Starfactory, SpaceX's current largest spacecraft manufacturing complex — will vertically integrate much of the AI1 satellite supply chain on a single campus. The facility will manufacture solar ingots and wafers, solar cells, printed circuit boards (PCBs), silicon-based electronic components, user terminals, gateways, and the AI1 satellites themselves. The 1,000-acre site will also include dedicated satellite development and testing facilities, warehousing and logistics infrastructure, and a large-scale AI satellite production line.

According to Musk, the solar manufacturing facilities are already under construction, while the AI satellite production building is about to break ground. SpaceX expects to be producing a "reasonable volume" of these orbital data centers by the end of 2027. While each satellite will carry 150 kW of compute power, the company aims to achieve 1 GW per year of space AI compute in that same time. That would mean launching on the order of more than 6,000 AI1 satellites in a single year. For context, Starlink has about 10,500 active satellites as of June 2026.

Musk hopes to scale to 100 GW per year by 2030 and even has eyes on Terrawatt-level computing, completely solar-powered in space. "This is what we are going to try to do and think we probably can do, which is to get to roughly an annualized rate of a gigawatt per year by the end of next year at Space AI compute. And then aspirationally, scale that by an order of magnitude per year. So in two and a half years, hitting an annualized rate of 10 gigawatts a year at Space. And three and a half years, maybe 100 gigawatts," he said, while also expressing a desire to one day scale to a terawatt per year, depending on progress in chip making.

The largest AI data center anyone has actually announced is Meta's Hyperion in Louisiana, designed to scale up to 5 GW and house roughly 2 million GPUs at full buildout, at over $100 billion — and even that only reaches its first 2 GW phase by 2030. xAI's own Colossus 2 in Memphis just expanded to nearly 2 GW of capacity, with 555,000 GPUs, for about $18 billion, making it the world's largest single-site AI installation. So 100 GW/year is approximately 20 Hyperions or 50 Colossus-2s per year.

Achieving this will take an unprecedented volume of chips, and Musk's answer is Terafab — a SpaceX/Tesla/xAI venture in Austin aiming to fab 1 terawatt of compute a year, roughly 100 to 200 million advanced chips, in a 100-million-square-foot plant. However, the project itself is being met with widespread skepticism. For starters, none of the three companies has ever made a chip, yet they're starting at the bleeding-edge 2nm node. Our in-depth Terafab industry analysis explores various other reasons why this skepticism is valid.

Multi-Gigawatt ambitions aside, Gigastat is a significant step towards orbital data center goals that are increasingly seen as a potential viable solution to the extreme power consumption of on-ground data centers. Not unexpectedly, SpaceX is leading the race. Manufacturing solar arrays, wiring, and satellite bodies at volume is largely conventional work — much of it built on technology SpaceX already produces for its Starlink V3 satellites, just at far greater scale — which is part of why a 2027 production start is credible.

ASML has confirmed that a group of disgruntled workers is pushing back hard against an invitation for Elon Musk to address the equipment maker’s closed annual tech conference in Den Bosch this Thursday. According to Dutch publication Eindhovens Dagblad (via DutchNews), employees have been airing their grievances on the company’s internal comms platform, with several threatening to skip his appearance altogether because of his involvement in U.S. politics and what staff described as "Nazi sympathies."

If the invitation is accepted, Musk could join CEO Christophe Fouquet by video link to discuss AI, robotics, space exploration, and semiconductor production, with the conversation centered around Terafab, the SpaceX and Tesla chip venture whose proposed Texas fab can’t reach production without ASML's machines.

Backlash began to surface on Viva Engage, ASML’s internal employee platform, where workers have argued that giving the controversial tech leader a platform clashes with the company’s stated commitment to an inclusive culture. In particular, ASML staff have specifically highlighted Musk’s political activities around the Trump administration, general anti-European rhetoric, and what staff described as “Nazi sympathies.” In March, an Amsterdam court banned the Grok image-generation app due to its abusive “undressing” image generation capabilities.

ASML is the sole monopoly supplier of EUV lithography systems, the tools needed to pattern transistors at the leading edge; fabs targeting advanced nodes have no viable alternative.

Every major foundry and memory maker, including TSMC, Samsung, SK hynix, Micron, and Intel, already holds orders with the company. Terafab, by contrast, has no path to leading-edge production that bypasses ASML, and the company’s CEO said last month that Musk is "very serious" about the project, noting that ventures like Terafab and Starlink would strain equipment makers' capacity over the coming years.

ASML hasn’t addressed the internal revolt directly, saying only to employees via Viva Engage that it’s committed to a workplace where everyone feels respected and is free to express their opinion.

Musk announced Terafab in March as a SpaceX and Tesla joint venture with an initial $20 billion budget, targeting 2nm production and a terawatt of annual compute for AI, robotics, and orbital data centers. SpaceX has since filed a property tax abatement application in Grimes County, Texas, for a fab costing $55 billion in its initial phases and up to $119 billion if all expansions proceed, with Intel set to contribute its 14A process technology. Outside estimates run higher, with Bernstein analysts arguing the full terawatt project would require roughly 358 fabs and around $5 trillion.

“ASML regularly invites technology leaders from the industry to share their insights. The invitation to Elon Musk took place in the context of the Terafab project, which is highly relevant to the semiconductor industry,” the company said.

SpaceX has won a 100% property tax abatement from Grimes County, Texas, for its proposed, massive $55 billion Terafab semiconductor chip manufacturing facility. According to a local news outlet, KBTX, Grimes County commissioners voted 4-1 on Wednesday, June 4th, to approve both a reinvestment zone designation and the tax abatement, which will fully exempt SpaceX from property taxes tied to the project in exchange for a regular payment instead.

In exchange for the tax exemption, SpaceX will instead remit a lump sum of $10 million to the county, followed by $20 million annually for 35 years. The vote — centered on a three-part proposal for the SpaceX Terafab facility, consisting of an obligation framework for jobs and infrastructure, a Tax Increment Reinvestment Zone designation, and a 35-year PILOT tax abatement agreement — followed a hearing that drew in more than 100 residents to the Grimes County courthouse.

Proponents of the proposals, including local officials, argued that the project promises massive economic investment, job creation, and national security benefits. Conversely, residents opposed to the project cited a lack of transparency in the approval process, potential strain on water and power infrastructure, and the destruction of rural identity. Commissioner David Tullos, who cast the lone “No” in the vote, questioned SpaceX’s absence in earlier meetings, saying that the hearing was the first time it had sent a representative since the project was announced. Tullos also questioned the company about the size of the proposed reinvestment zone and the company's plans for portions of the land included within it.

Several residents reportedly expressed deep concern about environmental impacts and land loss amid tears. In response, John Federspiel, senior director of Starlink Product Engineering at SpaceX, said: "We recognize that large projects bring legitimate questions about infrastructure and environmental stewardship. Our company is committed to proactively ‌addressing those ⁠concerns and taking care of them responsibly.”

Despite an apparent majority opposition from residents who urged the commissioners to delay the vote on the tax abatement — which public records show SpaceX applied for just last month — all proposals passed. A number of local residents expressed disappointment in the outcome, with many insisting that a 100% tax rebate — for a company about to launch the largest IPO in history with an expected $1.75 Trillion valuation — was too far.

Defending the arrangement in an X post on Wednesday, SpaceX owner and world's richest man, Elon Musk, said that the exemption was necessary to remain competitive with other chip manufacturers. “The reason SpaceX asked for this, which is standard practice for massive capital investments, is because Terafab will have a large number of extremely expensive machines for making chips. Property tax on these crazy money machines would put us at a serious competitive disadvantage relative to other chip fabs in the world,” Musk said.

He also implied that the county stands to significantly benefit from the project. “Something that perhaps isn’t clear is that, if this location works out (other locations are still in the running), SpaceX will still be paying an annual amount that increases tax revenue for Grimes County by ~25% and will be by far the biggest source of revenue for the county,” Musk argued.

SpaceX is advancing the Terafab project as part of its plans to expand beyond aerospace into computing infrastructure and domestic chip production, a strategic pivot investors see as vital for future growth. The company already owns Colossus 1, a massive data center it rented out to Anthropic, and is currently building Colossus 2.

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

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

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

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

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

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

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

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

Last week, Anthropic announced that it had struck a deal with SpaceX to lease all of the latter's Colossus 1 data center, with over 220,000 GPUs and 300 megawatts of compute capacity. The deal immediately raises questions, foremost among them: why would Musk lease one of xAI’s most aggressively hyped AI assets to a direct rival? With SpaceX's IPO just around the corner, a related strategy appears to be at play, but it also turns out that the system's mixed architecture with different types of GPUs may be a key reason Musk has decided to lease the system.

Anthropic says the newly acquired capacity will primarily be used to ease long-standing usage bottlenecks across Claude’s paid ecosystem. According to the company, the additional compute will enable significantly higher Claude Code limits, the removal of peak-hour throttling for Pro and Max subscribers, and substantially increased API request limits for Claude Opus models used by developers and enterprise customers.

The seemingly unlikely partnership — a complete turnaround of Musk's earlier stance on Anthropic — also reveals Anthropic is straining under the Claude ecosystem’s compute demands. The company says it needs the entire 300 MW AI supercluster just to improve the experience of using Claude.

Anthropic appears to have hit the compute wall

The earliest signs that Anthropic was struggling to keep up with the computing demands of its growing user base were the increasingly aggressive usage limits placed across Claude’s services. Free users frequently complained about rapidly exhausting tokens — the units Claude assigns for processing tasks. However, the restrictions extended beyond the free tier. Paid Pro, Max, Team, and Enterprise users also regularly encountered message caps, peak-hour throttling, API rate limits, and strict time-based usage ceilings on Claude Code sessions, particularly during periods of heavy demand.

It was clear that Anthropic was running out of inference capacity. While training an AI model is an expensive, one-time computational undertaking, serving that model to millions of users simultaneously creates a continuous, round-the-clock demand for compute that scales directly with every new user and every new query. The apparent solution is to build more data centers, which Anthropic is apparently pursuing via massive gigawatt deals with Amazon, Google, Microsoft, and Nvidia.

However, modern hyperscale AI data centers can cost tens of billions of dollars and take years to build. Utilities are increasingly struggling to supply sufficient electricity for AI projects, while land, transformers, cooling infrastructure, and high-end GPUs themselves remain constrained. There is also growing sentiment against AI infrastructure from local communities. We recently reported that a U.S. senator got physical with a reporter after a confrontation on a data center issue.

Anthropic's compute capacity problem was immediate and urgent, but the solution was significantly long-term. If only there were a massive AI supercluster with hundreds of megawatts of compute power just sitting there. Turns out there was: SpaceXAI’s Colossus 1. Following the deal, Colossus 1’s entire computing power now belongs to Anthropic — for now.

Musk, xAI, SpaceX, and an upcoming IPO

When Musk unveiled Colossus, it was framed as one of the clearest signs that xAI intended to compete seriously with OpenAI, Anthropic, and Google at the AI frontier. The Memphis-based cluster became famous for how quickly it was assembled. Tens of thousands of Nvidia GPUs were reportedly brought online in record time, eventually scaling to over 220,000 accelerators. Musk repeatedly boasted about xAI’s future compute ambitions, including plans to expand toward million-GPU-class systems through Colossus 2.

So why does he seem to have wrapped the whole thing in a neat little bow and handed it over to Anthropic, xAI's rival? One possible answer is utilization. Reports suggest that Colossus 1 may have had more available capacity than Grok’s current user base required. However, according to a detailed report by Mirae Asset Securities — a major South Korean investment bank — the bigger utilization issue was architectural. Colossus 1 is a heterogeneous cluster, mixing roughly 150,000 H100s, 50,000 H200s, and 20,000 GB200s — three different generations of Nvidia silicon running under one roof. This was largely a byproduct of how fast xAI assembled the cluster, with different GPU generations coming online as supply allowed, rather than a deliberate design choice.

For AI training, the heterogeneous configuration creates a significant efficiency problem. Distributed training requires every GPU in the cluster to complete each computational step simultaneously before the system can advance. When the faster GB200 chips complete their work first, the entire cluster waits for the slower H100s to catch up — a well-known bottleneck known as the straggler effect. At 220,000 chips, this effect is exponential.

