Elon Musk is planning to build an unprecedented vertically integrated semiconductor complex, with a potential investment scale significant enough to reshape the global semiconductor equipment market landscape.
According to a research report from UBS, the projected wafer fabrication equipment (WFE) procurement for the Terafab project under SpaceX over the next five years is estimated to be equivalent to the total size of the global WFE market this year. Analysts anticipate that Terafab could commence its pilot production line as early as 2027, with equipment orders already beginning to be placed with suppliers, with an initial scale of around $50 billion.
The advancement of this project implies that the ceiling for the global WFE market will be substantially raised. The report notes that if Terafab proceeds as scheduled, global WFE spending could approach $300 billion by 2029, and around 2030-2031, a new scale could emerge where a single customer makes annual purchases exceeding $50 billion.
Terafab: A Full-Chip-Supply-Chain Superfactory
The core rationale for Terafab mirrors Tesla Motors's past strategy of building its own battery supply chain. Musk stated clearly at the Terafab project launch in March 2026 that the combined AI computing power output of all global wafer fabs is only about 2% of SpaceX's target demand. He pointed out that the expansion pace of existing suppliers—including TSMC, Samsung, and Micron—is far slower than the growth rate of SpaceX's needs, stating "either build Terafab, or have no chips."
In terms of product planning, Terafab will focus on two types of chips: one for edge inference, primarily for the Optimus humanoid robot, and the other high-power chips optimized for the space environment. Musk estimates that ground-based computing demand is about 100 to 200 gigawatts annually, while space-based computing demand could reach as high as about 1 terawatt per year.
In terms of factory architecture, Terafab plans to integrate mask manufacturing, front-end logic and memory processes, advanced packaging, and testing all under one roof, aiming to achieve an ultra-fast "design-manufacture-test-iterate" closed loop. The report characterizes it as a truly vertically integrated semiconductor complex.
WFE Expenditure Forecast: $135 Billion Over Five Years, Peaking at Over $50 Billion Annually
Analyst John Hodulik predicts that the total capital expenditure for SpaceX's AI business (excluding satellite internet and rocket launch operations) over the next five years will be approximately $1.1 trillion, with about 20% of that, or roughly $225 billion, allocated to Terafab. Based on the industry-standard ratio where about 60% of capital expenditure translates into WFE, the cumulative WFE procurement over five years is estimated at around $135 billion—roughly equivalent to the total global WFE market size this year.
Looking at the timeline, Terafab's WFE spending is expected to start with a pilot line of about $5 billion in 2027, increase to about $10 billion in 2028 (already factored into the ~$250 billion global WFE forecast for 2028), and surge to over $50 billion annually around 2030-2031. This would effectively add a new WFE buyer comparable in scale to TSMC on top of the existing market.
Regarding location, SpaceX has filed for tax abatement in Grimes County, Texas. Related documents indicate an initial semiconductor wafer fab investment of $55 billion, with the total investment potentially expanding to $119 billion if all planned phases materialize. Based on this, Terafab's initial capacity layout might include around 80,000 wafers per month of memory capacity, along with two logic/foundry fabs of about 20,000 wafers per month each, supported by mask shop and backend packaging and testing capacity.
Intel Could Be "Technical Knowledge Holder"; Memory Suppliers Face Potential Impact
On the technical cooperation front, the report indicates that Intel is in discussions with SpaceX regarding Terafab. Its role could be similar to the historical technology transfer framework between IBM and AMD—that is, providing Terafab with process technology flows, manufacturing IP, PDK design rules, and tool recipes through licensing, while retaining underlying technology ownership and collecting licensing or royalty fees. UBS also suggests another possible scenario: if the pilot line proves successful, Intel might contribute its "Ohio One" campus to the Terafab system via a joint venture. That campus is of sufficient scale to support the operation of two leading-edge process fabs. Furthermore, if Terafab ultimately does not succeed independently, the likelihood of SpaceX making further investments in Intel would increase.
In the memory sector, Musk explicitly listed memory chips as one of Terafab's production targets during the launch. However, the source of memory IP remains unclear, as existing memory suppliers have limited motivation to license core IP to a potential competitor. Nevertheless, the report also notes that if Terafab successfully establishes scaled memory production capability, it could conversely pressure Korean companies to accelerate the deployment of front-end memory capacity in the US—Samsung already has ample land available for expansion in Taylor, Texas, with Micron and SK Hynix also being within scope.
In summary, regardless of the final form Terafab takes, the project's advancement is viewed as a positive overall support for semiconductor equipment suppliers and is expected to be a core market theme for several upcoming earnings seasons.
Comments