Two Men. One War. One $25 Billion Bet.
Picture this. It's early 2026. A war has just broken out in the Middle East. American data centers have been hit by missile strikes. A Dutch factory that makes one essential machine is now a geopolitical target. And two of the most powerful men in technology are staring at the same problem — and reaching completely opposite conclusions about what to do next.
On one side: Elon Musk. Founder of Tesla, SpaceX, and xAI. A man who has spent the last three years writing enormous cheques to Nvidia just to keep his AI ambitions alive. He just announced he's done doing that. He's going to build his own chip factory — from scratch — for $25 billion. He's calling it TERAFAB.
On the other side: Jensen Huang. The CEO of Nvidia. The man whose company is now worth more than most countries. He heard about TERAFAB and, with the calm confidence of someone who has been making chips for 30 years, offered a single sentence of response: "Advanced chip production ranks among the most technically demanding industrial undertakings in existence."
That's the tech world equivalent of a pat on the head.
This is the story of TERAFAB — what it actually is, why Musk is betting everything on it, and whether Jensen Huang is right to be so relaxed. Strap in. This one goes deep.
Wait — What's a FAB? Start Here.
Before we can understand why TERAFAB is such a big deal, we need to understand the word that's hiding inside it. Because "FAB" is the most important word in this entire story.
A FAB — short for fabrication plant — is the factory where computer chips are physically made. Think of it as an extraordinarily complex, impossibly clean kitchen where the recipe is a microchip and the ingredients are silicon, chemicals, and light.
The chips inside your phone, your laptop, and every AI model you've ever used were all "baked" inside a FAB. The problem? These factories are extraordinarily difficult and expensive to build. A single advanced chip FAB costs $10–20 billion just to construct — and takes 4–5 years to become operational. There are fewer than a dozen on Earth capable of making the most advanced chips. The world's largest is operated by a Taiwanese company called TSMC. They've spent 35 years and hundreds of billions of dollars to get where they are today.
When Musk says he's building TERAFAB, he is saying: I am going to build one of those, but bigger — in a fraction of the time.
Now you understand why Jensen Huang didn't exactly start sweating.
When engineers talk about chip sizes in nanometres, they're describing how tiny the components inside a chip are. A nanometre is one billionth of a metre. A human hair is roughly 80,000 nanometres wide. The chips Musk wants to make inside TERAFAB are 2 nanometres — which means the transistors (the tiny switches that power all computing) are almost incomprehensibly small.
The smaller the transistor, the more you can fit on a chip, and the faster and more efficient it becomes. 2nm is the most advanced chip technology entering commercial production right now. Only TSMC and Samsung can currently do it. TERAFAB wants to do it from day one — which is rather like saying you've never baked before, and your first recipe is a seven-tier wedding cake.
What TERAFAB Actually Is — The Full Picture
Tesla and SpaceX have jointly announced TERAFAB — a $25 billion semiconductor fabrication facility in Austin, Texas. Unlike a normal chip factory, which specialises in one part of the process, TERAFAB is designed to do everything under one roof: chip design, the actual printing of chips, memory production, packaging, and testing. It targets 2-nanometre technology — the cutting edge of what's possible today.
TERAFAB will make two types of chips. The first is optimised for edge and inference computing — the kind of processing that happens inside Tesla's electric vehicles, robotaxis, and the Optimus humanoid robots. The second is a high-power chip designed for space — to be used by SpaceX and Musk's AI company, xAI. And here's where it gets genuinely sci-fi: Musk says 80% of TERAFAB's total output will eventually go into orbit.
Cloud computing means your device sends a request to a powerful server somewhere far away (a data center), that server does the thinking, and sends back an answer. It's like asking a librarian in another city to look something up for you.
Edge computing means the thinking happens right there — inside your device, your car, your robot. No internet required, no waiting. It's like having the librarian sitting next to you. For a Tesla driving at 120km/h, waiting for a server to decide whether to brake is not an option. The car needs to think for itself, instantly. That's edge computing — and it needs a specific kind of chip to do it.
The space ambition is where the numbers become almost comical. Musk wants to eventually support one terawatt of computing power per year from orbital satellites — powered by solar energy in space, where sunlight is five times more intense than on Earth's surface. To put that in context: all the AI compute on Earth today is roughly 20 gigawatts per year. Musk wants fifty times that — in orbit.
Power is measured in watts. Your phone charger is about 20 watts. A microwave is 1,000 watts (1 kilowatt). A large power station produces about 1 gigawatt (1,000,000 kilowatts). A terawatt is 1,000 gigawatts. It is, in short, an almost unimaginable amount of power. The entire world currently uses about 18 terawatts of energy in total — for everything. Musk wants to dedicate 1 terawatt just to computing. In space. This is why analysts are simultaneously impressed and concerned.
Was This About the War — or Was It Always Coming?