As a result of these issues, xAI's real-world GPU utilization reportedly sat at just 11% — meaning 89% of the cluster's theoretical computing power was going to waste. For context, Meta and Google typically operate at 40% or above.

AI GPUs are not static assets that quietly sit on shelves, gaining value over time. They depreciate rapidly, consume enormous amounts of electricity, and require expensive maintenance and cooling infrastructure. Unused GPUs are effectively burning money.

From that perspective, Anthropic may have arrived at exactly the right moment. The company had exploding demand and an urgent need for ready-made compute, while SpaceX/xAI had a gigantic, not-so-great first-generation AI cluster. For Anthropic, however, the same cluster looked quite different. The company needed compute power for Inference — running queries through an already-trained model, which does not require the tight synchronization that training workloads demand. So, what was a structural inefficiency for xAI's training workloads is a workable infrastructure for Anthropic's inference needs.

Multiple reports suggest xAI is now heavily focused on Colossus 2, a far larger next-generation cluster reportedly aimed at gigawatt-scale AI infrastructure. Unlike Colossus 1's chaotic mix of chip generations, Colossus 2 is built entirely on Nvidia's Blackwell architecture — a homogeneous cluster where every GPU is identical. In a uniform cluster, every chip completes each training step at roughly the same time, allowing GPU utilization to theoretically surpass the range in which Meta and Google currently operate. xAI can also properly optimize its software stack for a single hardware generation rather than trying to serve three simultaneously.

According to the Mirae Asset report, xAI has already moved its core training workloads entirely onto Colossus 2, effectively treating Colossus 1 as a retired first-generation asset. In other words, Colossus 1 may have transitioned from "cutting-edge frontier training weapon" into a monetizable first-generation compute asset, while Musk continues to build towards xAI’s “takeover” with Colossus 2.

Musk has long treated his companies less like isolated entities and more like interconnected pieces of a broader ecosystem. Tesla technologies appear across SpaceX projects. SpaceX infrastructure supports xAI ambitions. xAI products increasingly feed into Musk’s wider platform strategy.

The deal also hints at another possibility: Musk could be positioning SpaceX/xAI as more of an AI cloud infrastructure provider. That would not be entirely surprising. xAI has already launched Grok Business and enterprise-focused offerings featuring APIs, security controls, audit logging, and corporate integrations. This also aligns with Musk’s reported plans for broader structural changes at SpaceX and xAI ahead of the company's upcoming IPO.

Earlier this year, Musk publicly attacked Anthropic and Claude, calling the company “misanthropic and evil.” Yet this week, he claimed he approved the deal after speaking with Anthropic executives and determining that “no one set off my evil detector.”

Mirae Asset’s analysts attempted to estimate the value of the Anthropic deal, using estimated hourly lease rates for different Nvidia GPU types. The analysts projected that Colossus 1 could theoretically generate roughly $5–6 billion in annual revenue. That nearly perfectly offsets xAI's annualized net loss of approximately $6 billion as of Q1 2026, effectively pulling the company to breakeven in a single contract.

For Anthropic, the analysts applied CEO Dario Amodei's own publicly stated estimate that roughly half of all AI industry compute spending goes toward inference, and that inference compute converts to revenue at a 3x multiplier. On that basis, the $5 billion being directed toward inference capacity could generate approximately $15 billion in incremental ARR — a significant addition to Anthropic's already rapidly growing revenue base.

Stellar ambition

Another critical aspect of the announcement involved “orbital AI compute capacity” — basically, data centers in space. Granted, it does sound like science fiction marketing language. But it directly ties into a core problem both companies, alongside several other AI giants, are increasingly facing: AI infrastructure is beginning to outgrow terrestrial constraints. So when a joint announcement comes from the world's largest AI company and the company that built the world’s largest reusable rocket system and operates thousands of active satellites in orbit, you best believe we may soon have data centers floating around in space.

Despite Mirae Asset’s analysis, the factual financial details of the Colossus deal are not publicly available. However, Anthropic recently raised $30 billion in a Series G funding round, valuing the company at $380 billion. It would not be too wild a guess to say some of that cash may have gone into funding the Colossus agreement. Then again, the company said last month that its annualized revenue run rate had already surpassed $30 billion, highlighting the staggering scale at which Claude’s business is now operating.

xAI built Colossus 1 fast — too fast, it turned out. The resulting mixed GPU architecture created structural training inefficiencies that made the cluster hard to justify as a long-term platform. With Colossus 2 now operational and built properly on uniform Blackwell hardware, Colossus 1 became a first-generation asset in search of a better use.

Anthropic, with explosive demand and not enough compute, provided exactly that. The deal converts what was effectively a depreciating liability into roughly $6 billion in annual revenue — enough to bring xAI close to breakeven. For Anthropic, the same compute could unlock an estimated $15 billion in additional ARR. Both companies got what they needed, and Musk gets a compelling infrastructure story heading into a potential IPO.

Anthropic announced in a press release on Wednesday that it has signed a deal with Elon Musk’s SpaceX to use the company's massive Colossus 1 data center. According to a corresponding announcement from SpaceXAI, the deal will give Anthropic access to all of the massive supercomputer. That's over 222,000 Nvidia GPUs — including powerful H100 and H200 chips alongside next-generation GB200 accelerator systems — and 300 megawatts plus of compute power.

Anthropic says this additional capacity will go toward improving the experience for paid Claude users — Pro, Max, Team, and Enterprise subscribers — via three key changes. Effective yesterday, Claude Code’s five-hour rate limits have doubled for all paid tiers. The company has also removed the peak hours limit reduction for Pro and Max. Lastly, it has “considerably” raised the API rate limits, the volume of requests developers can make, for Claude Opus models.

The announcement highlighted other similar agreements with Amazon, Google, and Microsoft, all aimed at building gigawatts of additional capacity. Anthropic also expressed interest in partnering with SpaceX to “multiple gigawatts of orbital AI compute capacity, basically data centers in space, as, according to SpaceX, “The compute required to train and operate the next generation of these systems is outpacing what terrestrial power, land, and cooling can deliver on the timelines that matter.”

The Colossus 1 deal means that the whole first-generation cluster, originally built to power xAI’s own Grok models, is now powering one of its direct AI rivals, as the company focuses on building Colossus 2. In an X post, Musk said he gave the green light to lease Colossus 1 to Anthropic after spending time with senior members of the company to “understand what they do to ensure Claude is good for humanity.” He claimed he “was impressed,” saying, “No one set off my evil detector.” The statement is a stark reversal of comments Musk made earlier this year about Anthropic, when he called Claude “misanthropic and evil."

In the press release, Anthropic also signaled that its compute expansion strategy extends far beyond raw processing power. The company said future infrastructure deployments will increasingly target international regions, such as Europe and Asia, to meet growing enterprise demand for local data residency and regulatory compliance, particularly in sectors like healthcare, finance, and government. It also stressed that future capacity partnerships will prioritize politically stable, democratic countries with secure AI supply chains, while exploring ways to offset increases in electricity costs and reinvest in communities hosting the data centers.

In addition to the compute deal announcement, Anthropic also unveiled on Wednesday a new Claude feature called 'dreaming,' designed to help its AI agents improve themselves between sessions. The feature allows Claude-powered agents to review previous work, identify recurring patterns and mistakes, and reorganize memory files that contain user preferences and contextual information, enabling the system to refine its behavior over time rather than starting each session from scratch.

Leaked excerpts from SpaceX's confidentially filed $1.75 trillion S-1 form shed some light on the company's business plans, which includes plans for it to build its own GPUs. The listing notes its intent to invest billions in production of some of the processors it needs internally as it does not have long-term supply agreements with its silicon suppliers. Interestingly, the silicon to be produced in-house is said to be GPUs, not specialized ASICs for AI acceleration, according to Reuters. However, the naming convention is still up for debate. The news comes on the heels of yesterday's blockbuster announcement that Musk will use Intel's 14A process node in its new TeraFab chipmaking venture, with SpaceX managing the manufacturing facilities.

The S-1 form seen by Reuters mentions 'manufacturing our own GPUs' as one of the reasons of the 'substantial capital expenditures' the company will incur in the future as it appears to lack ' long-term contracts with many of our ​direct chip suppliers.' While the fact that SpaceX will build and operate its own high-volume semiconductor manufacturing facility to produce silicon developed at Tesla was confirmed by Elon Musk, chief executive of Tesla and SpaceX, last night, the firm's plans to build its 'own GPUs' seems like something new.

Reuters recons that different companies name their AI accelerators differently. While AMD and Nvidia explicitly call them 'GPUs,' Google calls its AI accelerators 'TPUs,' Microsoft calls its Maia devices 'accelerators,' while SambaNova calls its AI chips 'RDUs.' Most of hyperscale cloud service providers and independent hardware vendors (IHVs) call their AI accelerators application specific integrated circuits, or ASICs. Since SpaceX does not clearly mention AI ASICs developed by Tesla, Reuters believes that SpaceX intends to design and produce something they call their 'own GPUs,' which would mean something different from Tesla's AI-series processors.

We have no idea whether SpaceX — or its xAI division — has the capability to design its own sophisticated custom silicon that would compete against AI GPUs/accelerators from AMD, Nvidia, Rebellions, or SambaNova. However, Elon Musk himself referred to Tesla's AI5 processor as a 'GPU' even though this chip cannot process computer graphics (at least not in a traditional way) due to a lack of special-purpose hardware.

"With the AI5, we we deleted the legacy GPU, or the traditional GPU, which is in AI4, but AI5 does not have the legacy GPU because it basically is a GPU," Musk said in a earnings conference call. "We also deleted the image signal processor. There is a long list of delitions that are very important."

Given the inconsistent naming of its AI accelerators at Tesla — mostly because they are aimed at very specific workloads — it is hard to verify whether SpaceX means producing Tesla's AI5 and/or AI6 hardware, or indeed in-house designed 'GPUs.'

Since the S-1 form has been filed confidentially, Tom's Hardware could not verify its content.

People working on Elon Musk's Terafab project are said to be aggressively tapping suppliers for pricing and delivery times for photomasks, substrates, etchers, and more. According to Bloomberg, his people have reportedly talked with Applied Materials Inc., Tokyo Electron Ltd., Lam Research Corp., and even chip manufacturing partner Samsung Electronics Co., with one source saying they reached out to a company during a Friday holiday for an estimate to be delivered the following Monday because Musk wants to move at “light speed."

According to the report, representatives for the project have been asking for "speedy" price estimates, while being decidedly coy about specific products they're planning to build. The inquiries show Musk is serious about his dreams of building his own chips, which he started talking about in late 2025. The billionaire founder may be one of the richest people on earth, but Nvidia’s Jensen Huang warns that building a fab is a complex engineering problem that requires more than money to solve. “Building advanced chip manufacturing is extremely hard,” Huang said. “It is not just building the plant, but the engineering, the science, and the artistry of doing what TSMC does for a living is extremely hard.”

Still, this seemingly did not deter Elon, who successfully made electric vehicles mainstream with Tesla and commercialized space flight by building low-cost, reusable orbital rockets through SpaceX. It seems that semiconductor manufacturing is his next big project, especially as he criticized existing chip fabs for not having enough capacity to meet the needs of his companies.

The project officially launched in March 2026, less than five months after he first publicly talked about the idea. Musk injected $20 billion into Terafab to kickstart it, although experts estimate that costs could hit more than $5 trillion. Less than a month after the announcement, Intel said it joined the Terafab project to “help refactor silicon fab technology” and “accelerate Terafab’s aim to produce 1 TW/year of compute.” It’s currently unclear what Intel will do for Terafab, but this tie-in allowed the chipmaker, which was on the verge of collapse just two years ago, to hit its highest market cap in 25 years.

The other companies Terafab reached out to all similarly experienced uplifts in stock prices, with Tokyo Electron closing at +5.3% in Tokyo, while Applied Materials and Lam Research were up 2% in pre-market trading, despite no fixed orders being placed by the startup just yet. We also do not expect the company to start churning out chips in the next couple of years. “It takes two to three years to build a new plant — no shortcut — and it takes another one or two years to ramp it up,” TSMC CEO C.C. Wei said during an earnings call. “There are no shortcuts.”

