Chip ManufacturingLOT 261007 · 3 min read
Musk: His Team Will Build and Run the Terafab, Not TSMC
Elon Musk says his own team will build and operate the planned Terafab, explicitly ruling out TSMC — but no capex, node, capacity or timeline figures accompany the claim.
Key takeaways
- Musk stated his own team will build and operate the Terafab, not TSMC.
- No capex, process node, wafer size, capacity or timeline figures accompany the announcement.
- The statement distances the Musk-affiliated mega-fab from the foundry that produces most leading-edge AI silicon.
- TSMC has issued no on-record response to the claim.

Elon Musk says his own team will build and operate the planned "Terafab," explicitly ruling out TSMC as the constructor or operator of the facility. The statement, reported by Briefs Finance, puts the Tesla and xAI chief's most ambitious manufacturing claim to date on record: the world's largest contract chipmaker is not part of the execution plan.
The declaration is short on engineering detail but long on implication. Musk's public positioning now assigns responsibility for fab construction, ramp and day-to-day operations to an in-house organization rather than to the foundry ecosystem that every major AI accelerator vendor today depends on — TSMC fabricates the overwhelming share of leading-edge AI training silicon on its 4nm- and 3nm-class processes, including chips designed by Nvidia, AMD and, historically, Tesla's own FSD hardware.
What did Musk actually commit to?
According to the report, Musk said only that his team — not TSMC — would build and operate the Terafab. The claim contains:
- No confirmed capex figure for construction or equipment.
- No process node, wafer size or monthly capacity target.
- No site, groundbreaking date or volume-production timeline.
- No clarification of whether "build" covers the shell and cleanroom or extends to process technology itself.
That last distinction matters commercially. Contracting a fab shell is routine; developing and operating a leading-edge logic process is something only a handful of entities — TSMC, Samsung Foundry, Intel and China's SMIC, under sanctions constraints — have done at scale. Musk's phrasing leaves open which of those two very different undertakings he means.
Why the TSMC distinction carries weight
Musk framed the statement as a correction of an assumption: that any Musk-affiliated mega-fab would naturally be a TSMC-built, TSMC-run facility. He said it will not be. The remark lands against a market in which AI compute demand has made foundry capacity a strategic bottleneck, and in which customers from Apple to OpenAI have explored deeper — and in some cases equity-based — ties to TSMC to secure wafers.
Going it alone would invert that logic. Instead of buying guaranteed capacity from the most advanced manufacturer on earth, Musk's stated approach would absorb the cost, yield-learning curve and talent war that in-house fabrication entails. Industry executives and analysts routinely price a single leading-edge gigafab above $20 billion; no such figure has been attached to the Terafab in Musk's statement, and Briefs Finance reports none.
What remains unconfirmed
Every hard parameter of the project is still open. Musk offered:
- No statement on whether the Terafab targets logic, memory or advanced packaging — the latter being a far lower bar than full leading-edge lithography and a path others have taken first.
- No indication of partners for lithography tools, without which no leading-edge fab proceeds; ASML remains the sole EUV supplier.
- No response from TSMC on the record.
Until Musk's team publishes capacity, node or spending figures, the Terafab exists as a claim about who runs it — not yet as a fab with a schedule. Whether in-house execution produces a functioning plant or a revised roadmap with foundry partners reattached will define how seriously competitors and suppliers treat the project over the coming quarters.
Google News: TSMC (Source)
More from Nathan Brooks
Show full bio
Senior reporter covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,201 articles