One of the most important questions surrounding TERAFAB is whether the Iran conflict triggered it. The short answer is no — but it gave Musk the perfect moment to announce it loudly.
Musk first publicly floated the TERAFAB concept in November 2025 — four months before any military action. He said his current chip suppliers TSMC and Samsung were "still not enough" and that Tesla might need to build its own factory. Construction work was already visible near the Gigafactory Texas site. Job postings for TERAFAB-related roles had appeared on Tesla's website. This was already in motion.
But when strikes hit American data centers, a Qatari helium plant, and the Strait of Hormuz was temporarily closed — the narrative around TERAFAB transformed overnight. What would have been reported as "Musk's Latest Ambitious Promise" suddenly became "National Security Infrastructure." TERAFAB stopped being an ego project and started being a geopolitical necessity. Musk didn't invent TERAFAB because of the war. He unveiled it because of the war.
"The war didn't create TERAFAB. But it gave Musk the most powerful possible backdrop to announce it as a national necessity rather than a personal obsession."— The strategic reality behind the headline
Musk vs Jensen — Same Diagnosis, Completely Different Medicine
Here's the genuinely fascinating part of this story: Elon Musk and Jensen Huang agree on the problem. Both men looked at the same data — the explosive, seemingly unlimited demand for AI compute — and reached the same conclusion: Earth cannot generate enough power or chip capacity to meet where AI is heading.
But their solutions could not be more different. Jensen's answer is to make better chips faster, lean on trusted partners like TSMC, and invest every dollar into software and ecosystem. Musk's answer is to own the entire chain — from the silicon wafer to the satellite in orbit — and make himself completely independent of anyone else.
Nvidia's genius, as one analyst put it rather brutally, was not building its own factories. By outsourcing manufacturing to TSMC, Nvidia could put all its capital into chip design and into CUDA — the software that every AI lab on Earth uses to run their models.
CUDA stands for Compute Unified Device Architecture. It's software created by Nvidia that allows programmers to use Nvidia's chips to do general computing tasks — not just graphics. When AI researchers train a model like ChatGPT or Gemini, they're almost certainly using CUDA to do it.
Here's the brutal truth: CUDA has a 20-year head start. Every major AI lab has built their entire workflow on top of it. Switching away from Nvidia chips doesn't just mean buying different hardware — it means rewriting years of code. This "software moat" is arguably more powerful than any single chip Nvidia makes. TERAFAB has no equivalent. Musk is starting from zero on software — which is the hardest part of all.
| Dimension | 🟢 TERAFAB (Musk) | 🔵 NVIDIA (Jensen) |
|---|---|---|
| Strategy | Own everything — design + factory + software | Own design + software, outsource factory to TSMC |
| Software Moat | None — starting from zero | CUDA — 20 years, industry standard |
| Target Market | Internal only (Tesla, SpaceX, xAI) | Every AI lab on Earth |
| Current Revenue | No external chip sales planned | $90B+ data center revenue in 2025 |
| Fab Experience | Zero | None (outsources) — but partners with experts |
| Geopolitical Risk | Low — 100% US-based | High — dependent on Taiwan-based TSMC |
| Timeline | Small batch 2026, volume 2027+ | Already shipping at massive scale today |
| War Narrative | Perfect — reshoring is now patriotic | Vulnerable — Taiwan/Korea exposure |
The one area where TERAFAB genuinely wins: geopolitics. If a conflict over Taiwan ever disrupts TSMC — which produces the majority of the world's most advanced chips — the entire global AI industry would grind to a halt. TERAFAB, built entirely in Texas, would not. That's not a small thing. That's potentially everything.
The Skeptic's Case — Why This Could Genuinely Fail
To be fair to Jensen Huang's calm smile: the history of Musk's manufacturing promises contains some notable gaps between announcement and reality. In September 2020, at Tesla's Battery Day, Musk promised a revolution in battery manufacturing. Five years later, the programme has been a significant disappointment — slower, harder, and more expensive than anything he described on stage. TERAFAB is Battery Day on steroids.
Every advanced chip factory on Earth needs EUV lithography machines — devices that use extreme ultraviolet light to print circuits onto silicon wafers. These machines are made by exactly one company in the world: ASML, a Dutch firm. ASML produces roughly 50 of these machines per year globally. The waitlist is years long. Musk cannot buy his way to the front of that queue. No amount of money changes the pace at which ASML can manufacture and deliver these machines. Without them, TERAFAB cannot make advanced chips. Full stop.
Imagine trying to draw lines finer than a single strand of DNA — on a piece of silicon the size of your thumbnail — millions of times, perfectly, every single second. That's what EUV (Extreme Ultraviolet) lithography does. It uses a specific wavelength of light — 13.5 nanometres — to "print" the circuits onto a chip, like an impossibly precise projector shining onto light-sensitive material.