Elon Musk on Wednesday showcased an image of one of the first samples of Tesla's AI5 hardware that will be used to drive AI applications in Tesla's cars, Optimus robots, and potentially xAI data centers. The AI5 processor is about half of the reticle size, uses industry-standard memory, and yet can be up to 40X faster than AI4 in certain scenarios, according to Elon Musk.

"Congrats to the Tesla_AI chip design team on taping out AI5," Elon Musk wrote in an X post. "AI6, Dojo 3 and other exciting chips in [the works]. […] And thank you to @TaiwanSemi_TSC and Samsung for your support in bringing this chip to production! It will be one of most produced AI chips ever."

The Tesla AI5 processor module features a fairly small ASIC die (about half the reticle size, according to Musk's previous comments) surrounded by 12 memory packages from SK hynix (most likely GDDR6/7). The module uses an organic substrate, and the memory packages are marked like industry-standard DRAM products. While we do not know how wide AI5's memory interface is, 12 memory packages clearly indicate that we are dealing with a fairly wide memory I/O. If we are indeed dealing with 12 GDDR6/7 memory ICs, then the Tesla AI5 ASIC has a 384-bit memory interface. Depending on the memory type used, the Tesla AI5 can offer memory bandwidth between 768 GB/s and 1.536 GB/s. Exact performance of the AI5 has not been disclosed, though Musk claims significant — up to 40X — improvements over AI5 in select cases.

"I think the Tesla chip team is really designing an incredible chip here: by some metrics the AI5 chip will be 40 times better than the AI4 chip," Elon Musk said during Tesla's Q3 2025 earnings call. "As a result of [outdated hardware] deletions, we can actually fit AI5 [on a] half [of a] reticle with good margin for the traces from the memory to the Tesla trip accelerators, the Arm CPU cores, and PCIe blocks."

Although Musk claims that the AI5 has just been 'taped out' (which means that the final chip design has been sent to a photomask house), he actually shows an already fabricated processor with a 'KR 2613' marking on it, which suggests that the ASIC was packaged on the 13^th week of 2026. Musk also mentions Taiwan Semiconductor (TSC) and Samsung for bringing the chip to production, though we are not sure that the producer of passive components has anything to do with bringing the AI5 processor to production. More likely, Musk meant Taiwan Semiconductor Manufacturing Co., better known as TSMC.

Previously, the head of Tesla, SpaceX, and xAI said that the AI5 would be made by both TSMC and Samsung Foundry, though we do not know which contract chipmaker fabbed the current sample. Assuming that Tesla got the chip in March or early April and no re-spin is required, it is reasonable to expect the company to deploy the processor sometimes in 2027.

Perhaps the most intriguing part of the announcement is that Tesla has apparently not given up on the Dojo system-on-wafer (SoW) processor for AI training, and the Dojo 3 processor is in the works. It was reported last August that the Dojo wafer-level processor initiative had been abandoned and the team behind it dismantled. Indeed, Peter Bannon, the head of the Dojo project at Tesla, retired last August, according to his LinkedIn page. Elon Musk said in July that the AI6 and Dojo 3 could feature a converged architecture (a converged ISA, we would speculate), which would enable the company to unify its software stack and could potentially allow the company to unify its hardware stack as well.

"I think about Dojo 3 and the AI6 as the first [converged architecture designs]," Musk said in a July 23 earnings call (via Investing.com). "It seems like intuitively, we want to try to find convergence there where it is basically the same chip that is used where we use, say, two of them in a car or an Optimus and maybe a larger number on a on a [server] board, a kind of 5 - 12 twelve on a board or something like that. […] That sort of seems like intuitively the sensible way to go."

In an unexpected turn of events, Intel on Tuesday said that it had joined Elon Musk's TeraFab project. The announcement mentions Intel's ability to develop, produce, and package advanced processors in high volumes, which could help Tesla, SpaceX, and xAI to get enough compute performance for next-generation AI and robotics applications. However, the announcement made in an X post you can expand below does not reveal how exactly Intel will help TeraFab.

"Intel is proud to join the Terafab project with SpaceX, xAI, and Tesla to help refactor silicon fab technology," a statement by Intel reads. "Our ability to design, fabricate, and package ultra-high-performance chips at scale will help accelerate Terafab's aim to produce 1 TW/year of compute to power future advances in AI and robotics. It was fun hosting Elon Musk at Intel this past weekend!"

The announcement is not accompanied by any press releases or SEC filings, which raises questions about the framework of the collaboration between Intel and TeraFab, as well as any possible legal bindings. In fact, the post in X is deliberately written in a way that barely reveals any concrete details about the structure of the partnership.

Officially, TeraFab is positioned as the "most epic chip-building effort ever" that is to combine "logic, memory and advanced packaging under one roof," which implies localized production in a massive facility. Furthermore, the company is hiring managers to build a greenfield semiconductor fabrication plant in Texas. By contrast, Intel's wording rather implies a virtual semiconductor production ecosystem, or even a consortium that involves chip design, manufacturing, and packaging at Intel and demand from Tesla, SpaceX, and xAI. How such a consortium would differ from a typical wafer supply agreement that large companies tend to have with their suppliers is something that is unclear at this point.

Considering the fact that Elon Musk wants his TeraFab project to ramp up production of chips as soon as possible, a reasonable way to achieve this would be to build a coordinated supply chain with a capacity pool that includes its own facilities as well as facilities belonging to other chipmakers, such as Intel. If we were to speculate further, we would envision custom co-invested fabs or at least product lines that would use unified process technologies, thus offering Elon Musk's companies dual or even triple sourcing to satisfy their demands.

In fact, Tesla, SpaceX, and xAI may even order Intel to develop custom silicon tailored specifically for their workloads, as Intel officially offers custom silicon development services, though this is speculation.

But while the Intel – TeraFab collaboration announcement lacks details on how Elon Musk plans to produce chips that would consume one terawatt of power on an annual basis, the chief executive of Intel expects Elon Musk to reimagine the entire semiconductor industry.

"Elon has a proven track record of reimagining entire industries," Lip-Bu Tan wrote in his own X post. "This is exactly what is needed in semiconductor manufacturing today. Terafab represents a step change in how silicon logic, memory and packaging will get built in the future. Intel is proud to be a partner and work closely with Elon on this highly strategic project."

Starlink has posted an update on satellite 34343, which it lost communication with on Sunday. The SpaceX-owned global internet firm says that the satellite “experienced an anomaly” while it was in LEO at around 560 km above the Earth. It also used wording to play down any concerns about risk to the thousands of other man-made objects at a similar altitude. However, orbital intelligence agency LeoLabs is less coy and describes Sunday’s incident as a “fragment creation event.”

Rapid Unplanned Disassembly

Following the anomaly on Sunday, Starlink’s official line (as above) was to play down any risks to the International Space Station, Monday’s Transporter-16 mission, or the upcoming launch of NASA’s Artemis II mission. It doesn’t go as far as admitting there is any post-rapid unplanned disassembly debris, but does say that it will “continue to monitor the satellite along with any trackable debris and coordinate with NASA and the U.S. Space Force.”

Naturally, SpaceX and Starlink teams are working to determine the cause of the anomaly and implement any changes necessary to prevent this from happening again. Sadly, the analysis of whatever happened towards the end of last year, when another Starlink satellite tumbled from space after an “anomaly,” didn’t stop this latest debris‑generation incident.

LeoLabs provides some further analysis regarding the fate of satellite 34343. Its radar system “immediately detected tens of objects in the vicinity of the satellite after the event.” Moreover, the independent space‑situational‑awareness (SSA) data provider reckons that the not-an-explosion was “likely caused by an internal energetic source rather than a collision with space debris” or another satellite.

On potential risks to other satellites and missions from the now-ex-satellite, LeoLabs doesn’t highlight any specific dangers. It reckons the fragments from the anomaly will probably de-orbit within a few weeks.

This was indeed a similar Starlink satellite incident to that which occurred last December, think the independent analysts. It goes on to recommend greater clarity to ensure safety in the operating environment.

SpaceX IPO

SpaceX is warming up Wall Street for what could be the largest IPO of all time. It reportedly wants to raise around $75 billion in its offer, which equates to a $1.75 trillion valuation.

It would be better for Elon Musk’s firm if fewer rapid unplanned disassembly incidents happened between now and the IPO, tipped to be this summer.

However, with FCC approval and plans for thousands more Starlink LEO satellites and even talk of a 1 million satellite Orbital Data Center System from SpaceX coming to light, one might assume that any unsolved anomalies become more frequent.

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.

British Army veteran Jon L. Noble has shared a heartwarming update on his first 100 days with a Neuralink brain computer interface (BCI) implant. Noble variously describes the journey from day 0 to the present as “science fiction… magic… brilliant… addictive… overwhelming and incredibly motivating.” Along the way, he’s moved from Mac newbie to power user. As for gaming, “I’m now raiding, and exploring Azeroth hands-free at full speed — no mouse, no keyboard, just intention,” enthuses the World of Warcraft fan, also known as P-18 (patient number 18).

For some background to Noble’s situation, the Financial Express reports that the ex-paratrooper was paralyzed from the shoulders down after a driving accident in 2016. He volunteered for Neuralink clinical trials and got the Neuralink N1 implant in London in December last year.

Noble’s first 100 days as an N1 implant patient have whizzed by, and he starts his recollective Tweet by summing up, “Looking back, the whole journey feels like science fiction that somehow became my everyday reality.”

On day zero, he went in for surgery, with a robotic system inserting 1,024 threads into his motor cortex. He was well enough - “alert and in good spirits” - to go home the next day. By day seven, the scar was already fading, and Noble says he “felt sharper and more positive than I had been in years.”

The fun started in week two, when the N1 was paired with a new Apple MacBook, says the ex-paratrooper. Neuralink engineers helped him get used to flexing his new BCI, and by week three, Noble says “scrolling, clicking, typing — all mind-controlled,” was second nature on the Mac. “I went from total Mac newbie to power-user faster than I ever expected,” he joked.

Riding into the realms of Azeroth

There’s a bit of a gap in the timeline shared, but by day 80, Patient 18 recalls he was “ready for the big leagues.” Of course, gaming presents quite a different experience to general computer interaction, with the stress on timing, accuracy, and not forgetting strategy. Noble admits he started off feeling clunky.

“Once my brain and the BCI synced, it was pure magic,” explained the ex-paratrooper. “I’m now raiding, and exploring Azeroth hands-free at full speed — no mouse, no keyboard, just intention. It’s honestly brilliant. The freedom is addictive.” You can see a video of Noble playing the game if you expand the embedded post above.

Noble also takes the opportunity to thank all the positive folks he’s had the pleasure to interact with on social media since his implant. It is good to hear of the overwhelming positivity and excitement communicated in the thousands of messages he has received.

With these first 100 days behind him, Noble relishes what the next 100 will bring. “The N1 didn’t just give me a new way to use a computer — it gave me a new way to live,” seems like a good summary of the life-changing experience gained from a capable BCI.

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.

Russia has at last begun to deliver on its plans to establish a domestic state-funded rival to Elon Musk’s Starlink, reports The Moscow Times. The launch of the first 16 satellites, said to form the foundation of a constellation of 900 by 2035, was supposed to have happened in Q4 last year. The so-called Rassvet system aims to expand to 250 LEO satellites next year, when it will become a domestic commercial alternative to Starlink.

A private aerospace outfit called Bureau 1440 appears to be the key tech partner of the government for this state venture. Scientists/engineers from this company are set to complete system checks on the newly launched 16 orbiters. When checks are finished, they will be moved into their target orbit positions.

So, 16 out of 900 might sound like a drop in the ocean of space, but this first step was probably one of the most difficult. Alexei Shelobkov, CEO of Bureau 1440’s parent company, ICS Holding, said that Rassvet’s deployment is going to see dozens more launches. A quick prod of the calculator suggests that just 15 more launches (with batches of 16 satellites) will be required to hit the 250 constellation target for sometime in 2027.

