SpaceX is pursuing an in-house chip initiative called Terafab to manufacture its own GPUs and AI accelerators, tying into xAI and Tesla, as revealed by Reuters through a review of the company’s filing.

The plan could leverage Intel’s upcoming 14A process once scaling allows, though details remain uncertain.

There is ambiguity over whether the term GPU is being used strictly or broadly to refer to AI chips, and SpaceX has not disclosed specific spending or production timelines.

Vertical integration in semiconductors carries risks like higher upfront costs and potential delays, since the industry benefits from specialization that improves yields and lowers costs.

The filing warns investors that substantial AI spending and potential supply risks could impact the IPO.

The motivation for in-house production is to reduce chip-supply risk and dependency on external suppliers by controlling the design to packaging and testing in one line.

The S-1 notes large planned capital expenditures to build in-house silicon due to a lack of long-term supply contracts with direct chip suppliers.

Other IPO contenders in 2026 include OpenAI, Anthropic, and Databricks.

The move underscores a broader market concern about securing AI compute power, as leaders rely on external foundries or have built custom chips, highlighting chip supply as strategically critical for AI development.

The filing warns of chip-supply risks, noting few long-term contracts with direct chip suppliers and potential difficulty in securing enough compute hardware from third parties.

SpaceX states it does not have long-term contracts with many direct chip suppliers and will continue to source hardware externally while pursuing more internal capabilities.

Building GPUs and controlling the full chip-supply chain entails massive upfront costs, uncertain timelines, and significant technical challenges.