SpaceX Prepares Own GPUs for AI, Targets Nvidia and AMD Market

SpaceX officially confirmed that it is considering developing its own GPU for artificial intelligence systems. For the company, this is no longer a side research idea, but part of major capital expenditures that it warned investors about ahead of the anticipated IPO, where the company's valuation could reach $1.75 trillion.

If the project reaches implementation, SpaceX will attempt to enter territory dominated today by Nvidia and AMD, with stakes tied not only to AI but to a broader attempt to control critical computing infrastructure. The plans became known from the S-1 registration statement filed with the U.S.

Securities and Exchange Commission. In the document, SpaceX directly mentioned the production of its own GPUs as a significant item in future investments. However, the company did not disclose the budget, timeline, or exact architecture of the future chips.

Therefore, it remains unclear whether this is about full-fledged graphics processors in the conventional market sense or specialized accelerators for machine learning tasks, which in corporate rhetoric are also often called GPUs.

The reason for such a move appears clear. SpaceX acknowledges that it depends on external chip suppliers and does not have long-term contracts with many of them. Against the backdrop of a shortage of high-performance AI accelerators, this becomes a strategic risk: companies building large models, data centers, and robotics platforms find it increasingly difficult to guarantee supplies in needed volumes at acceptable prices. For SpaceX, the issue is particularly sensitive because computing resources are needed by several directions at once — from its own AI systems to related projects involving xAI, Tesla, and robotics.

The development of chips is tied to the Terafab project — a joint initiative of SpaceX, xAI, and Tesla, overseen by Elon Musk. It is expected that an advanced manufacturing facility for producing AI chips will appear in Austin, capable of operating in cars, Optimus humanoid robots, and even space data centers. In a meeting with Tesla analysts, Musk also pointed to Intel's 14A manufacturing process as a probable foundation for bringing the project to an industrial level. This heightened expectations that Intel might become a key production partner, although the distribution of roles among participants has not yet been disclosed.

The main problem is that the GPU market is almost impossible to storm with money and ambition alone. Even Nvidia, the de facto leader in the segment, primarily designs chips and outsources production to TSMC, which has spent years building extremely complex manufacturing processes and invested billions of dollars in them. For a new player, the entry barrier here is huge: you need not just to design a competitive accelerator, but also to establish production with acceptable chip yields, stable packaging, memory, interconnect, and a software ecosystem.

If SpaceX truly aimed at the complete cycle, this is one of the most expensive and risky industrial projects in the AI sector. The scale of plans is also emphasized by Terafab assessments. The ultimate goal of the project is called the production of one terawatt of computing capacity per year — roughly double the current combined computing volume in the United States.

Bernstein analysts estimate the potential capital expenditures for such a level of capacity in the range of $5 to $13 trillion. Even if actual investments turn out to be lower or stretched over years, the very setting of the goal shows that SpaceX and related companies want not just to purchase accelerators for their own tasks, but to build their own computing vertical.

For the market, this is a signal in several directions at once. First, the largest technology groups increasingly fear dependence on Nvidia and a limited number of advanced chip manufacturers. Second, AI infrastructure is becoming such a scarce and expensive asset that companies are willing to go into production that was previously considered too complex even for giants. And finally, the news itself shows: the next competition in AI will not only be for models and products, but also for control over hardware, factories, and supply chains. SpaceX does not yet have a ready-made GPU, but the very fact of such an intention already changes the arrangement of expectations on the market.