Each ASML EUV machine costs around €200 million, weighs 180 tonnes, and contains over 100,000 individual parts. They are, without exaggeration, the most complex machines ever built by human hands. And if you want to make a chip smaller than 7nm — which is what all the best AI chips need — you cannot do it without one.
Chip FAB engineers are among the rarest professionals on Earth. TSMC, Samsung, and Intel have spent decades building their workforce pipelines — with universities, apprenticeships, and internal training programmes going back generations. TERAFAB would need thousands of these specialists. They don't exist in a pool waiting to be hired. Musk cannot simply offer more money and fill the roles. The people who know how to run a 2nm FAB are, almost without exception, already working at the three or four companies that currently operate them.
Tesla's car sales declined for two consecutive years heading into 2026 — its first-ever annual decline in China, a bloodbath in Europe. The company's CFO acknowledged that the full $20–25 billion TERAFAB cost is not yet incorporated into Tesla's already record capital expenditure plan for 2026. You cannot comfortably fund a $25 billion moonshot when your primary business is losing momentum. The financial math only works if Tesla's car sales recover, SpaceX's IPO goes well, and xAI attracts enough outside investment to share the load.
The Believer's Case — He's Done the Impossible Before
And yet. Before you write TERAFAB off entirely, it's worth remembering what Elon Musk's track record actually looks like when the experts say something cannot be done.
In 2008, every aerospace expert on Earth agreed that reusable orbital rockets were physically impractical. The economics didn't work, the engineering was too hard, and the established players had tried and given up. Musk ignored them. Today, SpaceX routinely lands rocket boosters on drone ships in the ocean and reuses them. The company is valued at $1.5 trillion. The experts were completely wrong.
In 2008, Tesla was weeks from bankruptcy. Industry veterans said the economics of electric vehicles would never work at scale. Today Tesla is the most valuable car company on Earth. Again — the experts were wrong.
There's also a strategic coherence to TERAFAB that people outside Musk's world might miss. This announcement ties together the xAI-SpaceX merger, Tesla's solar equipment ambitions, the Optimus humanoid robot programme, and Starship's development into a single, unified architecture. Chips designed in Austin. Launched by SpaceX. Powered by Tesla solar. Run by xAI. Extended eventually to the Moon. If even 30% of that vision materialises, it would be the most consequential industrial project since the Manhattan Project.
"Building fabs actually binds capital they could use to rival Nvidia. If Jensen was afraid of competition, you'd see him pushing Musk to spend his relatively limited funds building fabs that will never be competitive."— Industry analyst, making the sharpest point in the entire debate
That analyst makes a devastating point. Nvidia's stock didn't crash when TERAFAB was announced. Jensen didn't call an emergency board meeting. He smiled, said chip manufacturing is hard, and went back to work. That reaction is either the confidence of a man who genuinely isn't threatened — or the calm of a man who understands that the best outcome for Nvidia is to have Musk spend $25 billion doing something that won't hurt them for years.
The Verdict — Three Truths at Once
The honest answer — the one that requires holding complexity rather than choosing a side — is that TERAFAB is simultaneously real, premature, strategically brilliant, and operationally terrifying. Here are the three truths that can all be true at the same time.
Truth 1 — TERAFAB is real, not a stunt. Construction is already underway near the Gigafactory Texas site. Jobs are being posted. The Tesla-SpaceX-xAI merger gives it actual organisational teeth. This is not a slide deck at a conference. It is happening.
Truth 2 — The war accelerated the announcement, not the idea. Musk floated TERAFAB in November 2025. The Iran conflict simply gave him the best possible moment to announce it — transforming a personal obsession into a national security story. The timing was perfect. The idea was already old.
Truth 3 — TERAFAB is not a threat to Nvidia today, but it is a serious warning for 2028 and beyond. Right now, Musk is still writing large cheques to Jensen. But if TERAFAB delivers even half of what it promises, and if the war continues making TSMC's Taiwan location a strategic liability, the argument for American-made, vertically integrated AI chips becomes politically and commercially overwhelming. That is the scenario Jensen Huang has to plan for — not this year, but three years from now.
The real question isn't whether Musk can build a chip fab. With enough money and time, it's possible. The question is whether he can build one that works — that achieves competitive chip yields, delivers at a cost that justifies the investment, and does so on a timeline that matters. Because the AI race does not wait. Every month that TERAFAB remains a vision rather than a functioning factory is a month that xAI continues writing cheques to the very man Musk is trying to outcompete.
The one-line summary: TERAFAB is Elon Musk betting $25 billion that the future of AI belongs to whoever owns the entire stack — from the silicon to the satellite — and that the war just proved he was right all along. Whether he can actually build it is a different question entirely.
Never Miss a Decode
90 days of honest, clear writing about AI, tech, and the world around you. No jargon. No hype. Always free.
Subscribe Free →