By Nathan Brooks

Taiwan Semiconductor Manufacturing Co. has begun evaluating Texas as a possible second US manufacturing site, according to Nikkei Asia, with the foundry also weighing participation in Elon Musk's Terafab project. The discussions remain preliminary. The Arizona complex already under construction will absorb most of TSMC's American capital through the end of the decade.

What does Texas add to the map?

Texas offers something Arizona cannot match at TSMC's contemplated scale: water, grid headroom and a labour pool that runs from Dallas to Austin. State officials have spent three years courting advanced manufacturers with tax abatements and grid commitments. The state's industrial base, anchored by Tesla's Gigafactory outside Austin and Samsung's existing foundry, sits within trucking distance of multiple candidate sites.

A Texas footprint would also hedge the American programme against a single-grid outage, a single labour market and a single permitting regime. The Phoenix site depends on Maricopa County water supplies that have drawn scrutiny from state regulators. Texas sites would carry different constraints, chiefly around grid interconnect and cooling water for sub-3 nm lithography tools.

How serious is Terafab?

Terafab, in its current form, looks more like a strategic poster than a procurement contract. Musk has publicly framed the project as a captive AI-chip campus for Tesla and xAI compute demand, with on-site chip production adjacent to data centres. The project has no public process node, no foundry partner and no timetable that survives scrutiny.

TSMC's interest, by contrast, comes with balance-sheet weight. The Arizona complex includes four planned fabs, with the first 4 nm line scheduled for volume production this year and subsequent modules targeting 2 nm and beyond. Total Arizona commitment now sits at roughly $165 billion across the programme's life.

What does Musk actually want?

Musk's chronic frustration with Nvidia's allocation appears to sit behind the Terafab framing. Tesla has historically ordered large Nvidia GPU clusters for training and inference, then designed its own inference silicon in parallel. A captive fab would, in principle, free Tesla from waiting in Nvidia's allocation queue.

It would also free Musk from the TSMC queue, which currently spans roughly twelve to eighteen months for advanced-node wafers. Captive capacity carries its own problems — utilisation, yield ramp, capital cost — but it removes the foundry as a chokepoint. Musk has a track record of solving chokepoints vertically: battery cells, rockets, satellite internet, and now, potentially, wafers.

What changes for customers?

Apple, AMD, Nvidia and Qualcomm all source advanced-node wafers from TSMC today. A Texas line would offer a contiguous-US option for parts that currently route through Taipei or Kaohsiung, which matters less for cost and more for shipping under export controls. The US government has restricted certain advanced AI chip shipments to China; a domestic second-source option insulates customers against future restrictions.

Samsung's existing Austin fab, a logic operation that has run since the 1990s, would become a less lonely neighbour. The Texas semiconductor ecosystem would grow from a single fab operator to three if TSMC commits, with Samsung's Taylor facility under construction adding a fourth site by decade's end.

What happens next?

TSMC has not announced a Texas site, a Terafab partnership or a board-level decision on either. The Arizona programme alone will absorb tens of billions of dollars before the second Arizona module reaches mature yield. A Texas commitment, if it comes, would sit behind that work, not alongside it.

Nikkei's reporting suggests the Texas evaluation is preliminary enough that even the basic question — build, partner or simply observe — has no public answer. The foundry's executives prefer the second option in nearly every public setting. Texas, if it comes, would consume them.