TSMC Terafab Talks: Who Would Run Musk’s Chip Fab?
TSMC is discussing Musk’s Terafab, but says it must operate any fab. See what the dispute means for Texas chipmaking, capacity, and the project’s next steps.
Musk's Terafab Has a Critical Condition: TSMC Wants to Run It
A new semiconductor fab isn't just a building stuffed with expensive machines. It's a whole system - people, processes, suppliers, and daily operating decisions - that has to line up if the chip output is going to match what customers actually need. So in the discussions between TSMC and the team behind Elon Musk's Terafab, the real question isn't whether the companies will "work together." It's who will run the factory.
TSMC chairman and chief executive C.C. Wei says the chipmaker is talking with the customer building Terafab, but any fab would have to be operated by TSMC. Musk, meanwhile, has described a different plan: Tesla and SpaceX would build and operate the plant themselves, with TSMC potentially taking a limited role. No agreement has been announced.
That disagreement matters because Terafab is meant to supply chips for Musk's cars, robots, and data centres. It also sits inside a broader push to expand US semiconductor capacity. Here's what's been confirmed, why operating control is such a big deal, and what to watch as talks continue. For the original reporting, see The Next Web's coverage of the TSMC-Terafab discussions.
What TSMC and Musk have said
TSMC says it's open to working with the companies behind Terafab, but Wei was clear about one condition: the fab would need to be run by TSMC. He said this during the company's third-quarter earnings call, after JPMorgan analyst Gokul Hariharan asked about a potential group of TSMC facilities in Texas and whether Tesla or SpaceX were part of any talks.
Wei said large US customers have asked TSMC for more capacity. TSMC responded by sending teams to Texas to look for a site beyond its existing base in Arizona. He also confirmed TSMC was in discussions with the customer building Terafab about how that capacity might be used - but he didn't say a deal had been reached or give a timeline.
When asked if a possible arrangement would resemble TSMC's joint ventures in Japan and Germany - where TSMC operates the facilities - Wei didn't hedge: "We are still in discussion. But the fab got to be run by TSMC." In other words, it's a requirement, not a settled operating plan.
Musk's preference runs the other way. On X on 7 October, he said Tesla and SpaceX would build and operate the plant themselves. He left open the idea that TSMC could sublease part of Terafab, but he rejected a bigger operating role for the chipmaker.
That's the gap right now. One side is describing a customer-led facility where TSMC might have limited access. The other is saying that if TSMC is involved, it must control how the fab is run. Until that gap is resolved, it's fair to treat the "cooperation" talk as exploratory rather than a confirmed production partnership.
What is known about the project
Terafab is described as a joint chipmaking effort involving Tesla, SpaceX, and xAI. Its first phase is said to be a $16.8 billion fab in Grimes County, Texas, with construction beginning in August. The plan is for the project to supply chips for the companies' vehicles, robots, and data centres.
Intel is also part of the picture. Bloomberg reported that Intel chief executive Lip-Bu Tan said Intel would remain involved and is expected to supply its 14A manufacturing process. That said, Intel's planned process contribution doesn't, by itself, confirm that TSMC has agreed to operate Terafab - or that the parties have agreed on who controls the facility.
Those distinctions matter. Having a named participant, stating an intention, contributing a manufacturing process, and signing an operating deal are not the same thing. At this stage, the public record points to ongoing discussions and a project in development - not a completed plan for how the factory will be run.
Why operating control matters
In everyday language, "run" can sound like a management preference. In semiconductor manufacturing, it's closer to a day-to-day operating question. A fab has to coordinate manufacturing processes, equipment, materials, staffing, maintenance, quality checks, and production schedules. If those systems aren't managed as one integrated operation, the factory's capacity on paper doesn't reliably translate into enough chips delivered at the required quality.
That's why TSMC - or any company with deep manufacturing experience - would treat operational authority as practical, not symbolic. Wei's comments suggest TSMC isn't describing a model where it advises a customer while someone else makes the key operating decisions. The public comments don't lay out every reason behind that position, so it wouldn't be wise to assume a specific contract structure. Still, the principle is straightforward: TSMC says any fab it's involved in needs to be run by TSMC.
For Musk's companies, operating control could offer their own benefits. A customer-operated facility could give them more direct influence over priorities, how capacity is allocated, and how chip supply lines up with their product plans. Tesla, SpaceX, and xAI have different demand profiles - vehicles, robotics, and data-centre workloads aren't interchangeable. Bringing some manufacturing closer to those businesses could, in theory, help them steer planning more directly.
The trade-off is that owning or directing a factory doesn't automatically create the know-how needed to make it perform. A fab isn't a conventional factory where adding another shift reliably boosts output. Manufacturing depends on process control and repeatability across lots of interconnected steps. Building the site is only one milestone; operating it consistently is the harder challenge.
Control, accountability, and risk
A real partnership has to answer a basic question: who's responsible when production misses its targets? If one company owns the facility, another provides process expertise, and a third sets the expected demand, responsibilities can blur - unless roles are clearly defined. It becomes messy fast when each party expects a different level of control.
An operator usually needs authority that matches its responsibility. If TSMC is expected to stand behind manufacturing performance, it's reasonable for it to want operational control. If Musk's companies want operational control, they may prefer an arrangement that limits TSMC's authority to a narrower role. Those goals aren't impossible to reconcile - but they need to be spelled out in an agreement that hasn't been publicly described.