Did Starlink's new whitelist-only policy accelerate the rollout of Rassvet?

The Russian government has reportedly set aside the equivalent of $1.26b to help develop Rassvet. If you think it is good to see the Russian state investing in non-military infrastructure, remember that it coincidentally lost Starlink access only last month. That’s when SpaceX implemented its new whitelist-only policy to replace the previously ineffectual measures that were in place. Surely that would have put pressure on Rassvet development teams to get on with it.

Clearly, Rassvet, even at its fully fledged 900 satellite target, is going to be dwarfed by efforts from other countries, or groups of countries such as the EU. The U.S. is home to two of the biggest satellite internet firms: the aforementioned Starlink, with roughly 10,000 already in orbit, and a next-gen target beyond 30,000; plus Amazon Leo (the rebranded Project Kuiper) has over 200 satellites launched, but is targeting over 3,200.

Tesla and SpaceX CEO, Elon Musk, announced Saturday night that his TeraFab semiconductor project will be built on the Tesla campus in eastern Travis County, Austin, Texas, as a joint venture between the two companies.

In a livestream broadcast via X, Musk stated that the facility exists because the global chip industry cannot expand quickly enough to meet his projected demand across AI, robotics, and space computing. "That rate is much less than we'd like," Musk said from the defunct Seaholm Power Plant in downtown Austin. "We either build the TeraFab, or we don't have the chips, and we need the chips, so we build the TeraFab." The project reportedly carries a $20 billion price tag.

The Austin fab will house equipment for logic, memory, packaging, testing, and lithography mask production in a single building. Musk claimed that capability does not exist at any other facility in the world, and that having everything under one roof enables a rapid iteration loop: make a chip, test it, revise the mask, and repeat without shipping wafers between sites.

The facility is expected to produce two types of chips. One will be optimized for edge inference, primarily for Tesla's vehicles and Optimus humanoid robots. The other will be a higher-power chip hardened for the space environment, which Musk says will run hotter than “terrestrial” designs to minimize radiator mass on satellites.

Musk compared the project to the current global output of global AI compute, which he estimated at roughly 20 gigawatts per year. That figure, he said, represents about 2% of his companies’ eventual needs. On the terrestrial side, he projected 100 to 200 gigawatts per year of chip output; the remainder, up to a terawatt, would go to space-based AI compute aboard solar-powered satellites that SpaceX has already petitioned the FCC to launch.

“That's why I think it's probably a hundred to two hundred gigawatts a year of terrestrial chips, and probably on the order of a terawatt of chips in space," noted Musk. "Just because of power constraints on the ground.”

Musk said Tesla, SpaceX, and xAI — which SpaceX acquired in February — will continue buying chips from existing suppliers, including TSMC, Samsung, and Micron, adding that he would like them “to expand as quickly as they can.” He gave no timeline for when the TeraFab would begin producing chips or reach its target output, and while he has previously referenced 2nm as the target process node, he didn’t repeat that figure in the broadcast.

Elon Musk spent quite some time last fall complaining that existing foundries cannot meet his company's demand for high-performance AI processors and proposed an idea to build his own chipmaking venture. Apparently, this was not just a brag but rather an announcement of a long-term project. Now the project has gotten its launch date: March 21, 2026.

"Terafab Project launches in 7 days," Elon Musk wrote in an X post.

He did not elaborate on any details about the project, though his previous comments indicated that this is indeed a long-lasting semiconductor production facility project that would enable his companies — Tesla, SpaceX, and xAI — to get enough supply of AI accelerators.

Musk once mentioned that his companies might need 100 – 200 billion AI chips per year, and if it cannot get them from existing foundry partners, then the company will consider building its own fabs. Apparently, the Terafab Project seems to be the brand depicting the endeavor and scale, though it does not provide any reasonable portrayal of the nature of the project.

Speaking in an interview with Moonshots, Elon Musk argued that the semiconductor industry may be approaching cleanroom design incorrectly. Instead of keeping entire buildings ultra-clean, Musk suggested that fabs should focus on isolating silicon wafers themselves throughout the manufacturing flow, keeping them sealed from the surrounding environment at all times. He surmised that would allow him to eat cheesburgers in the cleanroom while chips were being made.

Rebuilding the whole supply chain for such fabs would take the industry a couple of decades, to say the least. For this, Musk argued that his planning horizon is closer to one to two years, and he rarely looks beyond three years, which makes the traditional semiconductor buildout cycle incompatible with his projected demand.

He added back then that if foundries could accelerate expansion and supply 100 billion to 200 billion AI chips per year within Tesla’s required timeframe, the company would gladly rely on external manufacturing instead of pursuing its own facilities.

Apparently, we are going to see details about the project in a week. The again who launces a multi-billion project on Saturday? This is what we are going to see next Saturday!

China could overtake the U.S. in brain computer interface (BCI) technology leadership in the coming months/years, thanks to a fortunate mix of government support and investor enthusiasm. In a recent interview with the Financial Times (FT), Tiger Tao, the founder of Shanghai-based NeuroXess, highlighted the rapid progress of the firm’s BCI technology. NeuroXess (no relation to Intel XeSS) was set up in 2021 and has already reported some successful human BCI trials.

The FT report comes less than six months after we covered news of China’s plans to outpace Neuralink with a state-backed BCI blitz, to coin a phrase.

At that time, a sweeping government policy document came into effect, with a fast, aggressive roadmap. It sought to coordinate and combine China’s huge resources and oil the wheels of progress by streamlining planning, regulatory approval, and financing. At the time of writing, there have already been 10 invasive BCI trial programs in China. The hope was/is to produce two to three ‘world-class’ BCI companies by 2030, and it looks like NeuroXess could be one of them.

As for NeuroXess’ success, it seems to have been rapid. As mentioned in the intro, it was only founded in 2021. Now just half a decade old, it has recently boasted of using its BCI tech to allow a paralyzed patient to control a computer cursor. Moreover, the patient was using their BCI for computing tasks just five days after the implant operation.

NeuroXess technology

NeuroXess trials have been using invasive BCI tech, and partly because of this, the focus has been on areas of strong medical need – like patients with paralysis or amyotrophic lateral sclerosis (ALS). No one without a strong medical need would be interested in a brain operation, and such a device being implanted.

The NeuroXess BCI uses a polyimide and metal mesh, but it sits on the patient’s brain and doesn’t pierce any brain tissue. This contrasts with Elon Musk’s Neuralink, which uses an implant bristling with microscopic threads to capture neural signals from within the brain. There have been concerns about electrodes going into brain tissue, causing scarring and thus the degradation of brain signal readings over time. Neuralink denies this is an issue, with its super-thin electrodes, butNeuroXess sidesteps the issue altogether.

Comparing Neuralink and NeuroXess in performance terms, the latest performance data shared by the FT suggests the former has achieved brain links at 10 bps in trials, and the Chinese tech is capable of 5.2 bps. You can see the NeuroXess speeds confirmed in the video linked in the intro.

While the race for the best invasive implant BCI continues, a BCI expert that the FT talked to noted that firms are using the latest knowledge from this frontier to the benefit of non-invasive systems. It is hoped that the advances could eventually improve non-invasive BCIs so much that operations/implants become minor or unnecessary, in general.

This optimization work is expected to be easier in China, with a virtuous circle coming into play: more data, lower costs, resulting in more users, and more data.

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.

A SpaceX FCC filing said that it plans to put a million satellites in orbit to build an Orbital Data Center system. The company said in the document that these will support AI, machine learning, and edge computing applications, taking advantage of the sun's energy without interference from the Earth’s atmosphere.

“Launching a constellation of a million satellites that operate as orbital data centers is a first step towards becoming a Kardashev II-level civilization — one that can harness the Sun’s full power — while supporting AI-driven application for billions of people today and ensuring humanity’s multi-planetary future amongst the stars,” the company wrote in its filing.

The filing went on to explain, “To deliver the compute capacity required for large-scale AI inference and data center applications serving billions of users globally, SpaceX aims to deploy a system of up to one million satellites to operate within narrow orbiting shells spanning up to 50 km (over 31 miles) each (leaving sufficient room to deconflict against other systems with comparable ambitions). This system will operate between 500 km and 2,000 km (310 miles and nearly 1,250 miles) altitude and 30 degrees and sun-synchronous orbit inclinations.”

SpaceX claimed that a million tonnes of satellites generating 100kW of compute per tonne would deliver 100 gigawatts of AI compute capacity without all the limitations of ground-based deployments, making orbiting data centers far more cost-efficient than their terrestrial counterparts.

The Kardashev scale measures the advancement of a particular civilization based on how it harvests energy. A Type I civilization uses all available energy on its own planet, which barely puts humanity and its current technology just below this level. On the other hand, a Type II civilization directly harvests energy from its nearby star, while Type III captures all the energy produced inside its galaxy.

The risk of Kessler Syndrome

While Elon Musk's plans to launch a million satellites into orbit come into view, a former Russian geostationary satellite has reportedly broken up in space. According to Space.com, the Luch/Olympic satellite, which the Russians use to observe other satellites in orbit, has recently been decommissioned and brought up to a graveyard orbit above its former geostationary altitude of more than 35,000 km or nearly 22,000 miles.

However, ground observers noticed the unit has fragmented, likely due to an external impact (see time-lapse video, embedded above). This incident produced more debris that is now orbiting the earth, which could collide with other satellites, further exacerbating the space junk problem.

Many experts are concerned about an event wherein multiple space collisions would produce so much debris that it would make it impossible to launch satellites or even keep them operating in orbit. SpaceX’s current fleet of 9,000 satellites already has experts concerned, especially as its competitors are also considering launching their own constellations.

Musk’s plan to launch a million satellites is likely a nightmare scenario for many scientists, as this would put the risk of a Kessler Syndrome event several magnitudes more plausible. Moreover, under such a cascade of space debris, humanity could effectively become trapped on Earth for generations, dashing the billionaire’s dream of landing astronauts on Mars.

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.

Nvidia tends to release its AI GPUs at an annual cadence, which keeps the company ahead of all rivals. AMD has invested a lot to keep up, so it also launches new AI accelerators on a yearly rhythm. Apparently, Elon Musk wants Tesla to progress even faster and release new AI processors every nine months to perhaps eventually catch up with AMD and then market leader Nvidia. There seems to be caveat with Musk's plans, but he seems to be looking forward a solution.

"Our AI5 chip design is almost done and AI6 is in early stages, but there will be AI7, AI8, AI9," Elon Musk posted on X. "Aiming for a 9-month design cycle. Join us to work on what I predict will be the highest volume AI chips in the world by far!"

Elon Musk's Tesla is not as prompt as AMD and Nvidia when it comes to releasing new hardware. There is an explanation for this: the company's processors are primarily meant for cars, which require redundancy and safety certifications. While redundancy is common for large high-performance AI processors that tend to be the maximum size possible (the reticle limit of an EUV lithography system), the safety required for cars is a whole different level.

Automation safety for automotive chips — particularly those used in advanced driver-assistance systems (ADAS) and autonomous driving — must comply with strict functional-safety requirements. The ISO 26262 standard serves as one of the governing specifications, but it is by far not the only one.

For advanced ADAS and automated driving (to a full-self drive degree), regulators increasingly require scenario-based testing (edge cases, failure modes), on-road testing permits (for higher automation levels), safety-of-the-intended-functionality, and cybersecurity compliance and software updates. After all, it goes without saying that developing a processor for a car is easier than building one for a data center.

Can the cycle be shortened, assuming that Tesla retains its processors to be both car- and data-center-bound? It seems to be feasible, but only with very strong constraints, and it will not look like a traditional 'clean-sheet' chip cycle. Let's unpack a bit.

A 9-month design cycle is realistic only if AI6, AI7, AI8, and AI9 are incremental, platform-based iterations, not clean-sheet designs. That means reusing the same core architecture, programming model, memory hierarchy, safety framework, and most IP, with changes limited to scaling compute, tuning SRAM, modest dataflow tweaks, or a planned node retarget. Any attempt to introduce something that goes beyond compute, such as a new memory type, compiler model, coherency scheme, or safety architecture, would immediately lengthen the schedule. On the competitive data center level dominated by Nvidia, these standards are redundant, though: performance and the software stack matter.

