Musk's $25B Terafab May Actually Cost $13 Trillion

Elon Musk announced Terafab on March 21 — a $25 billion chip factory joint venture between Tesla, SpaceX, and xAI targeting 1 terawatt of annual AI compute. Bernstein analysts immediately calculated the actual cost: $5 trillion to $13 trillion, requiring 140 to 360 new factories. The Albis Perception Gap Index scored this story 5.08, with US media framing it as visionary while Chinese outlets ran detailed cost breakdowns calling it structurally impossible.

That gap between $25 billion and $13 trillion isn't a rounding error. It's the distance between a press conference and reality — and it tells you more about the state of AI chip supply than any earnings call this year.

The man who's never made a chip wants to make all of them

Here's what Musk actually proposed from the defunct Seaholm Power Plant in Austin: a single facility that would produce 70% of TSMC's entire global output. At full capacity, 1 million wafer starts per month. At 2-nanometer process technology — the most advanced node only now entering production anywhere on Earth.

"We either build the Terafab or we don't have the chips," Musk told the crowd. "And we need the chips."

The ambition is staggering. The experience behind it is zero.

Tesla has never fabricated a semiconductor. It did once have a chip design team working on the Dojo supercomputer project, but Musk killed Dojo in August 2025, calling it "an evolutionary dead end." The lead architect, Peter Bannon, left with most of the team. Fast Company reported that the engineers who understood custom silicon at Tesla are now at a different startup entirely.

So Terafab is a chip factory proposed by companies that have never made a chip, staffed by a team that doesn't yet exist, at a cost that hasn't been budgeted.

The numbers that broke the analysts

Musk said $20-25 billion. The banks said: not even close.

Barclays analyst Dan Levy ran the figures on Monday and warned that even his own bull-case estimate of $50 billion was "dramatically low." He wrote that the actual capital requirements could be "many multiples higher, and likely well more than an order of magnitude" above what Tesla communicated. That's financial analyst language for: multiply by at least ten.

Bernstein's Stacy Rasgon went further. To reach 1 terawatt of annual compute, Terafab would need the equivalent of 140 to 360 leading-edge fabrication plants. Cost: $5 trillion to $13 trillion. For context, that upper figure is roughly 250 times Tesla's projected capital expenditure for 2026. Tesla ended 2025 with $44.1 billion in cash. The full Terafab vision would cost somewhere between 113 and 295 times that.

"A true Terafab feels like a stretch to us," Rasgon wrote, noting the output would match "the entire current global installed semi capacity."

Tesla's CFO acknowledged the $20-25 billion Terafab cost isn't even in the company's 2026 budget, which already exceeds $20 billion. Barclays separately forecasts Tesla's 2026 free cash flow at negative $3 billion — before any Terafab spending.

Why Musk is panicking (and why he's not alone)

The announcement didn't happen in a vacuum. On Monday — the same day Barclays published its Terafab skepticism — Broadcom dropped its own bombshell.

Broadcom's Natarajan Ramachandran told Reuters that TSMC has hit production capacity limits, calling it a bottleneck that has "choked the supply chain in 2026." He said AI chip demand is running roughly three times above supply. PCB lead times have surged from six weeks to six months. And TSMC won't meaningfully add capacity until 2027.

Ramachandran added something striking: he would have previously described TSMC's capacity as "infinite." Not anymore.

This is the context Musk was reacting to. He told the Austin audience that all the current fabrication facilities on Earth produce only about 2% of what his companies would need. He said he's personally told TSMC, Samsung, and Micron to expand faster and promised to "buy all of their chips." They said no — or rather, they said they'd expand at a pace Musk described as "much less than we would like."

When the world's richest person tells the world's most advanced chipmaker to go faster and the answer is "we'll move at our own pace," something has broken.

The real story isn't Terafab — it's what Terafab means

Moor Insights analyst Patrick Moorhead wrote that Musk probably won't build a chip fab at all. The announcement may be something else: a negotiating tactic aimed at TSMC, or a way to highlight the AI chip crisis ahead of SpaceX's expected IPO later this year.

But even as theatre, Terafab reveals something real.

The global semiconductor industry produced about $600 billion worth of chips in 2025. AI companies alone plan to spend $650 billion on data center infrastructure in 2026. Demand is growing faster than the factories that make the supply. TSMC's disciplined expansion — profitable, measured, shareholder-friendly — worked when growth was 10-15% per year. It doesn't work when demand triples in 18 months.

Meanwhile, the people trying to get around the supply crunch are getting creative in ways that make governments nervous. Last week, three people tied to Super Micro Computer were charged with illegally diverting $2.5 billion in Nvidia-powered servers to China using fake shipping documents and dummy equipment. ByteDance reportedly deployed 36,000 Nvidia chips in Malaysia through an intermediary called Aolani Cloud — a legal grey zone that effectively sidesteps US export controls.

Two different methods, same result: the chips found a way around the wall.

The perception gap: visionary or desperate?

US media covered Terafab as a classic Musk announcement — big vision, bigger questions. Headlines ranged from "most epic chip building exercise in history" (Musk's words, dutifully quoted) to "fever dream" (Bloomberg's framing).

Chinese tech media told a different story. Sina Finance ran detailed cost breakdowns comparing Terafab to TSMC's decades-long, hundreds-of-billions-of-dollars journey. Chinese analysts at Barclays and elsewhere noted that Tesla has zero fabrication experience and that even the initial 100,000 wafer-starts-per-month target would make Terafab larger than most existing fabs. The tone wasn't awe — it was clinical disbelief.

French media framed Terafab through strategic autonomy: reducing dependence on a global chip supply chain concentrated in Taiwan. Le Monde emphasised Musk's lack of semiconductor background. The European angle isn't "can he do it?" — it's "should anyone depend this much on one island for all their chips?"

Three markets, three framings. Musk is simultaneously a genius, a reckless gambler, and a strategic irrelevance depending on which continent you're reading from.

What actually happens next

The most honest reading of Terafab: it won't produce 1 terawatt of compute. It probably won't produce anything for years. What it will do is force a conversation about whether the semiconductor industry's expansion pace matches the AI industry's ambition.

Right now, it doesn't. Broadcom says so. Bernstein says so. Even Musk, in his own way, says so — he just expressed it by announcing a factory that would need $13 trillion and 360 new plants.

The chips that power every AI model, every autonomous vehicle, every satellite constellation all flow through a handful of factories in Taiwan, South Korea, and soon Arizona. When demand triples and supply inches forward, the options narrow to three: wait, steal, or build.

Musk chose build. Super Micro chose steal. TSMC chose wait.

The AI chip crisis isn't coming. It's here. The only question is which of those three strategies wins.