Tesla and SpaceX's Terafab chip-fab project is reportedly in talks with about ten South Korean materials, components and equipment makers, and the companies reportedly do not do business with Chinese suppliers. That is the account from Korean trade outlet The Elec, which says representatives visited Korean booths at SEMICON West in San Francisco on October 13. It is an unusually detailed look at how a chip supply chain gets built from scratch, and it is also a report, not an announcement: neither company has published a supplier list. This post separates the two.
For background on the project itself, see our explainer on what Terafab is and the TSMC "operational control" claimed-versus-verified post.
Update (October 20, 2026): Musk's stated Terafab timeline
Posts on X from a Terafab presentation, as relayed by market-watching accounts, quote Musk saying the first working test chip is expected around May or June 2027 at a research chip fab at Giga Texas, and that volume production of Tesla and SpaceX "super intelligence" chips at the main Terafab site is targeted for the third or fourth quarter of 2028. The same posts repeat a $16.8 billion Grimes County, Texas investment, a long-term goal of 1 million wafer starts per month, and that Chinese suppliers are being excluded. We could not find a company filing or press release for these dates, so they are Musk's spoken targets, not commitments. For scale, TSMC's chief executive said earlier this month that a new wafer plant takes two to three years, which makes a 2028 volume date aggressive.
TL;DR: reported facts and their status
| Item | Reported | Status |
|---|---|---|
| Site and budget | Grimes County, Texas; initial $16.8 billion | Announced by SpaceX and Tesla on August 6, per press coverage |
| Capacity goals | 100,000 wafers per month initial; 1 million within ten years | Reported; initial site expected before 2030 per The Elec |
| Korean outreach | About 10 firms approached at SEMICON West on Oct 13 | The Elec |
| Signed NDAs | KoMiCo (MiCo's cleaning arm) and Unisem named; at least one other firm received an NDA | The Elec |
| Firm equipment commitment | HPSP: one high-pressure hydrogen annealing system for the R&D line | The Elec |
| Hanmi Semiconductor | Supply deal per KED Global; talks per The Elec | Conflicting |
| Chinese suppliers | Not used | Stated by The Elec; no company statement found |
| Intel and TSMC | Intel on logic; TSMC talks ongoing | Reported |
Who is reportedly in the room
According to The Elec, Terafab staff from Tesla and SpaceX visited about ten Korean booths on the first day of SEMICON West, which ran October 13 to 15. Some firms had been talking to them for months. The statuses it lists:
- HPSP: agreed to supply one high-pressure hydrogen annealing system for the R&D line. This is the only firm equipment commitment in the report.
- KoMiCo (MiCo's cleaning subsidiary): signed a three-party NDA with Tesla and SpaceX about a month earlier, discussing cleaning services for an R&D pilot line in Texas.
- Unisem: signed an NDA to discuss environmental equipment such as scrubbers.
- Dongjin Semichem: two to three months of talks, focused on its thinner plant in Texas, with phosphoric acid also under discussion.
- Hanmi Semiconductor: discussing thermocompression bonders after a sales executive visited SpaceX at its request. KED Global reported a supply deal in September, so the two outlets disagree on how far along it is.
- ENF Technology, Asflow and GnG Intech: first contact at the show, covering process materials, piping and valves, and wastewater recycling.
- BMT: talking valves and fluid components, and already a SpaceX piping supplier.
- ISC: exchanging samples and design information on silicone rubber test sockets.
The Elec also says many of these firms arrived via two routes: former Intel staff on the Terafab team brought in Intel's suppliers such as ENF, Asflow, ISC and HPSP, while GlobalFoundries procurement alumni brought others such as Dongjin Semichem. That is a useful detail: Terafab is not inventing a supply chain, it is importing the contact books of its hires.
Big-name partners
The same report says Fujifilm holds regular meetings with Terafab procurement on CMP slurries, photoresists, etchants and cleaning chemicals, and that Lam Research and Applied Materials have set up dedicated Terafab teams. Fujifilm is quoted as saying what it supplies "will depend on which technologies are selected," which tells you the process choices are not final. Intel is reported as the logic partner, with TSMC talks ongoing and a memory process under development. An overseas equipment supplier tells The Elec that Musk first considered buying fabs from Intel, TSMC and Micron before moving to partnerships after being turned down; that is one anonymous source.
The "no Chinese suppliers" claim
This is the line that travelled furthest on social media, and it deserves care. The Elec states that Tesla and SpaceX do not do business with Chinese suppliers. It then quotes a Korean equipment-maker official saying that "excluding China, there aren't many suppliers, and Japanese companies are expensive," and another industry official saying Terafab wants prices even lower than Chinese firms offer.
