TSMC In, Intel Maybe Out: The Patent Problem Behind Musk's Terafab Talks

Semiconductors 4 OCT 2026 1,465 words 7 min read By Tech Desk

TSMC In, Intel Maybe Out: The Patent Problem Behind Musk's Terafab Talks

Introduction

Elon Musk has confirmed that Taiwan Semiconductor Manufacturing Company is in discussions with Terafab, the vertically integrated chip plant his companies are building in Texas. That confirmation arrived on Saturday in a report by Tom's Hardware citing Tim Culpan's Culpium newsletter, and it was deliberately thin. Musk wrote only that it was "just discussions, but something may come of it." TSMC has said nothing at all.

On the surface this reads as a straightforward capacity story: the world's largest contract chipmaker might help build more American fabs, joining a trend this site has covered repeatedly, including the Texas fab cluster TSMC was reported to be evaluating at the end of September. Read that way it is not especially new.

It is not the right frame. The interesting part of the Terafab discussions is not that TSMC might add capacity in Texas. It is that a TSMC relationship could quietly remove the process technology Intel was brought in to supply, and that the reason sits in a licensing arrangement nobody has publicly discussed.

The Arrangement Everyone Is Describing

Culpan's account, as reported by TechPowerUp, describes two structures under discussion.

In the first, TSMC would own and operate a new fabrication plant, with SpaceX and Terafab contributing capital, committing to guaranteed chip volumes, or doing both. Culpan considers this the more probable outcome, and it has a clear precedent: TSMC's JASM venture in Japan and ESMC in Germany both pair local partners who supply capital and anchor demand with TSMC contributing process technology and running the fab.

In the second, SpaceX holds majority ownership of Terafab while TSMC invests less and supplies operational expertise and process technology. As several reports have noted, that structure would be new for TSMC, which has consistently kept operating control of the fabs it helps finance.

A third possibility deserves attention because it is cheap for TSMC and solves only part of the problem: TSMC participates in one stage of the chain rather than all of it. TSMC is already the leading supplier of advanced packaging, so Musk's interest in making chips in Texas and packaging them in Texas is not a strained interpretation.

Why Intel Is Suddenly Vulnerable

Intel joined Terafab in April 2026 as a manufacturing and process-technology partner rather than an equity owner, contributing its 14A process, a 1.4-nanometre-class node. Musk later said 14A would be used at full scale, on the argument that by the time Terafab ramps the process would be mature. Tesla became the node's first major customer at the Austin complex.

That arrangement is now the one thing TSMC's involvement would most directly disturb. If TSMC brings its own process node, there is no reason to use Intel's 14A, and 14A's most prominent customer at Terafab simply evaporates. This is not a technical judgment about which node is better. It is a substitution, and it happens at the moment of greatest strategic leverage for TSMC.

Intel's broader position at Terafab has never been secured by equity. SpaceX's May 2026 IPO registration described Terafab as being at a very early stage, with Tesla and SpaceX having agreed only to a general framework, and with financial terms, intellectual property rights, and Intel's and Tesla's continuing participation all still subject to later agreements. Nearly six months on, that language has not obviously improved.

The Part Nobody Is Saying Out Loud

Here is the constraint that connects the two companies, and it is the most concrete thing in the whole story.

Intel holds a broad cross-patent license agreement with TSMC. Terafab does not.

Produce chips on an Intel-developed node at a facility Terafab controls, and you are running processes developed by a company that does not license its patents to Terafab. Some of what that process does may fall inside patents TSMC holds. The exposure is real enough that keeping TSMC close becomes the obvious hedge, which is precisely what a TSMC joint venture accomplishes: an equity stake, a process license, and an operating relationship arrive together.

This reframes the entire negotiation. Musk's public framing has been about supply, and the supply argument is legitimate. But a foundry tie is also the cheapest available way to put a patent-owner's skin in the game, and it does so while appearing to be a capacity deal.

