TERAFAB

TERAFAB is a planned chip factory that Tesla and SpaceX intend to build together, with the stated goal of producing more than one terawatt of compute per year. A fab is a factory that turns blank silicon wafers into finished chips. Elon Musk unveiled the project on March 21, 2026, at the historic Seaholm Power Plant in Austin. The plan is to make Tesla's own artificial intelligence (AI) chips at a scale no one has ever attempted, instead of renting factory space from outside chipmakers.

What is TERAFAB?

TERAFAB is Tesla and SpaceX's attempt to own the hardest part of the AI supply chain, the chips themselves. Tesla's own filings describe it as expanding into chip-making to keep its supply large enough and steady enough. Every chip is reserved for use inside the companies, with no plans to sell any to outside customers. Grokipedia also mentions xAI being involved, though Tesla's filings name only the Tesla and SpaceX partnership.

Why TERAFAB Exists

Tesla expects its need for chips, driven by the Optimus humanoid robot, the Robotaxi and Cybercab, and Full Self-Driving (FSD), to outrun every source of supply that exists or is even planned. Grokipedia puts that appetite at 100 to 200 billion custom AI chips a year. Tesla's Q1 2026 shareholder update says it is bringing logic, memory, and advanced packaging together in one place so it can move faster and lean less on outside chipmakers like Taiwan Semiconductor Manufacturing Company (TSMC), Samsung, and Micron. The deeper worry, per Grokipedia, is a squeeze in 2028 to 2029, when Tesla believes outside supply will run dry, plus the risk that comes with relying on chips made overseas. Most of what this factory makes is not even meant to stay on the ground.

That split, with roughly four-fifths reserved for space, shows that TERAFAB is less a parts supplier for cars and more the chip backbone for data centers and satellites in orbit.

How it Works

The first product tied to this effort is the AI5 inference processor, whose final design Tesla finished in April 2026. Tesla is aiming for a 50x jump in overall performance over the current AI4, made up of a 10x gain in raw computing power, 9x more memory, and a 5x gain on hardened block quantization and softmax, which are the math steps that let a chip run a neural network efficiently.

The fab meant to make these chips throws out decades of how the industry does things. Per Grokipedia, the design aims for a 2 nanometer (nm) node, meaning some of the smallest transistors made anywhere, and it drops the giant building-wide cleanroom in favor of sealing off each wafer on its own. Elon Musk sums up the idea in his usual style: "Tesla will have a 2nm fab, and I can eat a cheeseburger and smoke a cigar in the fab." Power is the other big constraint, since Grokipedia cites more than 10 gigawatts for the main site, roughly what several large cities draw.

The Economics of TERAFAB

Grokipedia puts the cost at 20 to 25 billion dollars, and Tesla's cash-flow statements note a 2.0 billion dollar SpaceX investment, a sign of how tightly the two companies are now tied together around this build. The math underneath is simple: how many good wafers you make, set against the chip bill Tesla would otherwise pay outside chipmakers at full markup. Owning the line cuts out that markup, shortens how long chips take to get, and rides the cost down as volume climbs, and every chip is kept for internal use rather than sold. Tesla frames the payoff as getting ahead of the 2028 to 2029 shortage it expects and lowering the risk of relying on overseas factories.

Current Status of TERAFAB

As of the middle of 2026, TERAFAB is still in early construction, not production. Tesla's Q1 2026 update says the work starts with the Tesla-owned Research Fab on the Gigafactory Texas campus, and that the larger chip-capacity build-out is underway. The AI5 design is finished, with high-volume production planned for 2027 and the follow-on AI6 chip targeted for 2028. The location for the main megasite is still being decided.

What Comes Next

The sheer size of the goal is what stands out. Grokipedia describes a start at 100,000 wafer starts per month (WSPM) climbing toward one million, a tenfold jump.

For perspective, a single top-of-the-line commercial fab today runs in the tens of thousands of WSPM, so one million would put a single site in a class all its own. The risk cuts both ways. A cleanroom-free, terawatt-scale, multibillion-dollar fab that works rewrites the economics of leading-edge chips, and one that does not becomes an expensive lesson. Then again, an expensive lesson is what people predicted for reusable rockets and cheap electric cars, and Elon proved them wrong both times.

The Bottom Line

TERAFAB is Elon Musk's bet that the cheapest way to guarantee enough AI chips is to stop buying them and start making them himself, at a scale and with a design that break most of the industry's rules. Whether that bet turns into a lasting cost advantage or a cautionary tale comes down to whether silicon, for once, cooperates, and few people have a better record of bending hard problems to their will.

Related

Keep reading: Colossus 1, Colossus 2. Zoom out to the SpaceXAI overview, or open the Quote Library.