From a carmaker's point of view, automotive requirements make this cadence easier, not harder: long lifecycles, determinism, and ISO 26262 safety force designs towards very conservative evolution and locked interfaces. Given the overlapping development (multiple generations in flight), vertical integration, and a single internal customer, Tesla could sustain this cadence.

Meanwhile, the 'highest-volume AI chips' clearly suggest that we are dealing with processors meant for chips deployed across millions of vehicles, which is a far higher unit volume than data-center AI accelerators.

Assuming that Musk's Tesla has enough chip designers (which it probably does not, given that calls for applicants in the post), the real bottleneck for the assumed 9-month cycle will be verification, safety cases, and software stability, not silicon design itself.

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.

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

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

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

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

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

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

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

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

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

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

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

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

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

No to invasive implants, yes to AI operating systems

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

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

Distant horizons

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

Amid widespread anti-government protests, Iran has shut down all methods of internet access within its borders, reports IranWire. It has done the same previously, during earlier, milder episodes of revolt. However, this time the authorities seem to have also been successful in disrupting the tens of thousands of Starlink receivers thought to be inside the country. On Sunday, U.S. President Donald Trump said that he would speak to Elon Musk, the owner of SpaceX and Starlink, about restoring internet access to the Iranian public.

You're likely well aware of the severity of the clashes between protesters and the Iranian authorities. The general worldwide news media are filling the airwaves and front pages with reports about the uprising, and shocking numbers of casualties and arrests over the last four days.

During previous notable times of public unrest, Iran has cut public communications like internet services, mobile services, and even phone services. The commonly accepted reason for such blocks is to prevent easy communication and organizing by protesters. The same actions serve to limit the flow of images and videos to the outside world. Thus, the internet blockade helps the Iranian government shape the narrative, or at least gives it a better chance of doing so.

Starlink disruption likely relies on acquired Chinese or Russian tech

IW highlights something quite different during this latest spate of organized government dissent. On Friday, it noted that “even satellite internet services like Starlink sharply dropped.” Internet researcher Amir Rashidi told IranWire that, this time, “military-grade jamming signals were detected targeting Starlink satellites.” This directed action by the Iranian authorities meant that up to 80% of Starlink traffic was disrupted. Rashidi characterized this jamming effort as unprecedented yet highly sophisticated, and suspects Chinese or Russian tech has been acquired for this purpose, though doesn’t rule out something domestically developed.

Internal to Iran, government-aligned internet sites and channels reportedly started to come back online, thought to be through a progressive white-listing system. The first beneficiaries were “government-aligned Telegram channels, state media accounts, and some university networks,” notes IranWire. The last time there was a crackdown on the internet and digital communications it lasted 12 days. Perhaps this could be longer.

President Trump will talk to Elon Musk about Iran’s internet blockade

U.S. government officials have been quite vocal in condemning the reported killing of more than 500 protesters and the 10,000+ that have been imprisoned over the last four days.

On Sunday, U.S. President Donald Trump told reporters he has plans to talk with Elon Musk about restoring internet services in Iran. Trump was answering a question from a reporter about restoring internet services, like Starlink, for the Iranian public. “He's very good at that kind of thing, he's got a very good company,” the President assured concerned reporters.

Elon Musk this week said that the chipmaking industry builds cleanrooms in the wrong way. The head of Tesla and SpaceX promised that once Tesla builds its own 2nm-capable fab, he would eat a cheeseburger and smoke a cigar in that fab.

"I think they are getting clean rooms wrong in these modern fabs," said Elon Musk in an interview with Moonshots. "I am going to make a bet here, that Tesla will have a 2nm fab, and I can eat a cheeseburger and smoke a cigar in the fab."

When asked whether Musk had sketched Tesla's fab in his mind and whether he found a way to protect silicon wafers from cheeseburger grease, he responded that all wafers should be contained at all times. "They just maintain wafer isolation the entire time," Musk said.

A modern fab is a large, integrated manufacturing facility that encompasses ultra-clean production cleanrooms where wafers are processed; sub-fab levels housing vacuum pumps, gas handling, and exhaust systems, dedicated tool service corridors for maintenance and utilities; centralized chemical delivery and waste management infrastructure; as well as office and control areas used for administrative work, engineering, and monitoring.

Meanwhile, cleanrooms are essentially buildings within fab buildings as they are completely separated from other segments of the fab shell. Cleanliness in cleanrooms is specified by ISO Class standards that define the number of particles per cubic meter of air at different particle sizes. For example, an ISO Class 1 cleanroom allows at most 10 particles ≥0.1 µm per cubic meter and 2 larger particles, whereas an ISO Class 2 cleanroom allows up to 100 particles ≥0.1 µm per cubic meter and 38 larger particles. The most critical operations — such as EUV or DUV lithography exposures or advanced gate formation are performed in ISO Class 1 and 2 environments. By contrast, an ISO Class 3 cleanroom allows up to 1,000 particles ≥0.1 µm per cubic meter, which is still vastly cleaner than typical fab environments and common in advanced fabs for less sensitive operations.

Naturally, smoking or eating a burger in ISO Class 1 – 3 environments is absolutely not allowed, as it renders hundreds of millions or billions of particles. In fact, one human breath produces millions of particles along with moisture, not to mention organic contamination with bacteria or viruses. So, even if wafer and tool isolation is perfect, human breath (not to mention smoke or food particles) can affect the environment, which will affect ultra-sensitive EUV mirrors and perhaps fab chemistry.

Eating and smoking in other areas of the fab is also prohibited due to contamination control and safety requirements. In theory, Musk could eat a burger and smoke his cigar in office areas, which is possible even today. Still, smoking is banned in most office buildings anyway.

This is not the first time that Elon Musk has blasted the foundry industry. While he admired Tesla's partners TSMC and Samsung Foundry, in recent months, he has criticized the foundry industry for slow fab buildout and insufficient capacities, which is slowing down the development of Musk's xAI artificial intelligence initiatives. Musk once mentioned that at some point, Tesla could build its own semiconductor production facility, but given the extreme complexity of such an endeavor, this is something that is unlikely to happen. Furthermore, given the comments made by Musk, it does not look like he had an expert-level understanding of how modern leading-edge semiconductor fabs work.

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.

Ukraine’s Defence Express news has highlighted the continuing issue of Russian drones with access to Starlink satellite internet guidance. Despite assurances from U.S. Pentagon officials from over a year ago, drones from the Russian aggressors are still flying into Ukrainian airspace to strike their targets – with Starlink hardware clearly strapped to them, claims the source (machine translation).

The new report, on what seems to be an ongoing issue, has been sparked by one of the first sightings of a crashed/disabled ‘Molniya’ (Lightning) strike drone. This is shown in pictures originally shared by Serhiy ‘Flash’ Beskrestnov, a Ukrainian electronic warfare and communications expert, who is active on Telegram social media.

In the images, a Starlink Mini is clearly perched atop the grounded drone. ‘Flash’ doesn’t insist this drone actually came from Russia, but the overall design and internal components (like the battery) suggest it was very likely a Russian device.

Perhaps even more serious than this latest spotting is the Ukraine Defense Express assertion that Starlink usage by Russian drones hasn’t been totally stopped since it was first documented in 2024.

Later that same year, the US Defense Department’s assistant secretary for space policy, John Plumb, told Bloomberg that the Russian military’s unauthorized use of SpaceX’s Starlink internet had been blunted. An IMI report says that Plumb wasn’t specific about the measures put in place to deny Russian access, but that it was a “good solution” for Starlink and Ukraine.

Russian Shahed-136 drones have previously been seen kitted out with Starlink gear. The Ukraine Defense Express says that Russia’s recently unveiled RD-8 mother drone is Starlink controllable, too. Evidence of this wasn’t reproduced within the source story.

If the Ukraine Defense Express is correct, though, it is unsettling to know that the incessant Russian drone onslaught hitting Ukraine’s military, civilians, and key infrastructure could be weakened with just a little more political will.

A Chinese study has outlined how the nation could jam Starlink access across the entirety of the island of Taiwan. It would require around 1,000 to 2,000 specially adapted electronic warfare drones for this hostile act to pay off, reports the South China Morning Post (SCMP). The research, taken in tandem with recent news about China’s advanced internet cable-cutting capabilities, ratchets geopolitical and world semiconductor ecosystem tensions even higher.

Zhejiang University & Beijing Institute of Technology ran simulations to determine how the CCP-controlled People’s Liberation Army (PLA) could deny their democratic foe’s access to Starlink. Musk’s constellation of 10,000-plus satellites has been a source of consternation among CCP strategists ever since Ukraine effectively made use of it to resist the Russian invaders. Access to tech like Starlink is just one of the speed bumps that have made Putin’s 3-day "Special Operation" extend towards a grueling near-four-year campaign.

1,000 – 2,000 electronic warfare drones

According to the Chinese scientists, the complex, ever-changing satellite mesh networking coverage provided by Starlink could only be countered by a broad distributed jamming strategy. “Hundreds or thousands of small, synchronized jammers would need to be deployed across the sky – on drones, balloons or aircraft – forming an electromagnetic shield over the battlefield,” reports the SCMP.

To reach their unhappy-for-the-PLA conclusion, the scientists used actual Starlink data to create a simulated dynamic satellite mesh the size of Taiwan over 12 hours. A mix of wide and narrow-beam electronic noise-generating jammers featured in the test simulation. Airborne Chinese jammers, situated around 3 to 6 miles apart from each other, could form an effective 12-mile-high internet blocking mesh, it is now thought.

Under ideal conditions, a successful Chinese Starlink blockade would require 935 coordinated interference nodes, suggests the research. With cheaper, more practical, lower-power drones, the number of airborne interferers would have to be scaled up to approximately 2,000 drones.

Taiwan’s drone defenses

Of course, hostile blanket drone coverage wouldn’t exist unopposed in Taiwan’s skies. The home of computer and semiconductor giants like TSMC, Asus, and MediaTek has been investing in both foreign-bought and domestically produced drone and anti-drone military equipment.

The ambitious and industrious Silicon Island (and aspiring 'AI island') might even be considering its own Iron Dome-inspired protective network, no doubt further infuriating its neighbor.

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.

It's no secret that Elon Musk has tremendous ambitions when it comes to artificial intelligence, but apparently, they are so tremendous that he wants to get more AI processors than the industry produces, or even can produce. As it turns out, Tesla might need '100 – 200 billion AI chips per year' and if it cannot get them from existing foundry partners, then the company will consider building its own fabs, which Musk discussed several weeks ago. Now he has elaborated on those goals further.

"I have tremendous respect for TSMC and Samsung, we work with both TSMC and Samsung at Tesla and SpaceX. They are great companies and we want them to make our chip as quickly as they can and scale up to the highest possible volume that they are comfortable doing," said Elon Musk, during his conversation with Ron Baron. "But it doesn't appear to be fast enough. When I asked how long it would take from start to finish to build a new chip fab built, they said five years to get to production. Five years for me is eternity. My timelines are one year, two years. […] I cannot even see past three years. This is not going to be fast enough. If they change their minds and say, yeah, they are going to go faster and they want to provide us with 100 billion, 200 billion AI chips a year in the time frame that we need them, that is great."

Musk did not say when Tesla and SpaceX would require those 100 to 200 billion AI processors a year, but that number is pretty insane, assuming that he meant units, not dollars. To put it into context, the industry supplied 1.5 trillion semiconductor devices globally in 2023, according to the Semiconductor Industry Association. Yet, this number is a bit misleading, because the term 'chip' covers a wide variety of devices, ranging from tiny microcontrollers and sensors to memory chips and logic devices. Logic devices like Nvidia's H100 or B200/B300 AI GPUs are huge pieces of silicon that are hard to make, and thus take the longest lead and production times.

Musk recently said he believed power consumption for his AI5 AI processors could drop to as low as 250W. The power rating (TDP) of a chip can often be used as a decent relative proxy for the size of a chip, and by comparison, Nvidia's B200 GPUs can consume up to 1,200W, or nearly five times more power, thus implying that the AI5 will be a much smaller chip. Regardless, there absolutely isn't enough production capacity to meet Musk's targets, even if his chips are much smaller.

