Intel Continues Terafab Work Amid TSMC Talks

Elon Musk answered a screenshot on X that Saturday. The image showed an article headlined that TSMC was exploring plans to work with Terafab. His reply was public and short. “Just discussions, but something may come of it.”

Until that post, markets had treated the manufacturing side of the venture as Intel-linked. Intel was the first foundry name attached to the project. Now Taiwan’s dominant contract chipmaker sat in the same frame. TSMC is the world’s largest dedicated foundry, the company that fabricates advanced processors for design firms across the industry and holds the bulk of global foundry market share. Its name beside a project built around SpaceX, Tesla and xAI changed what outsiders could read of the arrangement in a single weekend.

Musk confirmed the talks and stopped there. No structure was given. No timeline was given. The seat that had looked settled was suddenly open to another reading.

Intel stock fell 2.65 percent the session after the talk became public. It fell another 3.2 percent the following day. Across two sessions the decline reached 5.7 percent. TSMC rose 2.7 percent and closed at a record $485.80 a share.

The swings pointed at an unfinished rebuild. Intel Foundry recorded a $10.32 billion operating loss in 2025 even as revenue increased 3 percent. That figure captured the heavy cost of restoring advanced manufacturing strength and turning the foundry arm into a competitive contract business. Capital had flowed into next-generation process technology for years in the hope of winning outside work at the leading edge. The operating results still ran deep in the red. Progress stayed expensive, and the turnaround remained unproven.

Terafab had briefly changed the story. It supplied the one high-profile win investors could name—a project of real ambition that chose Intel’s manufacturing technology and gave markets a concrete test case against the losses. Talks that placed TSMC in the same conversation put the value of that win under fresh pressure. A crowded partner list left the original seat looking less certain.

In Tokyo on Wednesday, Lip-Bu Tan told Bloomberg reporters that Intel would remain involved in Terafab. He spoke ahead of an event marking the company’s 50th year of operations in Japan. Intel shares rose 2.18 percent in premarket trading after his comments, reaching $114.95.

Musk rejected the idea that TSMC would own or run the fab. “No, we will build and run the fab. Let there be ZERO doubt about that.” He left only one narrow opening: “Maybe TSMC subleases part of the Terafab if they want, but nothing more than that.”

Intel stayed on the project. Musk’s companies would build the fab and run it. TSMC’s ceiling, if it entered at all, was a sublease.

The main site sits in Grimes County, Texas. Terafab targets 1 terawatt of compute annually—a measure of yearly processing capacity meant to feed artificial-intelligence systems and robotics across Musk’s companies. The initial investment exceeds $16.8 billion. Total spending across phases could reach $119 billion.

Two factories are planned. One will produce the chips that go into vehicles and humanoid robots. The other will produce the chips that power AI data centers designed to operate in space. A workforce of at least 3,000 people is planned.

Intel joined on April 7, 2026, as the first named foundry partner, announced alongside SpaceX, xAI and Tesla. Later that same month, Tesla’s first-quarter 2026 earnings call indicated plans to use Intel’s 14A process. A process node is the manufacturing recipe that determines how densely transistors are packed onto a silicon wafer and how the finished chips perform. 14A is Intel’s next-generation node, the technology the foundry has been building to sell to outside customers at the leading edge. Tan had described the pre-announcement engagement with Musk as “highly strategic.”

Intel works with Musk’s team every week. The partnership is already a working routine of know-how moving back and forth.

Tan puts that routine inside a wider claim about how chips should be made. At the Splunk .conf26 keynote he said, “We should not be 95% dependent on one company based in Taiwan, and I think that is a very important issue.” He paired it with another line used the same way: “Everything is moving more and more toward a systems approach and packaging, and that is the future.” Advanced production, in his telling, is shifting toward whole systems and the packaging that binds them, and concentration of that work in one place has become the problem to solve.

SpaceX proposed spending $55 billion to begin construction, according to a Grimes County public notice. That figure sits above the initial outlay already disclosed and shows how much money is being lined up while the weekly process work continues and the first silicon remains years away.

Terafab’s first products are due mid-2028. More than eighteen months separate that date from the present. In the interval capital must still be spent, process work must still reach production, and the buildings must still go up. Sublease talks with TSMC remain unresolved. The 14A ramp has not yet produced finished chips.

Lip-Bu Tan stated what the interval requires. “It is not easy, it requires a very careful approach, and it requires a lot of capital expenditure... The good news is that things have improved over the past 18 months.”