Friday, 23 October 2026Global tech news & guides — what changed, and what it means for you.
AuTechWire
Original illustration: semiconductor fab complex at dusk in Texas, crane silhouetted against an amber sky, wafer-handling robots in the foreground, teal circuit traces arcing across the sky, teal and amber grade

Elon Musk Tried to Buy Chip Plants From Intel, TSMC and Micron First. They Said No — So He’s Building Terafab

by

In brief

  • Elon Musk first tried to buy existing chip plants from Intel, TSMC and Micron for his Terafab project and was turned down by all three, according to The Elec.
  • Rejected, he pivoted to partnerships and ground-up construction: a Texas semiconductor complex run by Tesla, SpaceX and xAI with an initial $16.8 billion budget and a projected $119 billion total cost.
  • The target: 100,000 wafers a month before 2030, scaling to 1 million a month within a decade — for Tesla’s self-driving chips, Optimus robots, Starlink satellites and planned orbital data centers.
  • Intel has joined as a partner on its 14A process; TSMC says it would only participate if it operates the plants — while Musk says TSMC’s maximum role would be subcontractor.
  • The scale reality check: TSMC alone expects about 120,000 wafers a month of leading-edge 2-nanometer chips by the end of 2026, so Terafab’s first phase is meaningful but not market-dominating.

Elon Musk’s plan to build one of the world’s largest chip factories from scratch was not his first choice. According to a report from the Korean outlet The Elec, published October 20, Musk initially approached Intel, TSMC and Micron about acquiring existing fabrication plants for his Terafab project — and was rejected by each of them before shifting to a partnership-and-construction strategy.

The claim comes from an official at an overseas equipment supplier quoted by The Elec: “My understanding is that Musk initially considered acquiring manufacturing facilities from Intel, TSMC and Micron, but shifted toward partnerships after being turned down.” It is a single-sourced report, and none of the three chipmakers has publicly confirmed the approaches. But it fits the pattern of everything Musk has said publicly about why Terafab exists: the existing chip industry cannot expand fast enough for what he wants to build, so he has decided to become the chip industry.

Why the world’s biggest potential chip customer can’t just buy capacity

Musk laid out the logic himself at the project’s March launch. “We’re very grateful to our existing supply chain, to Samsung, TSMC, Micron and others. But there’s a maximum rate at which they’re comfortable expanding. That rate is much less than we would like. And we need the chips, so we’re going to build the Terafab.”

The demand profile behind that statement is unusually large because it spans four companies at once. Terafab is meant to supply chips for Tesla’s autonomous driving systems and Optimus humanoid robots, SpaceX’s Starlink satellites, and the space-based data centers SpaceX has discussed — plus xAI’s AI training needs. Musk has described the complex as a combined design, manufacturing, packaging and testing operation intended to produce more than 1 terawatt of AI computing capacity a year, a figure that would dwarf current global output.

The conventional routes to that capacity are both constrained. Renting compute from hyperscalers means competing with Anthropic, OpenAI and enterprise customers for forward capacity that is already heavily committed. Buying chips from Nvidia means waiting in a queue where Nvidia has booked over half of TSMC’s advanced packaging capacity through 2027. The same capacity race reaches consumers at Apple’s October 28 event, where the touch-screen MacBook Pro’s M5 Pro and M5 Max chips are also built on TSMC’s leading-edge processes. The chips that power modern AI are allocated years in advance, and the allocation schedule favors whoever booked early and paid in advance. Musk’s answer is to stop being a customer of the chip industry and join the small club of companies that control manufacturing directly.

The partnerships — and the control dispute

After the reported rejections, the partnership map took shape quickly. Intel has joined as a partner, with its involvement centered on its 14A manufacturing process. Discussions with TSMC are ongoing, but they have exposed a fundamental disagreement about who runs what: Musk has said TSMC’s maximum role in the project would be as a subcontractor, while TSMC chief executive C.C. Wei said during the company’s latest earnings call that TSMC would need to operate the plants for a partnership to work. That tension — between Musk’s insistence on control and the foundry industry’s insistence on operational authority — is the fault line to watch.

Beyond the headline names, Terafab is reportedly in talks with about 10 Korean chip suppliers, and equipment giants Lam Research and Applied Materials are understood to have established dedicated Terafab teams. The discussions cover the unglamorous backbone of chipmaking: piping, polishing slurries, cleaning chemicals and other materials. A Fujifilm official told The Elec that most major materials suppliers have probably already held discussions with the project.

The execution speed nobody in the industry has seen

What has startled suppliers is not the ambition but the pace. Tesla, SpaceX and xAI announced the joint project in March; SpaceX signed a tax abatement agreement with Texas earlier this month and construction has begun. Tesla is also building a process R&D pilot fab at its Gigafactory Texas plant in Austin, with the primary objective of securing in-house chip supplies.

“When we send inquiries, even in the early morning hours in the US, we receive responses within 30 minutes to an hour,” an official at a Korean supplier told The Elec. “Engineers have decision-making authority, making the way they work completely different from dealing with ordinary customers.” Another supplier official said: “I’ve never seen this level of execution speed elsewhere in the semiconductor industry.”

Speed of procurement, however, is not speed of chipmaking. Building a fab shell is the easy part. The hard parts are the ones analysts keep flagging: access to EUV lithography tools (ASML’s machines, with years-long order books), ramping yields on leading-edge processes, and the sheer capital intensity. Musk himself has acknowledged the gap: he has said publicly that he does not know how to make chips. The Intel partnership on 14A is presumably meant to close exactly that gap — but Intel’s own foundry business has spent years struggling with the same yield and execution problems.

