SpaceX eyes building its own GPUs as AI plans expand

SpaceX has told investors that its AI push may force it into one of the hardest parts of tech: making GPUs.

In excerpts from the company’s S-1 registration, SpaceX listed “manufacturing our own GPUs” among the “substantial capital expenditures” behind its work on AI and other technologies. The filing comes ahead of a summer IPO expected to value SpaceX at about $1.75 trillion.

An S-1 is the document companies file with the U.S. Securities and Exchange Commission before going public.

The decision ties into Terafab, the AI chip complex being developed in Austin, Texas by SpaceX, its xAI unit, and Tesla. Elon Musk has said the site is aimed at chips for cars, humanoid robots, and space-based data centers.

Many details have stayed unclear, including what kind of AI chips the project may actually make. One key question is whether SpaceX means standard GPUs, or a wider label for AI processors.

SpaceX warns investors that outside chip supply may not keep up with growth

The filing says SpaceX may not have enough chip supply to support its growth. In the registration, the company said, “We do not have long-term contracts with many of our direct chip suppliers.”

It also said, “We expect to continue sourcing a significant portion of our compute hardware from third-party suppliers, and there can be no assurance that we will be able to achieve our objectives with respect to TERAFAB within the expected timeframes, or at all.”

The company has not said when it plans to start making its own chips. It is also unclear which groups inside the Terafab effort, or partner Intel, would handle the fabrication technology inside the plant.

Elon told Tesla analysts on Wednesday that by the time Terafab scales up, Intel’s next-generation 14A manufacturing process “will be probably fairly mature or ready for prime time” and “seems like the right move.”

Different companies are taking different paths on AI chips. Nvidia mainly makes GPUs, general-purpose chips used for heavy data work. Google, owned by Alphabet, uses TPUs, built for specific jobs tied to training AI models and running chatbots such as Anthropic’s Claude. SpaceX has not said which route it wants to follow.

SpaceX brings Cursor into its AI push after Microsoft steps back from talks

The AI plan includes Cursor, the coding startup tied to a $60 billion deal. Before SpaceX announced this week it had secured the right to acquire Cursor, Microsoft had explored a deal, according to CNBC.

But Microsoft chose not to proceed, and it is now trying to make its own AI tools more popular. It has gained ground with GitHub Copilot, but the AI coding market is now led by Cursor, Anthropic, and OpenAI. Microsoft’s main role there has been as an investor and cloud provider, putting billions into Anthropic and OpenAI, which committed to heavy spending on Microsoft Azure.

A company post said, “SpaceXAI and @cursor_ai are now working closely together to create the world’s best coding and knowledge work AI.” Cursor chief executive Michael Truell wrote on X that he was “excited to partner with the SpaceX team to scale up Composer,” referring to its AI model.

The SpaceX agreement came together so late in Cursor’s fundraising process that prospective investors were caught off guard. In the weeks before the announcement, SpaceX had already offered Cursor access to compute.

Making GPUs is extremely hard. Nvidia pioneered GPU design, but outsources manufacturing to TSMC in Taiwan. TSMC has spent years and billions building advanced chip processes.

Producing leading-edge chips takes exotic materials and more than a thousand steps done with atomic precision. Making billions of Apple iPhone chips gave TSMC the hands-on experience needed to keep producing advanced processors for years at scale.

If you're reading this, you’re already ahead. Stay there with our newsletter.