As one of the biggest clients of TSMC, Nvidia has supplied four million Hopper GPUs worth $100 billion (not counting China) throughout the active lifespan of the architecture, which was about two calendar years. With Blackwell, Nvidia has sold around six million GPUs, which equate to three million GPU packages, in the first four quarters of their lifespan.

If Musk indeed meant 200 billion units, then he would like to get orders of magnitude more AI processors than the industry (which is largely produced by TSMC) can build in a year. Yet, if he by any chance was referring to $100 - $200 billion worth of AI processors, then TSMC and Samsung Foundry could certainly produce that volume in the coming years. However, given that Musk is not satisfied with how quickly TSMC and Samsung build fabs, it looks like he indeed thinks he needs more than these companies can supply.

"We will be using TSMC fabs in Taiwan and Arizona, Samsung fabs in Korea and Texas," said Musk. "From their standpoint, they are moving like lightning. I am just saying that, nonetheless, it would be a limiting factor for us. They're going as fast as they can, but from their standpoint, it's 'pedal to the metal.' They just never had someone, a company, with our sense of urgency. It might just be that the only way to get to scale at the rate that we want to get to scale is to build up a real big fab, or be limited in output of Optimus and self-driving cars because of AI chip [supply]."

Whether Tesla and SpaceX really need 100–200 billion chips per year remains unclear. Tesla sold 1.79 million vehicles in 2024, so it does not need more than two million chips for its cars. Of course, the company might need millions more AI processors for its AI training efforts, though we have reasonable doubts that it can indeed build AI clusters powered by billions of GPUs any time soon. Also, while anthropomorphic Optimus robots, also powered by Tesla's AI hardware, could be a big market, it will take years to develop.

This free-to-access article was made possible by Tom's Hardware Premium, where you can find in-depth news analysis, features, and access to Bench. Subscribe now from just $7 per month for a limited time.

Elon Musk has confirmed that Tesla’s AI6 chip will be manufactured at both Samsung’s Taylor facility and TSMC’s Arizona fab. The update from the company’s CEO follows earlier comments in which he described Samsung’s site as technically superior — remarks that had suggested a one-fab strategy for Tesla’s most advanced silicon. That’s no longer the case.

According to Musk, both AI5 and AI6 are being built by Samsung and TSMC, with each foundry producing slightly different physical versions of the same design. “Our software works identically [on both chips],” he wrote, adding that the differences are a result of how each supplier translates designs to physical form.

Samsung has ‘slightly more advanced equipment’

In October, Musk said, rather plainly, that "Technically, the Samsung fab [in Texas] has slightly more advanced equipment than the TSMC fab [in Arizona],” referring to Samsung’s Taylor site and TSMC’s Fab 21 in Phoenix. According to Musk, Samsung’s tooling was slightly more advanced, which led him to suggest that AI5 would be produced there exclusively.

Naturally, that was never the full picture. Tesla had already lined up both Samsung and TSMC to manufacture AI5, and Musk’s own words now confirm it. While Samsung’s Taylor site is newer, TSMC brings proven process maturity and high-volume reliability. Tesla is using both. “Slightly different versions of the Tesla AI5 chip will be made at TSMC and Samsung simply because they translate designs to physical form differently, but the goal is that our AI software works identically.” Musk posted to X on November 4.

He also confirmed that Tesla’s AI6 chip will follow the same path, adding that it’ll “use the same fabs, but achieve roughly 2X performance.” This puts to rest earlier concerns that AI6 would require a different foundry altogether due to its more aggressive performance targets. Musk’s plan now is to stick with the same two suppliers and to maintain compatibility across both versions before moving to different fabs with the AI7.

Musk hasn’t detailed how the Samsung and TSMC variants differ at the hardware level, and the company is unlikely to highlight discrepancies unless they become relevant to functionality or deployment. In design terms, it seems like the chips will be tuned to abstract away fab-specific characteristics, enabling a consistent runtime environment for inference.

This isn’t unusual at the scale Tesla is targeting. Dual-sourcing is, after all, a standard practice among companies that need volume guarantees. What was unusual was Musk’s framing, which portrayed Samsung as having a clear lead and TSMC as the fallback. With the confirmation that both fabs are producing Tesla’s silicon, and that both will continue to do so for future generations, that’s clearly not the case.

Aggressive cadence, sparse detail

Aside from the more recent foundry confirmation, Musk also provided new performance figures. He claims that AI5 will be 40 times more performant than AI4. “Not 40 percent, 40 times,” he said during Tesla’s Q3 earnings call in October, offering no further explanation. The figure has not been tied to any specific metric. It could refer to throughput, latency, energy efficiency, model execution rate, or some weighted blend of those factors.

Tesla’s current silicon, often referred to as AI4 or Hardware 4 (HW4), already brought notable improvements to the FSD stack, including upgraded image signal processing and higher camera bandwidth. A 40 times gain over that platform (AI5), let alone another 40 times gain (AI6), would imply either a major architectural redesign or a substantial jump in how the chip interacts with Tesla’s AI workloads. But without specifications, it’s impossible to verify.

Musk says AI5 is still in development and has not yet taped out, and the chip is scheduled to enter production in 2026. AI6, which he describes as a “fast follow,” is set to arrive about a year later. The target is to double AI5’s performance using the same fabs.

Musk’s timeline would put AI6 in production sometime in 2027 or early 2028. He has not said whether it will be on a new node or what architectural changes are expected. But the commitment to a 2x gain over AI5, which itself is supposed to be 40x better than AI4, suggests that Tesla is aiming to push performance per watt and per dollar significantly higher in each cycle.

Demand setting the pace?

Tesla’s decision to split manufacturing between Samsung and TSMC reflects the scale of its ambitions more than any individual supplier’s advantage. AI5 and AI6 are expected to power FSD, Tesla’s humanoid robot platform, and potentially other unannounced systems.

That level of deployment requires guaranteed availability, and no single U.S. fab is likely to offer enough volume without risk. In practice, Tesla needs both fabs fully engaged just to hit expected production volumes. With AI5 not yet in tapeout and AI6 already planned, there’s little margin for delay.

This dual-source approach also gives Tesla flexibility. If one fab encounters yield issues, capacity limits, or node-specific challenges, the other can pick up the slack. It also gives the company better leverage in negotiating wafer supply and pricing, particularly as both fabs compete for high-value AI contracts.

Musk has suggested that AI7 is already in early development. He called it “more adventurous” and hinted that it might need a different fab arrangement. Whether that means moving to a more advanced node, adopting new packaging techniques, or changing foundry partners isn’t yet known.

For now, Tesla appears to be sticking to a system of iterating quickly, building for scale, and abstracting away from manufacturing differences that don’t affect performance. If the chips behave identically, Tesla doesn’t care where they come from. Ultimately, the company will use whatever capacity is available to ensure that chips keep rolling off the line.

Yesterday, OpenAI CEO Sam Altman took to X — Elon Musk's platform — to post three screenshots telling a story about how his 2018 order for the next-gen Tesla Roadster was mishandled. Altman emailed to cancel his reservation after several years of delays, asking for the $50,000 security deposit to be issued back to him, only to find out that the email address for reservations was seemingly no longer operational.

Musk, in a fiery reply, said "you stole a non-profit," following it up with another reply claiming how the order was refunded within 24 hours but that it's in Altman's "nature" to mislead. This exchange was symbolic of the world's richest man's general sentiment toward his former partner, stoking the fires of a bitter rivalry between these two AI bros once again.

Back in 2015, when OpenAI was founded as a non-profit, Elon Musk was one of the co-founders alongside Sam Altman, who now serves as the CEO of both OpenAI Foundation and OpenAI PBC — its recently restructured for-profit company. Musk famously left OpenAI in 2018 due to potential "future conflict of interest," with reports suggesting he was unhappy with the commercial direction OpenAI was treading in. After their breakup, Sam Altman and Elon Musk have always engaged in rivalry, one that has only intensified since Musk's xAI arrived on the scene.

For instance, Musk tried to block OpenAI from pivoting to a for-profit firm back when he was part of the board, suggesting a merger with Tesla who could fund the company's research. When the proposal didn't go through, he left, but the story didn't end there. Last year, the Tesla and xAI boss filed a lawsuit against OpenAI, citing his disdain for Altman and Brockman (another co-founder), saying they had "abandoned OpenAI’s founding mission," but there wasn't much legal ground to stand over, and it was later withdrawn.

In recent years, OpenAI has become the poster boy for the front-end of artificial intelligence, much like how Nvidia has for the back end. It wouldn't be an overstatement to say that ChatGPT ushered in one of the biggest technological revolutions of our lifetimes, sending many big players into a frenzy over how to compete. Companies like Google, who'd been previously known for AI or machine learning, were famously hesitant to deploy models to the public, but OpenAI changed that rhetoric almost overnight.

Elon Musk's xAI, the firm behind the controversial chatbot Grok that you find explaining basic concepts on X, has emerged as a key proponent in this race. Musk is building large data centers, buying up lots of AI GPUs, and planning to create his own in the future, all in a bid to compete with OpenAI (and others). There's a subtle irony in criticizing your peer for doing the exact same thing you're doing, but Musk would argue that there's a difference of intent.

OpenAI was founded to achieve AGI for the benefit of humanity, and it was turned into a profit-first company later on. OpenAI Foundation, the non-profit, still governs its commercial arm, but relies on its money to operate. It's a hybrid model that was recently lent clarification when Microsoft and OpenAI agreed to recapitalize the company, making it easier to secure funds and work more autonomously. Musk's statement, "you stole a non-profit" explicitly refers to that, contending that Altman has taken over OpenAI's original mission statement.

By contrast, xAI was always intended to make a profit. It's a standard organization developing models to be implemented inside X and commercialize through hyperscalers like Azure. There is no welfare plan in sight and Musk, therefore, has a safety net around his argument. Sam Altman never responded, but the spectators certainly got a show. Musk later replied with a laughing emoji to a screenshot of this exchange posted on X, implying a self-proclaimed moral victory over his AI brethren.

SpaceX is expected to receive $2 billion for President Donald Trump’s Golden Dome project, which sources say will include up to 600 satellites as part of an “air moving target indicator” system. According to the Wall Street Journal, this would represent another satellite project the Elon Musk-led company would have with the Pentagon, alongside a military communications network called Milnet, which uses SpaceX’s classified Starshield satellites and terminals, and a ground-tracking satellite system. The Pentagon and SpaceX did not comment on the news, especially as no official announcement has been made. However, funding for this project has already been approved as part of the “One Big Beautiful Bill” that Trump signed in July 2025, although it hasn’t yet been linked to a specific defense contractor.

A massive project

Golden Dome is an advanced missile shield that will use a sophisticated system of satellites, interceptors, command and control, and other related infrastructure. It’s expected to cost at least $175 billion, with some experts projecting total expenditures to be much higher. So, the $2 billion deal that SpaceX is set to receive would just be one giant drop in the even bigger bucket of the Department of Defense. We should also note that SpaceX likely won’t be the only company working on this project, as other startups and defense firms have pitched their products and services.

Some of these companies include new names such as Anduril Industries and Palantir Technologies, as well as giants such as Lockheed Martin, Northrop Grumman, and L3Harris. “What we’re relying on is industry to help us innovate by showing us the art of the possible—bringing ideas to us,” said U.S. Space Force Chief of Space Operations Gen. Chance Saltzman during an industry event in 2024.

Given the size of the project, neither the legislative nor the executive bodies are keen on relying on a single supplier like SpaceX. Sen. Rick Scott (R-FL) said he does not want to rely on a single company as Washington builds its missile defense umbrella. The Pentagon’s Defense Science Board called this monopoly ‘vendor lock’ and can “negate the strengths of the market by stifling innovation and inflating prices.”

