Musk says he’s building Terafab chip plant in Austin, Texas

On Saturday night, Elon Musk stood inside a shuttered downtown Austin power plant and told the world he was building a new one not for electricity, but for intelligence. He called it Terafab, a $20–25 billion joint chip manufacturing venture between Tesla, SpaceX, and xAI that he described as “the most epic chip building exercise in history by far.” The symbolism was intentional: a dead industrial relic giving way to the next industrial revolution.

What Terafab Actually Is And Where It’s Going

Musk officially launched TERAFAB on March 22, 2026, at the old Seaholm Power Plant in downtown Austin a venue chosen with obvious theatrical intent. The announcement confirmed what had been circulating for weeks in industry circles: that Musk’s companies are attempting something the semiconductor industry has never seriously attempted at this scale a single facility designed to perform every stage of chip production, from design and lithography all the way through memory manufacturing, advanced packaging, and final testing, under one roof.

The facility is planned for the North Campus of Tesla’s Gigafactory Texas in eastern Travis County the same sprawling campus that already houses Tesla’s headquarters and its main EV manufacturing line. Multiple job listings under the TERAFAB name began appearing on Tesla’s careers page in the days before the announcement, with roles based in both Austin and Palo Alto. Construction activity was reportedly already visible near the Gigafactory site ahead of Saturday’s event.

Terafab is technically a joint venture between Tesla, SpaceX, and xAI the AI company SpaceX acquired in an all-stock deal in February 2026. It will be run day-to-day by Tesla and SpaceX. Musk said the initial build will be an “advanced technology fab” a smaller, exploratory facility first, before the grander ambitions kick in.

The Core Problem Musk Is Trying to Solve

Here’s where the story gets interesting. Musk isn’t building Terafab out of ambition alone there’s a real supply crunch driving the decision. As Tesla pivots heavily toward robotics, autonomous driving, and AI, it has outgrown what even its biggest suppliers can comfortably provide.

Tesla already has chip agreements with Samsung Electronics, TSMC, and Micron Technology. According to reporting by Bloomberg and Fortune, Musk told the Austin audience directly: “We’re very grateful to our existing supply chain, to Samsung, TSMC, Micron and others. But there’s a maximum rate at which they’re comfortable expanding.” His assessment of the gap is blunt the current global output of AI compute runs at around 20 gigawatts per year. Musk says his companies need something closer to a terawatt. That’s roughly 50 times the entire current global output, from one facility.

We either build the Terafab or we don’t have the chips, and we need the chips, so we build the Terafab Elon Musk, TERAFAB Launch Event, Austin, March 22, 2026

You can debate whether that logic holds up and we will but you can’t deny the internal consistency of the problem statement. Companies building humanoid robots at scale, running full self-driving compute on millions of vehicles, and powering a global AI model need a lot of silicon. More than any existing foundry relationship can guarantee.

Two Types of Chips, Two Very Different Bets

Terafab’s production roadmap calls for two distinct chip types, and the difference between them tells you a lot about where Musk sees his empire heading. The first is an edge-and-inference chip, optimized for lower power draw and on-device computation. These are the chips that will go inside Tesla vehicles, the Cybercab robotaxi fleet, and the Optimus humanoid robot line. They need to be efficient, reliable, and manufacturable in truly massive quantities we’re talking hundreds of billions per year at full capacity.

The second is something far more speculative: a high-power chip designed specifically for space-based deployment. This directly feeds SpaceX and xAI’s emerging orbital computing vision, where AI workloads are run on satellites in low Earth orbit, powered by constant solar irradiance. Musk argued during the presentation that solar energy in space is roughly five times more abundant than on Earth’s surface, and that radiating heat in a vacuum environment makes thermal management far more scalable than it is in terrestrial data centers.

For the space chips, Musk wants Terafab to eventually support a full terawatt of orbital computing a satellite network that would, in effect, turn low Earth orbit into the world’s largest distributed AI data center. He unveiled a speculative rendering of a “mini” AI data center satellite, describing it as one node in a much larger system.

Why This Is a Massive Gamble Not Just a Factory

Let’s be direct about what makes Terafab different from, say, the CHIPS Act investments flowing to Intel, TSMC’s Arizona fabs, or Samsung’s Taylor, Texas facility. Those projects are being built by companies that have fabricated chips for decades. They have the process engineers, the equipment operators, the yield improvement culture the institutional knowledge that takes a generation to accumulate.

Musk has none of that. Tesla builds cars and software. SpaceX builds rockets. xAI builds large language models. None of these companies have ever etched a transistor onto silicon at commercial scale. The semiconductor fabrication business is not like building cars or launching rockets it is arguably the most technically demanding manufacturing process in human history, with yield rates that can take years to optimize and equipment lead times measured in years, not months.

At full capacity, Terafab is targeting roughly one million wafer starts per month. For context, as Electrek reported, that would represent approximately 70% of TSMC’s entire current global output from a single facility that has never manufactured a chip. It is, on paper, an extraordinary overreach. The whole semiconductor industry has been humbled time and time again by companies that underestimated fab complexity, from IBM’s exit to GlobalFoundries’ struggles to Intel’s years-long delay getting its own advanced nodes competitive.

What’s notable here is that Musk has actually acknowledged the staged approach. He explicitly said Terafab will begin with a smaller advanced technology fab before scaling. That’s smarter than it sounds it buys the teams time to learn fabrication basics without betting everything on a facility that doesn’t exist yet. But the gap between “learn to make chips” and “produce 70% of TSMC’s global output” is not a gap you close in a year or two.