It also explains the sequencing. A TSMC relationship is easier to arrange before a fab is built than after, and Grimes County filings from September put the construction start at December 1, 2026 with completion targeted by the end of 2028. If a licensing structure is going to be negotiated at all, the window for negotiating it is open now.

The Vertical Integration Problem

Terafab's ambition is to put logic fabrication, memory, advanced packaging, testing, and mask production under one roof. As Wikipedia's summary of the project records, Musk pitched the site on March 21, 2026 at the defunct Seaholm Power Plant in Austin, describing a goal of more than one terawatt of AI compute capacity per year and stating that all current fabrication facilities on Earth produce roughly 2% of what Tesla and SpaceX will need.

That integration claim is where TSMC's contribution would be most valuable and most difficult, because advanced packaging is precisely the stage with the fewest substitutes.

One telling detail: Musk has said TSMC, Samsung, and Micron could keep supplying Terafab even as internal production develops. A project billed as vertical integration that relies on outside suppliers is better described as coordination around a demand centre. TSMC joining would not contradict that arrangement. It would formalise it.

What the Numbers Actually Say

The scale claims attached to Terafab need care. Musk's long-term goal is one million wafer starts per month and 100 to 200 billion custom AI and memory chips annually. The facility is planned at up to 100 million square feet. Texas filings in May 2026 discussed roughly $55 billion initially and as much as $119 billion across all phases, while an August 2026 announcement put the first phase at about $16.8 billion with 3,000 jobs, backed by a $30 million Texas Enterprise Fund grant and JETI program qualification.

Those are announcement figures, not committed capital. TSMC's own chief executive has been publicly sober about the clock. On an April earnings call, C.C. Wei said a new fab takes two to three years to build and another one to two years to ramp, with yield improvement being the actual work.

Terafab also intends to run on its own on-site natural gas generation and battery arrays rather than Tesla solar equipment. For a plant whose entire commercial justification is a guaranteed supply of advanced logic, powering it outside the grid is a substantial additional engineering problem, and one no joint venture partner has committed to solving.

Conclusion

Two things should be watched, and neither is another round of statements of interest.

The first is a described operating structure. Whether TSMC ends up owning the plant, operating a Terafab-owned plant, or supplying only packaging and process support tells you how much Musk is actually conceding. A technology-services contract suggests Terafab is not going anywhere on its own. An equity position with volume commitments suggests something closer to a real foundry.

The second is the fate of 14A. Intel joined a project with one flagship customer and no equity. If TSMC brings a node, the partnership reduces to a footnote, and Intel will have spent a year as the presumptive process supplier of the most discussed fab in Texas.

For everyone tracking AI infrastructure, the significance is narrower and more concrete than the headlines suggest. The question was never whether America would build more leading-edge fabs. It is who controls the process technology inside them, and on whose patents. Musk has answered that question by talking to the company that owns the broadest patent position in the industry. That is worth more to him than any incremental wafer count.

Images

Illustrative: a research microfabrication cleanroom at IEMN in Villeneuve-d'Ascq, France. Not the Terafab site.

Advanced packaging is the stage Terafab most wants in-house, and the one with the fewest substitutes. Diagram comparing 2D, 2.5D and 3D packaging approaches.

A patterned silicon die surface under oblique illumination, showing metal contact arrays and test structures.

A wire-bonded integrated circuit with the package opened, showing the die and bond wires. A 1974-era 2048-bit UV-erasable PROM, illustrative only.

System-in-package assemblies, an example of the multi-die integration work Terafab wants to bring under one roof.

TSMC In, Intel Maybe Out: The Patent Problem Behind Musk's Terafab Talks

Introduction

The Arrangement Everyone Is Describing

Why Intel Is Suddenly Vulnerable

The Part Nobody Is Saying Out Loud

The Vertical Integration Problem

What the Numbers Actually Say

Conclusion

Images

References