What we could not find: a Tesla, SpaceX or Terafab statement on supplier policy, or a regulatory filing that bars Chinese vendors. US export and procurement rules and political risk make avoiding Chinese suppliers plausible, and it matches the broader pattern in our AI chip supply-chain coverage. But "do not do business with" is a strong statement, and it comes from a trade report relaying unnamed industry sources. Treat it as likely direction, not confirmed policy.
The pricing demand is the more interesting claim. If accurate, Terafab wants non-Chinese suppliers at Chinese-level prices, which is hard to deliver and could limit who signs.
Speed as a stated differentiator
The Elec quotes an overseas supplier: "I've never seen this level of execution speed elsewhere in the semiconductor industry," and a Korean supplier says replies arrive within 30 to 60 minutes, even in early US mornings, because engineers have decision-making authority. These are supplier impressions, and positive ones from firms hoping for contracts. They suggest a fast, flat decision structure, but they say nothing about yield, schedule or cost, which are what decide whether a fab works.
How a fab supply chain is assembled, and why these categories
A semiconductor fab is not one purchase. It is hundreds of tool and consumable categories, each with a small number of qualified vendors, and every material has to be qualified against the specific process. The categories in the report show where a new entrant starts.
Process tools such as annealers and bonders determine what the line can physically do. HPSP's high-pressure hydrogen annealer is a niche tool used to improve device characteristics, and buying one for an R&D line is a way to learn the process before committing to volume. Thermocompression bonders, Hanmi's specialty, are central to stacking memory and advanced packaging, which fits Terafab's stated plan to put logic, memory, packaging and test under one roof.
Chemicals and gases are consumed continuously. Photoresists, etchants, CMP slurries and thinners are qualified lot by lot, and a plant near the fab, like Dongjin Semichem's Texas thinner facility, shortens logistics. Fujifilm's regular procurement meetings are about exactly this.
Facilities equipment such as scrubbers, wastewater recycling, valves and piping is unglamorous but gating: a fab that cannot treat its exhaust or water cannot run. The fact that a pilot line comes first, at Gigafactory Texas, lets Terafab test these suppliers at small scale.
Test hardware such as sockets sits at the end of the line. ISC's silicone rubber sockets are consumables sized to specific chip packages, so samples and design exchanges happen early.
This breadth explains why Terafab would work through Intel and GlobalFoundries alumni: a new fab cannot afford to discover its vendors one by one. It also explains the risk. Every category introduces a qualification cycle, and a vendor swap mid-build can reset months of work. That is why the reported demand for low prices and the reported exclusion of an entire geography, if both hold, make the schedule harder rather than easier.
What this means for AI hardware and for you
Three practical readings, none of which change what you pay today:
- Supply is the constraint. Our coverage of ASML's AI demand exceeding EUV supply and Michael Dell on a multi-year memory shortage show that AI demand is outrunning equipment and memory capacity. A new fab is a long-dated answer, not a near-term fix.
- Vertical integration is the bet. Terafab's chips are intended for Tesla's robots and cars and SpaceX's planned space data centers. Success would mean in-house silicon for those products; failure would leave them with the same TSMC and Samsung dependencies they have now.
- Timelines are aspirational. The $16.8 billion figure is only the first phase, with a reported $119 billion ceiling across all phases (see our Terafab explainer), and the initial 100,000-wafer facility is only expected before 2030. Do not plan around Terafab output in the next few years.
Questions people are asking
Is a fab really buildable by 2030? Industry fabs typically take years from groundbreaking to volume. The reports describe construction started and an R&D pilot fab at Gigafactory Texas, but no process node, yield or customer-qualified product has been shown.
Does Intel's involvement mean Intel 14A? Press coverage says Intel is slated to provide its 14A process technology for logic, but the contribution has not been spelled out by Intel in the sources we found.
Why Korea? Korean firms are strong in specific niches such as test sockets, cleaning, and certain bonders, many already supply Tesla or SpaceX, and Intel and GlobalFoundries alumni know them.
What is still unknown
- Any official supplier list or policy from Tesla, SpaceX or Terafab.
- Whether Hanmi has a signed contract.
- The corporate entity and ownership structure; the March announcement named Tesla, SpaceX and xAI without one.
- Process technology, yield targets and the timing of the memory process.
Related reading
- What is Terafab.ai?
- TSMC and Terafab: operational control claim vs verified
- AI chip supply chain: TSMC record, GlobalFoundries interposers
- ASML: AI demand exceeds EUV supply
- Michael Dell on the memory shortage
- Samsung HBM4 yield
Details are from reports dated October 2026 and may change as the companies comment.