Musk Just Pledged 16.8 Billion to Build the Biggest Chip Fab on Earth — the Fine Print Tells a Different Story

Musk Just Pledged 16.8 Billion to Build the Biggest Chip Fab on Earth — the Fine Print Tells a Different Story

Let me tell you something that's been sitting wrong with me since Thursday, when Tesla and SpaceX finally put a pin in the map for Terafab. This is the moment the chip shortage officially stopped being a supply problem and became a vertical integration problem. When the biggest buyers on Earth decide the industry can't build fast enough for them, they don't wait in line anymore. They build their own factory. And not just any factory — a 100-million-square-foot megafab in a Texas county that used to cool a coal plant, promising over a terawatt of compute a year.

The best breakdown I've seen this week is from Solving The Money Problem, a 24-minute walk through the numbers that even the optimists can't square with the grid. Watch it, then read the fine print yourself. Because I read it. And folks, the gap between the tweet and the fine print is wide enough to park a data center in.

What Tesla and SpaceX Actually Confirmed

Let's start with what's real, because there is real news here. Terafab is going to Grimes County, Texas, about an hour northwest of Houston. The companies put the first phase at roughly $16.8 billion and said the finished site will cover more than 100 million square feet — which they're calling the largest chip manufacturing facility on the planet. At least 3,000 people would work there, most of them local hires.

The pitch is vertical integration at a scale nobody has ever attempted: logic, memory, packaging, and testing all under one roof, producing chips for Tesla's Optimus robots and Cybercabs, plus the "space-based data centers" SpaceX keeps talking about launching. Tesla's own words on X: "Both Tesla & SpaceX will need far more chips than current & future global production can supply." That's why they're targeting over 1 terawatt of compute per year.

And there's a genuinely smart detail in the site selection. Terafab sits on the Gibbons Creek Reservoir, which used to cool a coal-fired power plant that shut down in 2018. SpaceX says it'll pull water from the reservoir rather than local groundwater — a nod to a community already uneasy about water use and the data centers spreading across Texas. That's the kind of concession you only make when you've learned the community-consent lesson the hard way.

The Fine Print That Tells the Real Story

Here's where my founder's alarm bells start ringing. That $16.8 billion is not the number Musk led with. When Terafab was announced back in March at the defunct Seaholm Power Plant in Austin, the headline was $25 billion. By May, SpaceX's own IPO paperwork was estimating $55 billion for the initial phase and $119 billion total. Now it's a $16.8 billion "initial phase," with "future expansion phases bringing total investment much higher." The number keeps moving, and it only ever moves in one direction.

And that S-1 filing — read it. It described Terafab as only a "general framework" with no binding commitments, no finalized IP split, and no obligation for either side to keep participating. A month before that, Electrek reported Tesla wasn't really going to build the fab itself, with Intel lined up to handle the actual manufacturing. Musk has since confirmed Intel's 14A process — the 1.4-nanometer node — will be used at full-scale Terafab.

So let me be straight with you: the confident tweet naming a county sits on top of a paper trail that says most of this is still undecided. Both of those things can't be fully true.

The Secondary Bottleneck Nobody's Talking About — a Terawatt of Compute Is a Power Problem

Now the part that keeps me up at night. A terawatt of compute a year is not a chip-industry number. It's a physics number. The entire United States runs on roughly half a terawatt of average electricity generation. Musk is talking about building something that computes harder than the whole country currently generates power. No fab on Earth is built to output that, because that number doesn't come from the chip industry — it comes from space ambitions. The 100-kW "AI Sat Mini" concept. The orbital data centers. The galactic civilization speech.

Put the footprint in perspective. One hundred million square feet is more than ten times the floor space of TSMC's biggest gigafabs. TSMC is spending roughly $165 billion to build six fabs in Arizona, and those are a fraction of Terafab's stated footprint. The largest semiconductor manufacturer in history is building six factories for less than what Musk is describing as phase one of one factory.

And then there's the water. The reservoir detail is clever, but a fab of this scale is a water monster. Texas is already fighting over groundwater, reservoir rights, and who gets to cool what. Grimes County residents were nervous before SpaceX reps started sitting down with them. The community-consent fight that has stalled data centers from Virginia to Ohio is coming to a chip fab near Houston.

Why Vertical Integration Is the Most Expensive Habit a Founder Can Develop

I've been running infrastructure for over a decade, and I'll tell you the truth: vertical integration sounds like the ultimate power move until you price it. A foundry is a money furnace. It eats capital, talent, and time for years before it produces a single profitable wafer. TSMC has spent decades and hundreds of billions learning how to do this, and they still make mistakes.

Here's the part nobody wants to say out loud: Intel needs Terafab as much as Musk does. Intel is the foundry in distress — one of only three manufacturers on Earth producing sub-5nm chips at scale, and the one bleeding market share. A 1.4-nanometer process deal with the biggest compute consumer on the planet is a lifeline. So when you hear "Tesla, SpaceX, and Intel are building Terafab," what you're actually hearing is a car company, a rocket company, and a foundry in crisis all hoping the other guy pays for it.

Meanwhile, Tesla's actual chips are already contracted out. Samsung's Taylor, Texas fab is in production on Tesla's AI5 chip. Tesla signed a $16.5 billion deal with Samsung last year to build the next-generation AI6. TSMC covers the rest. So Terafab isn't about Tesla's near-term supply at all — it's a bet on a future where Musk's companies need more compute than the entire industry can produce. And the car business that's supposed to fund it has stopped growing.

What This Actually Means for Independent Hosting Providers

So what does a Texas megafab that may or may not exist do to your business? More than you'd think.

First, treat the announcement as marketing until the permits finalize. The S-1 called Terafab a "general framework" three months ago. A county on a map is progress, but it's not a binding commitment. Don't reprice your hardware roadmap off a Musk keynote.

Second, watch what this does to merchant chip supply. Here's the counterintuitive part: if Musk actually builds, he stops competing with you for TSMC and Samsung wafers — he's making his own. That could loosen the merchant market. If he doesn't build, the demand stays and the squeeze continues. Either way, his paper promises shouldn't move your order book.

Third, watch Intel's 14A ramp as the real signal. Terafab runs on Intel's process. If Intel delivers 14A at scale, that's a foundry comeback that changes server pricing for everyone. If it slips — and Intel processes have a habit of slipping — the whole Terafab timeline slides with it.

Fourth, understand that every one of these megaprojects lands on the same grid. A fab this size doesn't just need chips' worth of power — it needs a power plant's worth of power. Texas ERCOT is already at the edge. Rate pressure, interconnection queues, and power-quality headaches are coming for every operator in the region.

Fifth, and this is the one I keep coming back to: don't build your capacity plan around somebody else's terawatt. Build it around your own order book. The hyperscalers and the Musks of the world are playing a different game with different money. Your game is serving customers who need machines today.

The Bottom Line

Musk's own line at the March announcement was the most honest thing anyone has said about this whole cycle: "We either build the Terafab, or we don't have the chips, and we need the chips, so we build the Terafab." That's not confidence. That's desperation wearing a cape. It's the same desperation driving every hyperscaler to gas plants, ocean platforms, and now their own foundries — the buildout has hit the point where the biggest players are building the entire supply chain from scratch.

Maybe Terafab happens. Musk has done impossible things before, and the site selection is more real than anything he showed us in March. But a founder doesn't get to claim the largest factory on Earth before pouring a foundation, ent? The fine print says this is a framework. The tweet says it's the future. I know which one I'd bet my capex on — and it's not the tweet.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.