Inspired by other missile defense systems

Trump’s missile defense project sounds similar to, and is likely inspired by, Israel’s Iron Dome interceptor system, though on a much larger scale. The latter has a reported 90% success rate and can intercept various threats, including missiles and mortars. However, there’s an even older project from the 1980s that is similar to the White House’s plan. Then U.S. President Ronald Reagan proposed the Strategic Defense Initiative (SDI) in 1983, intended to protect the United States against incoming ballistic missile threats from the Soviet Union using lasers and other systems. However, it’s been deemed that the required technology for it to run effectively was too advanced for its time, and with the dissolution of its main rival in 1991, the project fell into obscurity.

The advent of new technologies, like hypersonic missiles, meant that the U.S.’s current defenses aren’t equipped to handle this new threat. However, Golden Dome is more than just a new missile system — it’s an integrated defense network that would combine ground, sea, space, and, likely, even cyberspace systems, with SpaceX being just one of the cogs in this massive defensive machine.

During Tesla’s Q3 2025 earnings call, the firm’s CEO, Elon Musk, proposed that the cars take part in "a giant distributed inference fleet" to tap into their incredible compute power "if they are bored." Musk went on to estimate that, at some point, the advanced car fleet could summon "100 gigawatts of inference." Predictably, Musk’s latest musings have met with a mixed response on social media. So, let’s take a closer look at exactly what Musk said.

Expanding Tesla production and incentives to buy

During the Q&A session, Emmanuel Rosner from Wolfe Research asked about Musk’s intentions to expand the production of Tesla vehicles. He also queried what kind of incentives would be required to make such a production hike a reasonable business proposition.

Musk answered that an annualized production rate of three million vehicles should be achievable within 24 months. He added that the “single biggest expansion in production will be the Cyber Cab, which starts production in Q2 next year.” This will be a comfort-optimized automated transport vehicle, obviously targeting the cab market.

Killer app – allowing people to be lost in their smartphone screens, while driving

Beyond that project, the Tesla boss asserted that his team is looking closely at a killer app for new model cars with advanced processing. “If you tell someone, yes, the car is now so good, you can be on your phone and text the entire time while you're in the car. Anyone who can buy the car - will buy the car - end of story.”

Then, not for the first time, Musk heralded an “Autopilot safety game changer.” Elaborating on this, the Tesla CEO pledged, “I am 100% confident that we can solve unsupervised full self-driving at a safety level much greater than a human.”

Musk backed up his confidence by talking of the capabilities of the Tesla AI4 computer, also known as Hardware 4 (HW4). He indicated that, despite its muscle, the AI4 is already set to be eclipsed by the AI5, which outperforms it as much as 40-fold in tests. That sizable shot of extra performance boils down to facilitating autonomous driving systems that are 10x safer, it was suggested in the Q&A.

"A giant distributed inference fleet… [with] 100 gigawatts of inference"

Using talk of computing power as a springboard, Musk then openly pondered whether the upcoming systems “might almost be too much intelligence for a car.” To address the decidedly first-world problem of owning a car “that might get bored,” the Tesla CEO went off on an interesting tangent about tapping into idle car processing power, effectively turning the Tesla fleet into a giant distributed inference network.

“One of the things I thought: if we got all these cars that maybe are bored… we could actually have a giant distributed inference fleet,” Musk said.

Obviously, plucking numbers from the air, the Tesla boss went on to optimistically project that this fleet could expand to, say, 100 million vehicles, with a baseline of a kilowatt of inference capability per vehicle. “That's 100 gigawatts of inference distributed with power and cooling taken with cooling and power conversion taken care of,” Musk told the financial experts on the earnings call. “So that seems like a pretty significant asset.”

At its core, Musk’s idea beckons comparisons with classical distributed computing platforms, like SETI@home and Folding@home. But this Tesla fleet proposal could make an interesting commercial moon shoot idea for investors and business analysts.

Meanwhile, users will probably be more concerned about their bought and paid for vehicles being used for someone else’s advantage, perhaps using extra electricity, and their computer systems enduring longer heat stress, and so on. There’d probably have to be a clear benefit for end-users to incentivize them to sign up to such a compute power-sharing scheme.

Last week, Elon Musk announced that Tesla would make its next-generation AI5 processor in the U.S. and would double-source this system-on-chip from both TSMC and Samsung Foundry. This is down to Tesla aiming to have plenty of AI5 silicon available, since it will be used in a wide range of applications, including cars, robots, and, somewhat unusually, data centers. Musk also said that Samsung Foundry's fab in Taylor, Texas, is more advanced than TSMC's Fab 21 phase 1 in Arizona, which is a strange comment from a fabless chip designer.

We analyze what Samsung's Texas fab is capable of, how it compares to TSMC's fab 21 phase 1, and whether Musk's upcoming AI5 chip will really need the advanced capabilities that Samsung Foundry can offer.

More AI5 chips, more performance

Tesla's AI5 will be considerably more advanced than the AI4, according to Elon Musk. Since the company plans to use them broadly across its products, it wishes to maintain a surplus.

"I am confident [the AI5] will be winner, next level [of AI performance], so it makes sense to have both Samsung and TSMC focused on AI5," Musk said during an earnings call with investors and financial analysts this week.

Based on Musk's earlier comments, AI5 is a massive processor that can demand up to 800W (at peak). Without plenty of rudimentary IPs such as image signal processing or graphics processing, it leaves die space for low-precision tensor compute hardware to maximize performance and cost efficiency. Despite the high power consumption, the AI5 fits in 'a half reticle,' which equates to around 430mm^2, putting it in a happy middleground of not being too large, nor too small.

The exact configuration of Tesla's AI5 next-generation system-on-chip (assuming that it is a SoC, not a multi-chiplet SiP, or a series of products) is unknown; we only know it contains Arm CPU cores and plenty of proprietary Tesla hardware.

Due to this, it's difficult to ascertain just what lies under the hood of AI5, given its vast differences from AI4 in terms of power consumption (800W vs. 140W), lack of graphics processing and image processing capabilities, and the wide range of potential applications, which span from cars and robots to data centers.

"If we have too many AI5 chips for the cars and robots, we can always put them in the data center," Musk said. "So, we already use AI4 for training in our in our data center. So, [we] use a combination of AI4 and Nvidia hardware. So, we are not about to replace Nvidia, to be clear, but we do use both in combination, AI4 and Nvidia hardware. And the AI5 excess production, can always put in our data centers."

While Musk outlined vast volumes as one of the primary reasons behind the rather unusual decision of double sourcing a complex chip from two advanced foundries on two advanced nodes (which are yet to be determined), is perhaps another reason is to 'pipe clean' Samsung Foundry's American fab before assigning AI6 production exclusively (or maybe not exclusively) to SF, but Musk denied it.

Better equipment?

Musk also made an unusual comment regarding Samsung's Taylor, Texas fab tools, claiming that they are more advanced than TSMC's Fab 21 phase 1 in Arizona.

"Technically, the Samsung fab [in Texas] has slightly more advanced equipment than the TSMC fab [in Arizona]. These [AI5 chips] will be made in the U.S. […] by TSMC in Arizona, Samsung in Texas," Musk said.

Samsung's fab in Taylor, Texas, has been designed to produce chips on various manufacturing technologies, down to Samsung Foundry's SF2, which used to be called SF3P, but given its performance advantages over other SF3-series nodes, the company decided to bring it into the 2nm-class league. However, whether SF2 is competitive with Intel Foundry's 18A and TSMC's N2 remains to be seen, as other details, such as the number of EUV layers SF2 uses, are still a mystery.

While we do not know the exact configuration of Samsung Foundry's fab in Taylor, Texas, you can presume that it is tailored to make chips on 3nm-class process technologies, given its ramp timing in 2026 – 2027. So, we can expect Samsung to equip its fab for SF2P, SF2X, SF2A, and perhaps SF2Z (with backside power delivery), but this has not been officially confirmed.

Still, a fab capable of producing 3nm-class chips is indeed more advanced than TSMC's Fab 21 phase 1 in Arizona, which is designed to make chips on 5nm and 4nm-class production nodes. So, TSMC's fab 21 lags behind Samsung by design. Also, keep in mind that Samsung's leading-edge process technologies have relied on gate-all-around (GAA) transistors since SF3 in 2022, which require specialized tools to fabricate.

TSMC's Fab 21 phase 1 was designed and equipped between 2020 – 2024 (it likely follows TSMC's blueprints for its 1st-generation EUV-capable fabs). This is notably before ASML released its newer Twinscan NXE:3800E lithography systems, so the fab is equipped with Twinscan NXE:3600D machines, which can process up to 160 wafers per hour at a dose of 30 mJ/cm^2.

In contrast, Samsung Foundry's fab in Taylor is now being equipped, so if we were to speculate, we would imagine that Samsung equipped its fab in Taylor, Texas, with the more advanced Twinscan NXE:3800E, which processes up to 220 wafers per hour at 30 mJ/cm^2. Do bear in mind that ASML's scanners are upgradeable, so, theoretically, TSMC can boost the performance of the tools it uses if deemed necessary.

Because Samsung Foundry's fab in Taylor, Texas, is meant to produce chips relying on GAA transistors, it has more significant differences from TSMC's Fab 21 phase 1, which is designed to produce ICs with tried-and-true FinFET transistors.

GAA transistors introduce an entirely new level of complexity in both etching and deposition, because their architecture replaces a single vertical fin with multiple stacked horizontal nanosheets. Each sheet is separated by a temporary silicon-germanium (SiGe) layer that must be selectively etched away without damaging the adjacent silicon channels.

This requires atomic-layer etch (ALE) tools capable of removing material one or two angstroms at a time with extreme selectivity and directionality. These systems must also achieve near-perfect uniformity across 300mm wafers, since even slight over-etching can collapse or roughen the ultra-thin nanosheets. While TSMC hardly needs such tools at Fab 21 phase 1, Samsung does at its Taylor fab.

On the deposition front, conformal atomic-layer deposition (ALD) and selective epitaxy are key to device performance. The gate in a GAA device wraps fully around each nanosheet, requiring pinhole-free, uniform metal and dielectric films on all surfaces, even deep within narrow trenches. To produce them, chipmakers need tools that provide sub-angstrom control of high-k dielectrics (e.g., HfO₂) and metal gate layers (TiN, Ru, Mo) with near-perfect step coverage.

In addition, selective epitaxial deposition systems grow alternating Si/SiGe multilayers that form the nanosheet stack with atomic-scale thickness precision. Together, these advanced etch and deposition systems provide the precision and process stability necessary to manufacture GAA transistors reliably at the 3nm node and beyond. However, GAA transistors are not required for TSMC's N4 or N5 processes. TSMC is expected to debut GAA transistors in its 1.6nm-class 16A process.

Since the Samsung fab in Taylor, Texas, is at least one production node ahead of TSMC's Fab 21 phase 1, it features different metrology and inspection tools, too. Fab 21 phase 1 likely features a standard toolset from the late 2010s and early 2020s, so its in-line metrology/inspection stack is comprised of overlay/CD scatterometry, optical wafer inspection, and e-beam review tuned for 4nm and 5nm nodes. In contrast, Samsung's fab in Texas uses new generation tools, such as multi-angle scatterometry, X-ray, and e-beam or even multi e-beam metrology to ensure sub-angstrom control and uniformity across wafers.

But does Tesla AI5 need those advanced tools?

With the differences between the fabs out of the way, another question remains: Does Tesla's AI5 chip demand the advanced tools that Samsung Foundry's Taylor fab can provide? The simple answer is that we don't know, given the lack of specifics we have about AI5. The processor was expected to be released in early 2026, but then Tesla delayed it to late 2026. As Musk revealed last week, the hardware is still in development and has not taped out, so it is well over a year away from production.

"The AI5 chip designed by Tesla is, I think it is an amazing design, I have spent almost every weekend for the last few months with the chip design team working on AI5," said Musk. "I do not hand out praise easily, but I have to say that I think I think the Tensor chip team is really designing an incredible chip here. This is by some metrics, the AI5 chip will be 40 times better than the AI4 chip. Not 40%, 40 times. Because we have a detailed understanding of the entire software and hardware stack."

If Tesla tapes out AI5 this November, the earliest it can start mass production is December 2026. A more realistic timeframe is likely early 2027.

