Terafab Targets 2028 Chip Production as Suppliers Accelerate Equipment Shipments

| 2 Min Read
ASML and other suppliers are speeding Terafab deliveries as Musk targets volume production in 2028 and outlines a 13 GW AI compute buildout.

Terafab’s schedule depends on suppliers

Terafab’s schedule now depends on whether a crowded semiconductor supply chain can deliver equipment fast enough. Elon Musk says ASML and other suppliers are accelerating shipments for the Tesla and SpaceX chip project, which aims to begin volume production in Texas in late 2028. He cautioned that the target is highly aspirational, with “big error bars.” That qualification matters: faster deliveries can ease one constraint, but they don’t establish that the full manufacturing line will be ready on schedule. A fab needs many tools and processes to work together, so progress on individual shipments won’t by itself settle the larger question.

Musk discussed the schedule during SpaceX’s Q3 2026 earnings call, broadcast live on X on October 20. SpaceX expects to produce its first experimental test chip at a research fab in Austin around the second quarter of 2027. The primary Terafab facility is planned for Gibbons Creek, Texas. The Austin work gives the project an earlier checkpoint, but an experimental chip and volume production are different tests: one can show that parts of the process work, while the other depends on a coordinated manufacturing operation. If you're working in this space, the gap between those milestones is where much of the execution risk sits.

A broad equipment push

ASML is central to the plan: it is the only commercial supplier of the extreme ultraviolet lithography systems used for high-volume production of leading-edge chips. Each system costs about $400 million. Unofficial reports had placed Terafab’s first EUV tool delivery in mid-2027; Musk’s remarks suggest some deliveries may be moving faster, though he didn’t identify which equipment or shipment dates had changed. That leaves a key detail unresolved: an accelerated shipment could refer to a tool needed early in the build, or to equipment with a narrower role. Without a revised delivery schedule, the remarks signal movement but don’t make the timeline easier to assess.

The supplier roster extends well beyond lithography. Applied Materials, Lam Research, KLA Corporation, and Japan’s Tokyo Electron are expected to provide equipment for wafer processing, measurement, and process control. Tesla and SpaceX are also in discussions with roughly 10 South Korean equipment makers, including advanced-packaging supplier Hanmi Semiconductor and high-pressure hydrogen annealing company HPSP, alongside vendors of specialized piping and process valves. That breadth reflects how many distinct systems a semiconductor fab requires. Lithography draws attention because of the scarcity and price of EUV tools, but wafer handling, process measurement, and supporting infrastructure also have to fit the manufacturing flow. A delay in one part can limit the value of equipment already delivered elsewhere.

Intel joined the project in April 2026 to help develop fabrication technology around its 14A process. Terafab is planned as a vertically integrated operation covering logic, memory, and advanced packaging. Its stated goal is to produce enough chips for 1 terawatt of AI computing power annually, with traditional EUV lithography in the manufacturing flow. Bringing these functions under one project could give Tesla and SpaceX more control over how chips are designed and assembled, but it also increases the number of technical tasks Terafab must coordinate. The ambition is broad. The schedule has to account for that breadth.

Compute targets add pressure

The fab is part of a larger effort to secure processors for xAI clusters, Tesla’s self-driving vehicles and Optimus robots, SpaceX’s proposed orbital AI data center, and other customers. Musk has framed in-house production as a way to reduce dependence on external foundries such as TSMC, where access to advanced manufacturing capacity is constrained. That makes Terafab a supply strategy as well as a factory project: the intended output is tied to several businesses with different computing needs. Yet building a source of chips takes time, and the project’s near-term plans still depend on external equipment suppliers. Here’s the thing: reducing reliance on foundries doesn’t remove reliance on the rest of the semiconductor supply chain.

On the same call, Musk said his AI companies expect to have 2.5 to 3 gigawatts of computing capacity operating by the end of 2026. He expects another 10 GW to come online during 2027 and said he would be surprised if total capacity fell short of 13 GW by year-end. SpaceX and xAI are also procuring Nvidia’s next-generation Vera Rubin hardware, with plans for a 440,000-GPU training cluster to be online by the end of the second quarter of 2027. Those plans show why the chip supply question is tied to a wider infrastructure buildout: processors need data-center capacity around them, and both have to arrive in time to be useful. The scale of the targets is striking, but the call’s forecast doesn’t provide a detailed delivery schedule for every part of that buildout.

That scale would put Musk’s target ahead of several announced data-center projects. The restructured Project Stargate is planned for 10 GW, while Meta’s Hyperion is planned for 5 GW; both have timelines extending into the 2030s. Those comparisons underline how much infrastructure the 2027 compute forecast would require. They also need to be read carefully: announced capacity targets and operating capacity aren’t the same thing. A forecast of 13 GW by year-end is a claim about what will be running, while project plans can describe capacity expected on a different timetable. The comparison conveys the ambition, but it doesn’t verify delivery.

The scale remains unproven

Musk says Terafab’s eventual chip output could have commercial value comparable to the GDP of a top-20 country, a claim he illustrated with Switzerland’s roughly $1.14 trillion GDP. The project’s 1 TW annual capacity target is far larger than today’s installed AI data-center base: Gartner estimates global AI-optimized data-center capacity at about 44 GW in late 2026. EpochAI has reported annual growth of roughly 3.3 to 3.4 times since 2022, but that pace alone doesn’t establish that Terafab’s production target is achievable. These figures describe different things: a comparison of chip output’s possible commercial value, a manufacturing capacity goal, and the capacity of data centers. Put side by side, they communicate the scale of Musk’s ambition, not a direct measure of whether the project can meet it.

The numbers here are underwhelming as evidence of execution, even if they’re impressive as a statement of intent. A large market opportunity and rapid growth in AI infrastructure can help explain why Musk wants a dedicated source of chips. They don’t show that Terafab can produce the required volume, or that customers will be ready to use it as planned. That distinction is easy to lose when a project’s manufacturing target is described through the computing power it could support. The goal is clear; the path from equipment deliveries to sustained output is still largely ahead.

What the next milestones can show

Supplier acceleration is an early sign of execution, not proof that the production schedule will hold. The next meaningful milestones are the Austin test chip and delivery of the first EUV system; both will show whether Terafab can turn its ambitious capacity plans into a working manufacturing operation. The test chip can provide evidence that the research work is advancing, while the EUV delivery can indicate whether a central tool is arriving in time for the planned development. Neither alone would establish that the Texas facility is ready for volume production. And yet, each would give observers something firmer to evaluate than a broad target.

Terafab’s significance will depend on whether it can connect chip production to the compute plans that are supposed to justify it. If you're tracking the project, watch for specifics about which shipments are accelerating and what the Austin test chip demonstrates. The details matter because a schedule built around many suppliers can move unevenly: one tool may arrive early while another process remains unfinished. (And this is the part most people overlook.) An earlier delivery is useful only if it advances the work that follows.

For now, the project has a stated production goal, a planned site, an equipment roster, and an interim research milestone. What it doesn’t yet have, based on the information in the call, is evidence that the full chain can meet the late-2028 target. Ambitious, and still unproven.

Source: Etiido Uko · www.tomshardware.com

Comments

Please sign in to comment.