What the numbers actually mean

Industry observers expect Terafab’s initial output to reach roughly 100,000 wafers a month before 2030, with the long-term target of 1 million wafers a month within a decade. For context, TSMC alone is expected to produce about 120,000 wafers a month of its leading-edge 2-nanometer technology by the end of 2026. In other words, Terafab’s entire first phase — a $16.8 billion initial investment — would produce less leading-edge volume than TSMC will be making next year. The 1-million-a-month goal would be a different story, but it is a ten-year goal in an industry where ten-year forecasts rarely survive contact with reality.

The $119 billion total projected cost, reported via Wccftech, would make Terafab one of the most expensive industrial projects ever attempted — funded across Tesla, SpaceX and xAI at a moment when all three are burning capital on their own moonshots. It is worth remembering that this is the same industry logic that has driven the AI infrastructure boom all year: when we covered Microsoft’s exit from Nscale’s UK data center deal, the binding constraint was electricity. For Musk, the binding constraint is chip supply. The industry keeps discovering that the scarcest input is whatever it assumed would be available.

The million-times economy

Musk has connected Terafab to a much larger vision. In a podcast conversation with Walter Isaacson in August, he argued that physical AI — robots that can conduct experiments, manufacture goods and build infrastructure — combined with lunar development and space expansion, could eventually support an economy a million times larger than Earth’s. Terafab, in that telling, is not a chip factory. It is the first industrial foothold of an off-world economy: the chips that run the robots that build the Moon bases.

That framing is pure Musk — every factory is a step toward Mars — and it is doing real rhetorical work here. A $119 billion chip fab is easier to justify as the foundation of a million-times economy than as a way to make car chips cheaper. But the near-term economics will be judged on far more mundane terms: whether the Texas site can produce usable wafers, at acceptable yields, before 2030, without the operational control TSMC insists on.

The industry has seen this movie before. Every few years, a deep-pocketed outsider decides chipmaking can’t be that hard. It is that hard. The difference this time is that the outsider controls four companies with bottomless chip demand, has Intel’s process technology in his corner, and moves at a speed that has genuinely shocked suppliers. Whether that is enough to do what Intel, TSMC and Micron all declined to sell him remains the open question — and the most interesting industrial bet of the AI era.

Update, October 21, 2026 — Musk puts dates on the buildout

Terafab now has a public timeline. On SpaceX’s third-quarter earnings call on October 20, Musk said the Texas complex is targeting volume chip production in the third quarter of 2028 — with the fourth quarter of 2028 still in the frame — and that an SRAM test chip should be working around May or June 2027 at the project’s Austin research fab. The dates come heavily hedged: Musk called them “very aggressive” and “insanely fast timeframes for chip production,” and said he puts “big error bars” on the internal targets. He thanked ASML and other equipment suppliers for accelerating deliveries but would not confirm EUV lithography orders. Separately, he posted on X that Terafab’s output value would rank in the top 20 of country GDP, describing the long-term goal as “a terawatt of superintelligence chips per year.”

He also said SpaceX has “some significant potential breakthroughs in chip fabrication” that he did not detail. The timelines sharpen the bet: construction barely begun, a first test chip in roughly 18 months, volume production about two years after that. In the United States, cutting-edge logic fabs typically take three or more years to build.

Sources

  • The Elec, “Musk’s Terafab in Talks With About 10 Korean Chip Suppliers,” October 20, 2026 (primary report on supplier talks and the rejected acquisition approaches).
  • Wccftech, “Elon Musk Tried To Buy Chip Plants From Intel, TSMC And Micron Before Building Terafab,” October 20, 2026.
  • SemiWiki, “Musk Reportedly Sought Intel, TSMC and Micron Plants Before Terafab Push,” October 21, 2026.
  • Reuters, “SpaceX and Tesla’s Terafab targets 2028 for volume chip production, Musk says,” October 20, 2026 (earnings-call timeline: SRAM test chip May/June 2027, volume production Q3 2028).
  • TeslaNewswire, “Terafab timeline: test chip mid-2027, volume production Q3 2028,” October 20, 2026 (call transcript details).
  • Musk’s March 2026 Terafab launch remarks (supplier expansion quote), via trade reporting.

Related reading: Anthropic in Talks for Second $45B Nscale Deal: What It Means · Microsoft Walks Away From the Nscale £2B UK Data Center Deal — No Grid Power Until the 2030s

Sources: The Elec (2026-10-20 — primary report on the rejected Intel/TSMC/Micron acquisition approaches and the Korean supplier talks); Wccftech (2026-10-20 — $119B total-cost reporting); SemiWiki (2026-10-21 — reporting on the same claims); Reuters (2026-10-20 — earnings-call timeline: SRAM test chip May/June 2027, volume production Q3 2028); TeslaNewswire (2026-10-20 — call transcript details, top-20-GDP claim).

The rejection claim is single-sourced reporting via The Elec; none of Intel, TSMC or Micron has publicly confirmed the approaches as of October 21, 2026. Timeline figures are Musk’s stated targets from the October 20 earnings call, hedged by Musk himself as “very aggressive.”

Shoaib Hassan

Shoaib Hassan is the founder and editor of AuTechWire, an independent technology publication covering broadband, mobile networks, privacy, startups, and consumer tech.

Related Stories

Leave a Comment