The reporting so far leaves a few plausible outcomes on the table, none confirmed:
- TSMC operates the fab or a defined part of the site. This fits Wei's requirement, but it would mean Musk's companies accepting an operating role for TSMC.
- TSMC takes a limited or separate role. Musk has floated a sublease idea, but it's unclear whether TSMC would accept that as sufficient.
- The talks don't produce an agreement. In that case, the project could still move forward, but the information available doesn't say what the alternative operating plan would be.
The point isn't to guess which scenario is most likely. It's to recognize that the operating model is core commercial and technical language - not something you can "figure out later" once construction is underway.
Terafab in the wider chipmaking landscape
These discussions are happening while TSMC is looking at ways to add capacity in Texas. Wei said teams went beyond TSMC's Arizona base after US customers asked for more capacity. That shows site exploration and demand pressure - it doesn't prove that a Texas campus has been approved, funded, or specifically tied to Terafab.
A Texas location could matter to customers who want more US-based manufacturing. But location alone doesn't solve the real business problem. A new fab still needs a defined operator, a production plan, qualified processes, and customers whose demand justifies the investment. Putting those pieces together takes time. The fact that customers want capacity doesn't automatically say how much can be delivered, when it would be available, or which facility would supply it.
TSMC's earnings give some context for the company's scale, but they aren't evidence that a Terafab deal is close. TSMC reported third-quarter revenue of $46.7 billion, up 41% year over year, and forecast fourth-quarter revenue of $52.3 billion to $54.3 billion. Those numbers point to strong demand and a large operating business. They don't disclose the terms or status of Terafab talks.
Intel's involvement adds another layer. Tan said Intel would stay on the project and is expected to supply its 14A process. That suggests Terafab isn't being described publicly as a simple, single-company manufacturing setup. Still, the source reporting doesn't explain how Intel's role would fit alongside TSMC's involvement. It's safer to treat the companies' responsibilities as unresolved than to infer a complete technical plan from brief comments.
Why the US location is not the whole story
US semiconductor investment often gets talked about in terms of where factories are built. For customers, the more important questions are whether the facility can deliver the right chips at the right volumes, on a schedule that fits product plans - and who carries the risk if it can't.
Even if a facility is in your backyard, manufacturing capacity isn't suddenly interchangeable. A fab's output depends on its processes, equipment, workforce, and operating model. The investment case also depends on sustained demand. So a planned fab is not the same thing as available supply.
For Terafab, that distinction matters because the project is described as serving multiple businesses with different products. Demand from cars, robots, and data centres might sound like a broad base, but each category has its own requirements and timing. The public reporting doesn't provide a detailed capacity plan or production schedule, so there's no solid basis for estimating how much of each company's needs the project could cover.
What to watch next
The next meaningful update won't be another generic statement that the companies are talking. What would clarify things is evidence that they've agreed on responsibilities. Several specific details would help show whether this is moving toward an operating partnership - or staying in exploratory territory.
1. A defined operating structure
Watch for a clear statement about who operates the fab, and whether that applies to the whole facility or only part of it. Musk has described a model where his companies operate the plant, while Wei says TSMC must run any fab in which it participates. A concrete arrangement would show whether there's a middle ground or whether one position has shifted.
2. A formal agreement and scope
The companies have not announced an agreement yet. If one is signed, the scope will matter - whether TSMC is providing manufacturing capacity, operating the facility, taking a lease, or participating in some other way. Calling it a "partnership" won't be enough. The real question is how investment, control, and responsibility are divided.
3. A credible production timeline
Wei didn't provide a timeline for talks. Any schedule that appears later should be treated carefully, especially if it mixes construction milestones with when production actually starts. A building can be under construction while the operating model, production readiness, or customer allocation are still unsettled.
4. Clarity on Intel's role
Intel's expected 14A contribution is a significant piece, but public comments haven't explained how that role connects to TSMC's discussion. Look for information on which company is responsible for which part of manufacturing, rather than assuming Intel's participation automatically means a shared operating plan.
5. Evidence of capacity decisions
TSMC has said it is exploring a possible Texas site in response to customer demand. Site search is an early step. Confirmation of investment, a chosen location, or a capacity plan would be a more concrete signal. Until then, the fact that discussions exist shouldn't be treated as proof that a new TSMC Texas campus will supply Terafab.
If you're trying to keep track, a simple way to filter the noise is to separate announced facts, stated intentions, and open questions. The fab's stated cost and location, the companies' public positions, and TSMC's site exploration are reported facts or statements. The status of a completed TSMC operating deal, a production schedule, and a clear division of work are still open questions.
Conclusion
Terafab's biggest near-term question isn't whether TSMC sees an opportunity in Texas. It does. The real question is whether the companies can agree on who controls the factory - and who takes responsibility for how it operates. TSMC says it must run any fab it is involved in; Musk says his companies intend to run Terafab themselves. No agreement has been announced.
Until that changes, it's worth treating the project as a significant plan with unresolved operating terms - not as confirmed new chip supply. The next signals to look for are a formal agreement, a clear division of responsibilities, and a production timeline. For now, the best move is to follow what the companies actually announce, and separate confirmed commitments from stated ambitions.