Did you know Elon Musk looked at the global AI chip shortage and apparently thought, fine, I’ll build the factory myself?
That’s right! On August 6, Tesla and SpaceX announced an initial $16.8 billion investment in a Texas semiconductor complex called Terafab. It will sit in Grimes County, about an hour northwest of Houston, and make AI chips for Tesla’s Optimus robots and Cybercabs, plus the orbital data centres SpaceX wants to build.
Table of Contents
Terafab Would Dwarf Every Building on Earth
Musk called it the largest and most valuable building on Earth, and for once the hyperbole undersells it.
Plans call for more than 100 million square feet of manufacturing space. The current record holder, the New Century Global Center in Chengdu, covers about 18.9 million. So Terafab would be roughly five times the largest building humans have ever completed.
The design is unusual too. Most chipmaking splits logic, memory, packaging, and testing across separate suppliers in different countries. Terafab puts all four under one roof.
The Terawatt Claim Deserves a Second Look
You see, the justification for all this is demand. Tesla and SpaceX say their combined need will exceed one terawatt of compute per year, which they describe as larger than current global supply.
Now do the arithmetic. A terawatt is a thousand gigawatts. By Musk’s own estimate, the world’s entire AI compute output currently runs around 20 gigawatts a year.
So the claim works out to roughly fifty times global production, coming mostly from one building in rural Texas.
Musk put it plainly in March. Either they build Terafab or they don’t have the chips, and they need the chips.
It Will Not Touch the Power Grid
Here’s the detail that got buried. According to the executed tax abatement agreement, Terafab expects to run entirely on on-site power plants and not to use grid electricity at all.
Fabs need extremely stable voltage, because a flicker ruins a wafer. So self-supply makes engineering sense. But at this scale it’s close to unprecedented.
And the fuel is natural gas. SpaceX energy lead Riley Trettel told county commissioners the site will run on gas fired plants plus very large battery arrays.
Notice what’s missing. Tesla is a full partner on this project. Tesla sells solar panels, Powerwalls, and grid scale Megapacks. In June, Tesla and Sunrun launched a 16 gigawatt virtual power plant marketed specifically at data centres. None of that appears in the plan. Musk did, however, acquire gas turbine operator APR Energy earlier this year.
A factory built beside a retired coal plant’s reservoir, powered by gas, co-owned by a company whose stated mission is accelerating the transition to sustainable energy.
Terafab isn’t alone in this, though. Texas now has more than 70 gigawatts of bring your own power capacity charted out, as Google, Meta, and Microsoft all announce gas projects to sidestep interconnection queues. That rush has driven gas plant costs up 66%. So Terafab is an extreme version of an industry wide move rather than an outlier.
What Was Announced, and What Wasn’t
But here’s the twist. The announcement settled where the factory goes. It settled almost nothing else.
Nobody has said how Tesla and SpaceX split the investment, who owns the resulting entity, or when the first commercial wafer appears. Intel has confirmed it will contribute something while staying cagey about what.
The numbers keep moving too. The first phase was floated at $25 billion in March, SpaceX previously estimated $55 billion, earlier Texas filings contemplated up to $119 billion, and the announced figure came in at $16.8 billion. None of those have been reconciled.
Meanwhile SpaceX’s May IPO filing called Terafab a general framework with no binding commitment, and the company did not mention it on its first earnings call days before the announcement. The funding capacity is certainly there, since that IPO raised $87.5 billion.
Fabs Take Longer Than Anyone Promises
Here’s the reality check the announcement skipped.
TSMC announced its Arizona fab in May 2020, planning production for 2024. High volume production actually began in the final quarter of 2024, and the second fab slipped from 2026 to 2027 or later. Labour shortages, permitting, and the sheer difficulty of qualifying a new site caused the delays.
That is the most experienced fab builder on Earth, working on a conventional site, on a $40 billion budget. Terafab is five times the largest building ever completed, combines four manufacturing stages competitors keep separate, and has to build its own power station first.
Anyone modelling first silicon before the 2030s should show their working.
The broader pattern supports caution. Wood Mackenzie counted more than 1,000 gigawatts of data centre power requests across the US and expects barely a quarter of them to get built. Announcements in this sector are cheap. Concrete is not.
Grimes County Is Not Unanimous
Worth mentioning, the commissioners approved the reinvestment zone and abatement on a four to one vote, and a local petition opposed it. Hundreds of residents turned up to raise concerns about transparency and incentives.
SpaceX secured a 100% county tax abatement in exchange for spending at least $5 billion by 2030 and creating 1,800 full time jobs by 2035. Texas added $30 million. Note the job figure, because the state press release advertised 3,000.
SpaceX has committed publicly that the project will not raise electricity bills, deplete local water, or dry up the Navasota River, and says it chose the Gibbons Creek site for its rainwater storage rather than groundwater access.
Looking Forward
Certainly, AI started as a software race. Then it became a GPU race. Now we’re pouring concrete and building power plants.
Yet Terafab today is a site, a grant, a first phase estimate, and an enormous ambition. Watch for the investment split, the ownership structure, Intel’s role, and a production date. Until those land, the building is a rendering.
At this rate, the next AI startup accelerator might include a construction permit.
Finally, leave your thoughts in the comments below, because I’m curious. Is owning the entire stack the smartest move in AI, or the most expensive way to learn a lesson? Let me know what you think.
And if you want more of this, subscribe to our newsletter and join us on YouTube.For tech builders, check out CloudMelonVis, where we dive deeper into the technologies shaping what we build. And for your weekly insights of AI, cloud native, startups, and venture capital, follow CVisiona, where we break down what happened, why it matters, and what to watch next. See you in the next one!
Source : Tesla and SpaceX will invest $16.8B to start building Terafab chip factory in Texas from TechCrunch, Musk’s planned chip factory is five times bigger than the world’s current largest building from Fortune, SpaceX plans to power $16.8B Terafab without the grid from Data Center Knowledge, and Tesla and SpaceX commit $16.8 billion to Texas Terafab, but the deal structure is still opaque from MLQ News
















Your First AI Agent Will Fail. Here’s How to Build One That Doesn’t
Podcast Episode: Open versus Closed AI
Anthropic Custom Chips Are a Margin Play, Not an Nvidia Divorce
AI agents on Kubernetes: who controls their actions?
Open-source AI infrastructure: where can a startup ACTUALLY make money?
The Open Weight AI Debate Is No Longer About Safety Alone