Days after reports emerged that TSMC was reassessing plans for a wafer fab in Texas, Tim Culpan, a former Bloomberg columnist and founder of independent media outlet Culpium, disclosed that the chipmaker is in discussions with Elon Musk's Terafab over potential cooperation.
After the X account Whole Mars Catalog shared the report, Musk personally responded, "just discussions, but something may come of it," confirming that talks are indeed taking place, although no agreement has been finalized.
Taiwan-based facility engineering suppliers said they have not yet received instructions from TSMC to begin preparing manpower, equipment, or facility resources for Terafab.
Based on TSMC's previous fab projects in the US, Japan, and Germany, suppliers involved in facility engineering, cleanrooms, gases, chemical supply, and equipment installation typically begin allocating personnel and resources once an overseas fab reaches a more concrete construction stage. This suggests the discussions between Terafab and TSMC remain at the contact and evaluation stage.
However, further inquiries indicate that TSMC's eventual participation cannot be ruled out. Amid US policy pressure, growing chip demand from Tesla, SpaceX, and xAI, and the process and mass-production challenges Terafab would face once construction begins, some facility suppliers estimate the probability of a partnership at more than 80%.
From Terafab's launch in early 2026 to Intel's announcement that it had joined the project in April, and now reports that TSMC has entered the discussions, the project is moving beyond Musk's ambitious vision of building an in-house chip supply chain and confronting the practical realities of wafer manufacturing, process integration, equipment deployment, and volume production management.
Musk launches Terafab to build vertically integrated Texas chip supply chain
Musk announced on X in March that the long-planned Terafab Project would officially move forward. He said Tesla, SpaceX, and xAI would require enormous volumes of chips in the future and that external supply might not be sufficient, creating a need to establish their own semiconductor manufacturing capabilities.
Terafab is envisioned as more than a standalone wafer fab. The project aims to establish a vertically integrated semiconductor manufacturing ecosystem in Texas spanning chip design, wafer fabrication, memory, advanced packaging, testing, and printed circuit boards (PCBs).
The resulting chips would support Tesla's electric vehicles and autonomous driving systems, SpaceX's space operations and Starlink network, and xAI's computing requirements.
The scale discussed in the market is substantial. Initial wafer capacity has been estimated at around 100,000 wafers per month, potentially rising to as many as 1 million per month over the longer term, with annual computing output exceeding 1TW. Initial investment has reportedly been estimated at US$16.8 billion, with total long-term investment potentially reaching US$119 billion.
Musk's push for Terafab stems largely from rising demand across Tesla, SpaceX, and xAI for high-performance computing, autonomous driving, robotics, and space-computing chips. Concentrating critical semiconductor manufacturing in Asia would also create greater supply-chain security risks, making chip supply assurance a central objective of Terafab from the outset.
Intel joins Terafab as Musk names 14A as planned process
Intel announced in April that it had joined Terafab, citing its chip design, manufacturing, and packaging capabilities.
Musk subsequently said Terafab planned to use Intel's 14A process. The parties have not disclosed how manufacturing, packaging, and investment would be divided.
The partnership would also carry strategic significance for Intel. Its 14A node is central to efforts to restore the company's competitiveness in advanced processes and expand Intel Foundry, while the US government currently holds an equity stake in Intel. Whether Intel's foundry operation can secure major customers is also an important part of US efforts to rebuild domestic advanced semiconductor manufacturing.
However, the two sides have not subsequently disclosed a process adoption timeline, fab capacity, investment structure, or formal mass-production schedule.
That does not demonstrate that Intel has withdrawn from Terafab, nor does TSMC's involvement in discussions necessarily mean it would replace Intel.
As Terafab moves toward actual fab construction, it remains unclear whether Intel's proposed process, construction, and manufacturing support fully meets the project's requirements or whether the overall partnership structure needs to be adjusted.
Musk's confirmation of discussions with TSMC nevertheless indicates that Intel may not be Terafab's only potential manufacturing and operational partner—and could also suggest that Terafab is exploring additional capabilities beyond what Intel currently offers.
Terafab joins Musk's growing Texas packaging and PCB footprint
SpaceX has already begun building semiconductor back-end manufacturing capabilities in Texas, including a fan-out panel-level packaging (FOPLP) line and the relocation of PCB production from Los Angeles to Texas.
The first phase of SpaceX's FOPLP operation is planned around 700mm × 700mm panels. Volume production was originally expected around the end of the third quarter of 2026 but is now reportedly delayed until mid-2027.
SpaceX's FOPLP team remains relatively small, with around 10 core personnel, according to industry sources, and additional time will be required for process integration and yield improvement.
Equipment can be purchased quickly and factories can be constructed rapidly, but moving from equipment installation through process integration to yield ramp-up requires substantial manufacturing experience.
SpaceX's PCB operation faces similar challenges. Its Texas PCB line currently has an estimated yield of around 60%, according to market sources, well below the more than 90% achieved by mature Taiwanese PCB production lines.
