Terafab targets 1 million wafers a month after 2032

By Grace Kim

Terafab, the Tesla-SpaceX joint venture, has told equipment makers informally that it seeks production of 1 million wafers per month after 2032, according to Financial News. The proposed mix allocates 700,000 wafers, or 70%, to memory, and more than 300,000 to logic chips.

Procurement officials visited Japan earlier this month to meet local suppliers and work through equipment imports. Some suppliers were asked to deliver within a few months, underscoring the project's reliance on near-term vendor responses.

At SEMICON West, Terafab representatives visited booths operated by about 10 Korean materials, parts and equipment companies, and signed nondisclosure agreements with some, The Elec reported. Industry members were skeptical when Musk announced the plans in March; the subsequent orders have led more companies to view the project as a business opportunity.

It usually takes three years or more to move from planning to volume production at a leading-edge fab. Terafab targets volume production in the second half of 2028. On a SpaceX call last month, Musk called the dates “very aggressive” and said he put “big error bars” on the timeline.

The first SRAM (static random-access memory) test chip is targeted for around May or June 2027 at the Austin research facility. A pilot plant could be completed as early as July 2027. The first phase in Grimes County, Texas, is budgeted at $16.8 billion, while later phases could lift total investment to as much as $119 billion.

Memory production generally splits between DRAM (dynamic random-access memory) and NAND flash (non-volatile storage). Terafab's 700,000-wafer allocation covers both categories, while Samsung and SK hynix figures count DRAM only, so it could rival Samsung's entire estimated DRAM output without necessarily replacing DRAM supply.

Samsung's DRAM capacity was estimated last year at 650,000–700,000 wafers per month; SK hynix was reportedly at about 550,000, with a goal of 1 million by 2030. Terafab's memory allocation would be 3.5 times the reported 200,000 per month at SK hynix's Wuxi fab in China.

The memory type remains unresolved. Terafab is considering technical help from a Taiwanese company, but has named no partner and disclosed no process technology for the 70% memory allocation.

A Tesla job listing for a Terafab Memory Process Integration Engineer, reported in August, covers end-to-end DRAM integration and node development and supports MRAM (magnetoresistive random-access memory), RRAM (resistive random-access memory) and 3D DRAM (three-dimensional dynamic random-access memory).

Terafab has not said how it would divide DRAM output between HBM (high-bandwidth memory) and LPDDR (low-power double data rate memory), both used for artificial intelligence and both mentioned in its job postings. Nor has it said whether it will make and stack DRAM itself or package chips it brings in.

Jiji Press reports that Terafab is working with Intel to adopt its 2nm and 1.4nm process nodes. Building a leading-edge logic fab in the United States is a separate challenge because Tesla and SpaceX lack chip-manufacturing experience.

TSMC's role remains open. TSMC says any fab with its involvement must be operated by TSMC, while Musk said his companies would build and operate Terafab. That combination could leave TSMC as a space tenant.

What does this change for chip supply?

It changes no near-term wafer supply. The first test chip is still targeted for May or June 2027, the pilot plant for July 2027, and volume production for the second half of 2028; the 1 million-wafer goal applies after 2032. Buyers should treat the figure as a procurement target, not committed capacity.

If most of the 700,000 memory wafers become DRAM, Terafab could add substantial chipmaking capacity after its ramp. If the mix centers on NAND flash, the comparison with Samsung and SK hynix DRAM output does not hold. Undisclosed process technology, unresolved packaging plans and Musk's error bars leave the scale and timing uncertain.