SpaceX Reportedly Aiming To Manufacture Its Own GPUs Amid Chip Supply Concerns

The company apparently has no long-term contracts with many of its direct chip suppliers.

According to a report by Reuters, SpaceX is looking to manufacture its own GPUs sometime in the future. While details on this supposed move are unclear, it seems that the company is considering this route as part of a broader push into AI infrastructure. So by GPU, we're looking at AI accelerators, and not gaming hardware.

This information originates from SpaceX's S-1 filing, which the outlet managed to acquire. For context, the company is planning to go public this summer, with an expected IPO of US$1.75 trillion (~RM6.95 trillion). Part of this involves filing an S-1 registration with the US Securities and Exchange Commission to disclose the company's finances and risks.

In this document, the Elon Musk-led company listed "manufacturing our own GPUs" as one of the "substantial capital expenditures" the company will undertake. Basically, it's warning prospective investors that it plans to spend a lot of money on this, among other technologies.

That said, it's uncertain when SpaceX will put the plans into motion. What's more, the specifics of these in-house GPUs are also up in the air. The company may be using "GPU" as a catch-all term for AI processors. Chipmakers have varying approaches to AI processors, each with different naming conventions. NVIDIA, for instance, makes general purpose GPUs. Meanwhile, Google has tensor processing units, which are suited for specific tasks.

Tom's Hardware pointed out that Musk has referred to Tesla's AI5 processor as a GPU despite its lack of specialised hardware, so the filing could mean the AI-series processors. However, the Reuters report implied that this may be a different undertaking.

Previously, Musk confirmed plans for the Terafab, an advanced chip manufacturing complex. This project, which is a joint effort between a joint effort between SpaceX's xAI unit and Tesla, aims to produce chips for various purposes including cars, humanoid robots, and space-based data centres. Intel is expected to play an integral role here. Recently, Musk revealed that Terafab will be using the chipmaker's upcoming 14A technology.

Of course, making GPUs is one of the biggest challenges in the tech industry. It's a costly endeavour involving highly specialised materials and multiple controlled steps. So, SpaceX's supposed plans are looking pretty ambitious. Which begs the question of why it would consider this.

Well, it seems that SpaceX is concerned about chip supply. In the S-1 registration, the company noted that it does not have long-term contracts with many of its direct chip suppliers. Additionally, it expects to continue relying on third-party suppliers for a significant portion of its compute hardware. The company went on to warn that it may not be able to reach its goals regarding Terafab on time, or at all.