The experience with PCBs and packaging illustrates the greater manufacturing challenges Terafab would encounter as it moves upstream into wafer fabrication, strengthening the need for a partner with comprehensive semiconductor manufacturing experience.
Semiconductor industry sources noted that a wafer fab must simultaneously manage process development, equipment integration, facility systems, cleanrooms, chemicals and gases, yield management, process control, and customer qualification. Problems in any one area can disrupt the mass-production schedule.
Nvidia CEO Jensen Huang has previously highlighted the complexity of semiconductor fabs, noting that equipment can be purchased but integrating it and mastering process technologies are the difficult parts. Semiconductor manufacturing remains one of the world's most complex manufacturing activities.
TSMC chairman and CEO C.C. Wei similarly said when discussing foundry competition during the company's first-quarter earnings call that there are "no shortcuts" in the foundry business, which requires technology leadership, manufacturing excellence, customer trust, and service capabilities.
Based on TSMC's experience, bringing a wafer fab from construction to mass production can take two to three years, followed by another one to two years to ramp capacity and yields.
This is precisely the challenge now facing Terafab.
Should TSMC ultimately deepen its involvement, cooperation could extend beyond process technology licensing. TSMC possesses capabilities spanning fab construction, process integration, manufacturing operations, and cost control.
One model currently discussed in the market would have TSMC lead fab construction and operations, while Musk provides capital, long-term purchasing commitments, or both.
Another possibility would involve Musk-controlled companies such as SpaceX and Tesla holding a larger equity stake, with TSMC taking a minority position while providing process technology, fab construction expertise, operational management, and volume-production support.
Sony precedent exists, but Terafab would be considerably more complex
TSMC has recent experience with similar overseas partnerships.
In August 2026, TSMC and Sony Semiconductor Solutions signed a definitive agreement to establish Advanced Vision Semiconductor Manufacturing Corporation in Kumamoto, Japan.
Sony will hold the leading position in the venture and invest approximately JPY465 billion (US$3.1 billion), while TSMC will contribute approximately JPY282 billion in cash. The fab will use TSMC process technology and is expected to begin mass production in 2029.
Terafab, however, would involve Tesla, SpaceX, xAI, and US efforts to reshore semiconductor manufacturing, giving it a very different industrial and policy backdrop from the Kumamoto project. Its eventual ownership, technology, and operating structure therefore remains uncertain.
Terafab could stretch TSMC's engineering resources as US expansion accelerates
TSMC's announced US investment has already reached US$265 billion, with its Arizona footprint ultimately planned to expand to 10 wafer fabs, two advanced packaging facilities, and one R&D center.
Only the first Arizona fab has entered volume production, while the remaining facilities are under construction, in planning, or awaiting subsequent development. Adding Terafab to TSMC's US footprint would expand its American manufacturing commitments even further.
At the same time, TSMC continues to increase investment in advanced processes and advanced packaging in Taiwan.
Taiwan's Ministry of Economic Affairs previously said TSMC already has 19 operational sites related to sub-5nm advanced processes and advanced packaging in Taiwan, with another 13 under construction. Expansion projects are also progressing at different stages in Tainan's Shalun area, Longtan Science Park's third phase, Hsinchu, Southern Taiwan Science Park, Central Taiwan Science Park, and Kaohsiung.
TSMC is therefore simultaneously undergoing major expansion cycles at home and overseas.
Another large US fab would require not only additional capital expenditure but also experienced process engineers, equipment integration teams, volume-production management personnel, and substantial facility engineering and supply-chain manpower.
TSMC's long-established competitive advantages include advanced process R&D, equipment integration, process development, yield management, and an ecosystem encompassing equipment, materials, packaging, and IC design companies.
A wafer fab itself can be replicated overseas, but the R&D infrastructure and accumulated engineering expertise that support continued advances in process technology take much longer to reproduce.
If TSMC's US fab footprint continues expanding, the company will inevitably need to deploy more experienced engineers to support overseas construction and production ramps. This could create concerns over dilution of advanced-process R&D resources in Taiwan and accelerate the overseas migration of advanced manufacturing expertise.
From wishing Musk well to potentially supporting Terafab
The US has continued pushing for more advanced semiconductor manufacturing to return domestically. President Donald Trump has repeatedly criticized Taiwan for capturing what he views as an excessive share of US semiconductor industry interests and has called for more wafer manufacturing to take place in the US.
Washington has also continued signaling the possibility of higher tariffs on imported chips, including tariff threats targeting semiconductor products not manufactured in the US.
At TSMC's June 4 shareholders' meeting, Wei was asked whether Musk's move into wafer manufacturing through Terafab could pose a challenge to TSMC. His response was simply: "I wish him well."
Only three to four months later, Musk has confirmed that TSMC and Terafab are discussing potential cooperation.
Supply-chain sources said the shift underscores the growing geopolitical pressure confronting TSMC. With US semiconductor policy increasingly focused on domestic manufacturing and supply-chain security, they described the prospect of TSMC helping establish advanced US process and packaging capacity outside its own wholly controlled operations as increasingly difficult to avoid.
Article edited by Ysi Chen