The Consolidation Narrative Why Tesla, SpaceX, and xAI Are Merging in Everything But Name

Terafab didn’t arrive in a vacuum. It’s the latest and largest step in a slow-motion convergence of Musk’s three major companies that has been accelerating throughout 2025 and into 2026. Consider the timeline: in January 2026, Tesla announced a $2 billion investment in xAI alongside a formal framework agreement for cross-company collaboration. In February 2026, SpaceX acquired xAI outright in an all-stock deal, making xAI a wholly owned SpaceX subsidiary. Tesla has already integrated xAI’s Grok chatbot into some of its vehicle infotainment systems and sells Megapack battery storage units to xAI for its data centers.

Now Terafab adds a shared chip manufacturing layer to this stack. In effect: Tesla builds the robots and cars, SpaceX launches the satellites, xAI runs the models, and Terafab feeds all of them silicon. If it works, it’s a vertically integrated AI hardware stack that no other company in the world not even Google or Amazon can fully match, because no one else owns a rocket company, a humanoid robot program, an AI lab, and a chip fab simultaneously.

This is also why some analysts are reading Terafab as a strategic narrative play as much as an operational one. As we covered when Jeff Bezos announced his $100 billion AI manufacturing fund, the billionaire AI infrastructure race is intensifying, and announcements like Terafab are partly about signaling positioning in that race particularly as SpaceX’s IPO looms at a potential $1.5–1.75 trillion valuation.

What This Means for the US Chip Landscape

If even a fraction of Terafab’s ambitions are realized, the geopolitical implications are significant. The US semiconductor industry has been on a forced-march rebuild since the CHIPS and Science Act passed in 2022, pouring roughly $52 billion into domestic fab expansion. But most of that money has gone to established players Intel’s fabs in Ohio and Arizona, TSMC’s massive Phoenix facility, Samsung’s Taylor campus. Terafab would, if operational, be a new player in a very small club.

The Austin angle specifically matters. Texas has been making a concerted push to become a semiconductor hub Samsung’s $17 billion Taylor facility, which broke ground in 2022, is already in the area, and Musk reportedly already has a chip supply agreement with that Samsung plant for upcoming Tesla products. Having Terafab on the same regional grid would create a meaningful concentration of chip manufacturing talent and infrastructure in the state.

Texas Governor Greg Abbott, who was present in the audience at Saturday’s event and thanked on stage by Musk, clearly sees the political win here. Abbott posted on X afterward that Musk’s vision is “powerful” and expressed Texas pride. The state has historically moved quickly to offer economic incentives to large manufacturers though, notably, Tesla’s relationship with Travis County over its Gigafactory has not been entirely smooth, with county officials reportedly still awaiting documentation on parts of their existing agreement.

The Optimus Connection Robots Building the Robot Factory

Perhaps the single most audacious claim in Saturday’s presentation was Musk’s assertion that Optimus humanoid robots would help build and operate Terafab itself. Tesla has been publicly developing Optimus since 2021, and the robots have been deployed inside Tesla’s own factories in limited capacities during 2025. Musk’s pitch is essentially a closed loop: Terafab produces chips for Optimus, and Optimus builds more Terafabs.

This matters beyond the theater of it. If Tesla can demonstrate Optimus performing meaningful semiconductor fab tasks even auxiliary ones like materials handling, quality inspection, or clean room logistics it would be a landmark moment for industrial robotics. Semiconductor fabs are among the most demanding manufacturing environments on the planet, with extremely tight contamination controls and process precision that makes them challenging even for specialized human operators.

As we covered in our earlier reporting on Tesla’s Optimus and Grok integration, the robot has been gaining new AI capabilities at a steady clip. Terafab would be its most demanding real-world deployment yet and a compelling proof of concept for the entire humanoid robotics industry.

The current output of AI compute is roughly 20 gigawatts per year. All of the rest of the output from Earth is about 2% of what we need Elon Musk, TERAFAB Launch, Austin, March 22, 2026

What Comes Next and What to Watch

The immediate next steps are going to be telling. Terafab needs to file for construction permits that means actual dates and facility specs becoming public record in Travis County. It needs to acquire lithography equipment, which means orders from ASML, the Dutch company that supplies the extreme ultraviolet (EUV) machines required for sub-5nm chip production. Lead times for ASML’s most advanced EUV tools run well over a year, so order placement timing will be one of the first verifiable signals of whether this is on a real track or still a roadshow.

It also needs to hire. Semiconductor process engineers don’t grow on trees they come from Intel, TSMC, Samsung, and the handful of universities with serious fab programs. Multiple job listings have already appeared under the Terafab name, concentrated in Austin and Palo Alto. What those listings say about the role profiles and experience levels required will tell the industry a lot about where Terafab is in the actual technology development stack.

The AI compute race is accelerating, and billionaires are not waiting for existing industry players to catch up. We’ve already reported on how OpenAI is consolidating its products into a single desktop superapp that kind of software-layer consolidation is happening in parallel with this hardware-layer consolidation, and the two trends reinforce each other. More compute enables more capable models. More capable models justify more compute investment. Terafab is Musk’s answer to that loop a bet that the only way to guarantee he doesn’t fall behind is to control the entire silicon supply chain himself.

Whether that bet pays off depends on execution, not vision. And execution is exactly where Musk’s track record is most complicated. Keep watching Austin.