Elon Musk’s “Terafab” in Austin aims to build chips for Tesla and SpaceX, starting in 2027

Elon Musk is betting billions that the next bottleneck in cars, robots, and space isn’t steel or software, it’s silicon.

At a site just north of Tesla’s massive Gigafactory Texas outside Austin, Musk has unveiled “Terafab,” a proposed semiconductor complex designed to pull chip design, manufacturing, and advanced packaging under one roof for Tesla, SpaceX, and his AI company xAI. The venture is pegged at an initial $20 billion to $25 billion investment, with first chip production targeted for late 2027.

The pitch is classic Musk: go vertical, move faster, and stop waiting on outside suppliers. But building a chip fab is one of the hardest industrial projects on Earth, and the timeline leaves little room for mistakes.

A three-company joint venture with one goal: control the chips

Terafab is structured as a joint venture among Tesla, SpaceX, and xAI, an unusual corporate mash-up that signals how central chips have become across Musk’s empire. Tesla needs silicon for driver-assistance and robotics. SpaceX needs hardened electronics that can survive radiation and brutal temperature swings in orbit. xAI needs compute, lots of it.

The plan is to create an “integrated fab” on Tesla’s North Campus near Gigafactory Texas, consolidating steps that are typically spread across multiple vendors and locations: chip design, fabrication, memory production, advanced packaging, and testing. Musk’s argument is that tighter integration cuts logistics delays and speeds iteration between engineers and the factory floor.

Musk has also talked up the sheer scale. He’s suggested the Terafab building could surpass Gigafactory Texas, already one of the world’s largest factories, and claimed it could be more than 12 times the size of Samsung’s chip plant in Taylor, Texas, a local reference point for just how big this could get. Early on, though, part of the site may be used for a smaller design-focused facility.

Two chip lines: Tesla “edge” AI and SpaceX space-hardened compute

Terafab isn’t being sold as a general-purpose chip factory. It’s being framed around two specific chip families.

First: an “edge” inference processor optimized for onboard AI, aimed at Tesla’s Full Self-Driving system, the Optimus humanoid robot program, and any future robotaxi fleet. The idea is to run more AI directly in the vehicle or robot, reducing dependence on constant back-and-forth with data centers while keeping power use under control.

Second: a higher-power, space-hardened variant intended for SpaceX, especially satellites and concepts that amount to orbital computing infrastructure. In space, raw performance is only part of the equation. Radiation tolerance, fault resistance, and validation testing can dominate both cost and schedule.

One Texas-based semiconductor engineer quoted in the report put it bluntly: if you control design and packaging, you can optimize for your real-world use case, not a generic market. That’s the promise. The risk is complexity: launching two product lines at once can slow ramp-up at a fab, where early success often depends on stability, yield, and repeatability.

$20 billion to $25 billion, and a big question about who pays

The price tag puts Terafab among the most expensive industrial bets in the U.S. A modern semiconductor fab isn’t just a building, it’s a clean-room ecosystem packed with specialized tools, chemical systems, water management, and quality control infrastructure. Even with “everything under one roof” messaging, each step requires its own expensive equipment and expertise.

One detail investors will zero in on: Musk has said Tesla’s 2026 capital expenditures won’t include Terafab. That suggests the funding may be handled separately through the joint venture or other entities, leaving open questions about cost-sharing and how the spending hits each company’s books.

That matters because fabs can burn cash for years before they reliably ship high volumes of qualified chips. Wall Street loves bold narratives, right up until timelines slip and budgets balloon.

Musk’s eye-popping “1 terawatt of compute” claim, and why it’s controversial

Musk has attached a headline-grabbing metric to Terafab: a goal of 1 terawatt of compute capacity per year, with some statements suggesting 100 to 200 gigawatts annually “on Earth” and a terawatt “in space.” He’s also suggested roughly 80% of output could be dedicated to space-based AI computing, putting orbital infrastructure at the center of the plan, not as a side project.

Tesla has even implied that this would exceed the combined capacity of chipmakers today, and possibly even by 2030. Read that less like a spec sheet and more like a moonshot target meant to signal ambition and attract talent, partners, and capital.

But “compute” measured in power terms can blur critical details: chip efficiency, cooling, packaging, uptime, and the practical ability to feed systems with electricity and manage heat. In orbit, shedding heat is notoriously difficult. On the ground, power and water use are increasingly political, and expensive.

Austin hiring and today’s reliance on outside suppliers like Samsung

Terafab is also a response to a basic reality: Musk’s companies still depend on external chip suppliers, including Samsung, for key components. That’s normal in autos and aerospace. It becomes a strategic problem when you want faster hardware refresh cycles, or when you’re trying to reserve massive capacity in a market where leading-edge production is perpetually constrained.

On the ground in Austin, the project is already leaving footprints. Local observers have reported construction activity north of the Tesla plant, and job postings tied to engineering and supply chain roles suggest this is more than a stage announcement. Companies don’t hire supply chain teams unless they plan to buy equipment, manage vendors, and build an industrial organization.

For Tesla, the upside is tighter alignment between software and silicon, and less exposure to shortages. For SpaceX, it’s about availability and reliability under punishing qualification requirements. For xAI, it’s about feeding models that demand ever more compute. Putting all three under one roof also creates a new internal tension: when capacity is limited, who gets priority?

If Terafab works, it could give Musk a rare advantage: custom chips built on his schedule for his machines, from cars to satellites. If it stumbles, it risks becoming a single point of failure, an expensive one, at the center of three of the most ambitious hardware programs in America.