In any case, Samsung's Taylor fab is expected to start ramping up chip production sometime in 2026, so by late 2026 or early 2027, it will be ready for AI5. As for TSMC, the 3nm-capable Fab 21 phase 2 will only ramp in the second half of 2027. So, if Tesla plans to double-source AI5 from TSMC and Samsung Foundry starting in late 2026 or early 2027, it will likely rely on TSMC's N5A in Arizona and, presumably, the SF4A fabrication process in Texas.

In theory, Tesla could use SF2A, but we doubt it will be ready for large processors by early 2027, so SF4A is more likely, as it will be a mature automotive-grade fabrication technology by then. Furthermore, making the same chip on a 5nm-class process technology and Samsung's SF2 does not make much sense, as the final silicon will exhibit different power, thermal, and performance characteristics.

Faster production?

Given that Tesla's AI5 will almost certainly rely on FinFET transistors, it will make little use of the more advanced tools Samsung will have at its fab in Taylor, Texas, making Musk's comment largely irrelevant for the AI5 processor.

Perhaps, the main advantage of SF's fab for Tesla's AI5 is its Twinscan NXE:3800E lithography systems, which can process 40% more wafers per hour than their predecessors used at Fab 21 phase 1. This can potentially shrink cycle time and/or increase output. However, the actual cycle-time improvement across the entire fab is much smaller, as EUV exposure is only one part of a multi-hundred-step process.

Other tools, such as deposition, etch, inspection, and cleaning, dominate total wafer cycle time, which is around 12 weeks for 5nm-class products. So, at best, the newer ASML tools might cut total wafer cycle time by 5%–10%, but their performance advantage over predecessors will not translate into a 40% reduction in cycle time.

To sum things up, Musk's comment about Samsung Foundry's upcoming fab in Taylor, Texas, being 'more advanced' than TSMC's Fab 21 phase 1 is technically true, but it is largely irrelevant to Tesla's immediate production needs.

Last year, Tesla announced its new AI5 custom silicon designed for inference in its vehicles needed for autonomous driving. Like previous generations, it's designed in-house and was to be manufactured by TSMC, but now Samsung has also been added to the mix, with both foundries set to fabricate it at their U.S.-based facilities. In a new earnings call, Musk also detailed how the AI5 chip came to be and what it represents for the future of not only Tesla but also the intentional spillover that his other company, xAI, will benefit from.

“We are actually going to focus both TSMC and Samsung initially on AI5... Musk said. "Technically, the Samsung fab has slightly more advanced equipment than the TSMC fab. These will both be made in the US, one TSMC in Arizona, Samsung in Texas... Samsung, it's worth noting, does manufacture our AI4 computer and does a great job doing that.”

Musk spoke at a grand show about how the AI5 is superior to the outgoing AI4, saying that it performs up to 40x faster in certain tasks. Of course, only time will tell how true that claim is, or what the metric of comparison even is. He attributed the main driving factor behind this radical lead to Tesla's tunnel vision. Since the AI5 only has one client, it can be designed with absolute efficiency, focusing on current needs, while ditching legacy hardware that would otherwise hold back the silicon, denying it of its full potential. “We know what the chip needs to do, and just as importantly, we know what the chip doesn't need to do,” claimed Elon.

The plan is to produce more AI5 chips than necessary, creating a buffer that would spill into xAI, where the extra silicon would be put to use in data centers. "We already use AI4 for training in our data center. We use a combination of AI4 and Nvidia hardware,” said Elon. Nvidia's Hopper and Grace Blackwell hardware predominantly make up xAI's compute at the moment — with Elon already promising to switch to custom silicon — so, bringing in even more foreign chips from his car company could further jeopardize the GPU pool for the Green Team.

In many ways, Elon is going the opposite direction; where Nvidia has a myriad of eager customers, the AI5 has to satisfy no one beyond the world's richest person. That being said, Musk remains both grateful and appreciative of Nvidia in its technical ability to tackle contemporary silicon design:

"When you look at the various logic blocks in the chip, you increase the number of logic blocks, you also increase the interconnections between the logic blocks. So you can think of it like highways – how many highways do you need to connect the various parts of the chip?" he explained. Especially if you're not sure how much data is going to go between each logic block on the chip, then you kind of end up having giant highways going all over the place. It becomes almost an impossibly difficult design problem, and Nvidia has done an amazing job of dealing with almost an impossibly difficult set of requirements."

Originally, it was only TSMC that was supposedly contracted to build AI5 for Tesla, but in July of this year, Musk disclosed that Samsung had secured a $16.5 billion chipmaking deal with Tesla, which has now been confirmed as a manufacturer for the chip, alongside TSMC. Elon boasted that "this is a beautiful chip" that he's "poured so much life energy into" and is confident that it's "going to be a winner." Not carrying forward older hardware blocks has allowed Tesla to "fit AI5 in a half-reticle," suggesting impressive yields that would line up with his aforementioned oversupply ambitions.

Samsung is developing an AI-boosted modem chip to enable Earthbound devices to connect directly with Elon Musk’s Starlink service, according to insider sources talking to the Korean Economic Daily (KED).

Currently, connections between SpaceX’s Starlink hardware and tech on planet Earth are routed via a base station infrastructure. However, the insiders suggest that a Samsung modem with a built-in AI-accelerating NPU could revolutionize direct satellite-to-device connectivity.

The key innovation here will be that devices packing the new chip will be able to “predict satellite trajectories and optimize signal links in real time.” If it all plays out as expected, this innovation could upend the world’s telecoms architecture and enable Musk’s Starlink 6G non-terrestrial network (NTN) ambitions to be realized.

As recently as last month, SpaceX bought 50 MHz of wireless spectrum and global mobile satellite service (MSS) frequencies to facilitate a 6G NTN service. This was a considerable investment, reportedly worth $17 billion.

New Exynos NPU up to 55x faster at optimizing satellite connections

A presentation purportedly shared by Samsung execs on Thursday showed that the new Exynos modem can improve beam identification and channel prediction performance “by 55 and 42 times, respectively,” compared with current models, writes the Korean source publication. Obviously, this is enough to enable the real-time tuning capabilities that such modems currently struggle with.

Among the struggling current-gen modems might be those from Samsung’s Exynos Mobile 5000 series, which are currently advertised as being ready, “once NR NTN becomes commercialized.”

Such long-range real-time tuning sounds like quite an energy-intensive processing task. Thus, it will be interesting to see how the first 6G NTN smartphones, for example, fare when connected using this tech. Other potential use-cases of the first-gen 6G NTN modem include applications like automotive and robotics, where extreme power efficiency isn’t quite as crucial.

The KED also says that industry analysts have signaled this yet-another-Musk-collaboration shows Samsung’s semiconductor operations are serious about extending their reach beyond smartphones and memory. Previous evidence of this thrust is the joint sourcing of Tesla’s new A15 chip from TSMC and Samsung.

With so much negativity surrounding artificial intelligence nowadays, it is good to see another serious, compelling use case for AI here. But ultimately, perhaps much of the 6G bandwidth will be used to stream AI-slop cat videos and similar.

Earlier this week, Nvidia’s DGX Spark AI mini-PCs were released to the masses. Symbolic of the splintered state of the AI industry, though, Nvidia’s CEO Jensen Huang hand-delivered DGX Spark systems to Elon Musk and Sam Altman — separately. Altman and Musk once worked closely together, for a common cause, as co-founders of OpenAI, where Jensen once hand-delivered the original DGX-1 nine years ago, long before the AI boom began.

The last time we saw such a star-studded DGX photo-op was when Huang hand-delivered the original DGX-1 to Elon Musk, in his role as co-founder of non-profit OpenAI. Once partners seeking to further the development of safe AI for the benefit of humankind, Musk and Altman have become the fiercest of rivals. The rivalry isn’t very sportsmanlike, either, with the pair now regularly trading barbs and public insults, and even engaging in some acrimonious legal tussles. Since 2023, we have also had two distinct competitive AI products addressing the same market, albeit from different angles: ChatGPT and Grok.

Musk received his DGX Spark while wearing his ‘Chief Engineer at SpaceX’ hat. Huang quipped that he was “delivering the smallest supercomputer next to the biggest rocket,” at the Starbase, Texas facility.

We are reminded by the Nvidia blog that the GB10 Grace Blackwell Superchip-powered DGX Spark packs 128GB of unified memory and delivers a petaflop of AI performance. It is claimed to have enough resources and muscle to run models with 200 billion parameters locally. Moreover, the Green Team’s blog highlights that it has been nine years since Huang personally delivered the DGX-1 to Musk at OpenAI.

Sam Altman also reminisced about Huang’s previous little parcel. “Things have come a long way since the delivery of the DGX-1 9 years ago; amazing to see...” mused the OpenAI boss. Altman was commenting on the President and co-founder (another one) of OpenAI, Greg Brockman's, shared photo. The picture shows Huang nestled between Brockman and Altman.

The Jensen-powered delivery service

Huang famously started his career washing dishes, bussing, and waiting at Denny’s. Now he’s back as a server, but on the menu are slices of AI-accelerating silicon, delivered in person to a select few fellow tech industry giants.

Nvidia’s Jensen-powered delivery service isn’t standard, even though the DGX Spark has gotten a third more expensive since it was first announced. The Nvidia first-party DGX Spark MSRP is now $3,999, and it is shipping direct from Nvidia, Micro Center, and a number of partners.

The first batch of DGX Spark systems was also put into the hands of researchers at AI-processing hungry companies like Anaconda, Cadence, ComfyUI, Docker, Google, Hugging Face, JetBrains, LM Studio, Meta, Microsoft, Ollama, and Roboflow.

If you are interested in the DGX Spark, and its integrated Nvidia AI stack with full CUDA library support, the firm’s partners are also cooking up systems, featuring their own special saucy tech - but they all look fairly similar. DGX Spark systems are being made and marketed by Acer, Asus, Dell Technologies, Gigabyte, HPI, Lenovo, and MSI. You definitely won’t get Jensen-powered delivery with these, though, you’d be lucky to get a free MSI dragon plush toy.

SpaceX revealed its new V3 Starlink satellites, which it says will provide gigabit connectivity to users and enable 60 Tera-bits-per-second downlink capacity on the Starlink network. The company says each launch will add 20 times the capacity of its current V2 mini offering.

The above animation was first shared by SpaceX on Monday, during a live stream of the 11^th flight test of SpaceX’s reusable Starship vehicle, notes PCMag. And now we can see why SpaceX has started to use its more powerful Starship vehicle, in preference over the existing Falcon 9 rocket, which was used for prior generation satellite deployments..

As well as their obviously significantly larger physical form, a V3 satellite is estimated to weigh in at as much as 2,000kg (4,409 pounds). Compare that to the existing V2 Mini satellites, which are less than 600kg, and the V1 satellites at around 300kg.

We’ve established that these V3 Starlink satellites are big and heavy, so what are the network performance gains? SpaceX’s latest social media bulletin regarding the performance of these units is quite brief. However, it highlights that V3 will “bring gigabit connectivity to users and are designed to add 60 Tera-bits-per-second of downlink capacity to the Starlink network.”

Thankfully, there were some prior news releases and filings, which flesh out V3 details. An important one, for example, is that for every upcoming Starship launch, it will be possible to put 60 V3 satellites into orbit. “That's more than 20 times the capacity added with every V2 Mini launch on Falcon 9,” says SpaceX. Moreover, each new V3 satellite boasts 1,000 Gbps of download and 200 Gbps of upload bandwidth, making them better than 10x faster than V2 models.

With the V3 satellites deploying, probably early next year, we might begin to see SpaceX CEO Elon Musk’s original vision of a global broadband service materialize. The new satellites’ next-gen architecture should deliver the increased coverage and speed to iron out connectivity and capacity wrinkles some users are experiencing with the service, as it stands. However, it has previously been indicated that Starlink users will need new hardware to benefit from the best speeds V3 satellites will enable.

Hopefully, SpaceX is still designing even its latest, biggest, and heaviest satellites to completely burn up upon reentry, at the end of their working lives. With V3, there’s a lot more mass to be combusted on the way down to Earth, and as more satellites head into orbit, space debris — and that debris falling back to earth — remains a growing concern.

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.