Musk's Terafab Five-Year WFE Procurement Could Match Global Annual Market Size
Taylor Wilson
UBS estimates SpaceX's Terafab project will spend roughly $135 billion on wafer-fab equipment over five years — matching this year's entire global WFE market. This means the semiconductor equipment industry is gaining a buyer on par with TSMC.
Why is Musk building his own chip fab?
At a March 2026 unveiling, Musk cited a stark number: the combined AI compute output of every wafer fab on Earth equals only about 2% of SpaceX's target demand.
This means → even if TSMC, Samsung, and Micron expand at full speed, their capacity growth still falls far short of SpaceX's demand curve.
In plain terms = the logic mirrors Tesla's decision to build its own battery lines — suppliers can't scale fast enough, so you build it yourself. "Either we build Terafab, or we have no chips."
What will Terafab make, and how?
The fab targets two chip categories: edge-inference chips for Optimus humanoid robots, and high-power chips optimized for space environments.
Musk estimates terrestrial compute demand at roughly 100–200 gigawatts per year; space-based demand could reach approximately 1 terawatt per year.
Architecturally, Terafab plans to house mask-making, front-end logic and memory processes, advanced packaging, and testing all on a single site. This means → it is not an ordinary fab but a vertically integrated production line from design to final test.
How does UBS arrive at $135 billion in equipment spending?
UBS analyst John Hodulik forecasts SpaceX's AI-related capex at roughly $1.1 trillion over five years, with about 20% — roughly $225 billion — allocated to Terafab.
Applying the industry-standard 60% capex-to-WFE conversion ratio yields a five-year cumulative WFE spend of approximately $135 billion.
The ramp timeline: about $5 billion in 2027, rising to roughly $10 billion in 2028, then jumping past $50 billion per year by 2030–2031. In plain terms = at peak, Terafab alone adds a buyer the size of TSMC to the global equipment market.
Where will it be built, and what does initial capacity look like?
SpaceX has filed for tax abatements in Grimes County, Texas. Filings show an initial semiconductor investment of $55 billion, potentially expanding to $119 billion if all planned phases proceed.
Initial capacity includes roughly 80,000 wafers/month of memory production and two logic/foundry fabs at approximately 20,000 wafers/month each.
This means → Terafab covers both memory and logic lines from day one — it is not a single-product pilot facility.
What role might Intel play?
Intel is in discussions with SpaceX. UBS believes its role could resemble the historical IBM–AMD technology-transfer framework — licensing process recipes, manufacturing IP, design rules, and tool configurations while retaining underlying technology ownership and collecting royalties.
In plain terms = Intel would not build the fab itself but would sell "the recipe for making chips," keeping the IP and earning license fees.
UBS also outlines an alternative scenario: if the pilot line succeeds, Intel could fold its "Ohio One" site into the Terafab system through a joint venture.
What does this mean for memory suppliers and the broader equipment industry?
Musk has explicitly listed memory chips as a Terafab production target, but the source of memory IP remains unclear.
This reflects a potential chain reaction: if Terafab establishes large-scale memory production, it could pressure Samsung, Micron, and SK Hynix to accelerate front-end memory capacity buildouts on U.S. soil. Samsung already has ample land in Taylor, Texas, for expansion.
UBS concludes that regardless of Terafab's final form, the project is a net positive for semiconductor equipment suppliers overall. Whether the 2027 pilot line launches on schedule and equipment orders materialize will be the first critical checkpoint for this thesis.
Content is for reference only, not financial advice.