SpaceX may build its own GPUs to reduce chip supply risks

[midian182]

What we know so far: SpaceX may be about to take on one of the most difficult jobs in the tech industry: making its own AI chips. According to reports, the company has warned prospective investors that chip supply constraints and the cost of securing enough compute hardware could become a serious problem. As such, it's now considering manufacturing its own GPUs.

The disclosure appears in excerpts from SpaceX's S-1 filing ahead of its expected IPO this summer. Reuters says the filing lists "manufacturing our own GPUs" among the "substantial capital expenditures" the company is taking on.

The filing acknowledges that SpaceX still expects to rely heavily on third-party suppliers for a significant portion of its compute hardware. The company also said it does not have long-term contracts with many of its direct chip suppliers, leaving it more exposed to shortages or price spikes.

Elon Musk outlined a joint Tesla, SpaceX, xAI, and Intel chipmaking effort last month called Terafab, an advanced manufacturing complex planned for Austin, Texas. SpaceX's reported in-house GPU ambitions appear to tie directly into that same project, which is supposed to help produce the processors needed for cars, humanoid robots, and space-based data centers.

– Tesla (@Tesla) March 22, 2026

It seems the plan isn't just about reducing Nvidia dependence; it appears to align with Musk's broader push to expand in-house AI infrastructure across his companies.

Wanting to build GPUs and actually doing it are two very different things, of course. Producing cutting-edge chips requires billions of dollars, highly specialized materials, and a manufacturing process involving well over a thousand tightly controlled steps.

It's still unclear when SpaceX plans to manufacture its own chips, whether "GPU" is being used precisely or as a catch-all label for AI processors, and which company would handle the fabrication technology inside Terafab.

The report notes that SpaceX's filing frames compute hardware as a potential operational and financial risk, particularly given its reliance on outside suppliers and the lack of long-term contracts with some of them.

Reuters says SpaceX's filing identifies compute hardware as a potential operational and financial risk because of its dependence on outside suppliers and the absence of long-term contracts with some of them. It also leaves several questions unanswered, including what role Terafab would play in that effort.

Permalink to story:

SpaceX may build its own GPUs to reduce chip supply risks

 

[wiyosaya]

I'll believe it when I see it.

SpaceX is already significantly behind on producing a Lunar Lander for NASA and they are going to take on another technically very difficult venture like AI chip manufacturing?

I'll believe it when I see it. IMO, SpaceX is trying to do too many things at the same time and its overall leader isn't helping.

 

[return]

"Please, pay your subscription to unlock more cores to able land back on the Earth."

 

[Endymio]

SpaceX is already significantly behind on producing a Lunar Lander for NASA and they are going to take on another technically very difficult venture like AI chip manufacturing?

I'll believe it when I see it. IMO

All of Artemis is behind schedule; SpaceX's HLS is actually ahead of several other mission components. And SpaceX's Falcon series has just completed 642 total launches, 639 of which were complete successes, by far the best launch record of any government, military, or private company in the world. And their Crew Dragon is still the only reason we're not having to rely on Russia to send our astronauts to and from the ISS.

 

[GABulldog]

From a business perspective, makes perfect sense.

 

[wiyosaya]

All of Artemis is behind schedule; SpaceX's HLS is actually ahead of several other mission components. And SpaceX's Falcon series has just completed 642 total launches, 639 of which were complete successes, by far the best launch record of any government, military, or private company in the world. And their Crew Dragon is still the only reason we're not having to rely on Russia to send our astronauts to and from the ISS.

I have better information that you do. Throwing your straw man on the fire along with gasoline - as usual.

 

[Plutoisaplanet]

SpaceX will just rely on the Terafab for chip manufacturing. Most likely the language regarding SpaceX manufacturing its own GPUs is just standard risk disclosure statements since the Terafab is a future project. Risk disclosure statements are often answers to boilerplate questions like "what will happen if you can't do X listed plan?" It doesn't mean there are real plans to do these things.

 

[BuckarooBonzaii]

I think in the beginning they'll need a fab like TSMC to get things going. Then we'll see how far SpaceX will get.

 

[Plutoisaplanet]

I have better information that you do. Throwing your straw man on the fire along with gasoline - as usual.

You should provide a source if you want to say your source is better. Otherwise your refutation is no different from "trust me bro" in this case. Once you do, then people would be able to see the "better information".

I would agree that all of Artemis is probably not behind SpaceX for getting humans from lunar orbit to the moon's surface, but as far as I can tell the alternative option (Blue Moon) is behind Starship HLS. It relies on BE-7 engines which are still under development, while Starship's Raptor 3 engines have been used (even in space) many times successfully. Other lunar landers Nova-C and Blue Ghost have been launched by the Falcon 9, while the Blue Moon Mark 1 (not capable of transporting humans) has no schedule to be launched currently. New Glenn also needs to put a real payload in orbit or else Blue Origin can't even test its own lunar lander (without relying on SpaceX that is).

Meanwhile Starship V3 (in spite of many design changes in Starship Block 1 and 2) is set to launch in May and has completed full static fire testing. There will probably be a few failures, but there are no significant design changes planned anymore. There are two new launch pads designed for permanent usage now, heat shield design is tested, and so SpaceX plans to test Starship refueling in orbit whenever kinks are worked out with its new vehicle. All of the criticism with Starship so far has been due to continuing design changes, and these are all but finished.

 

[PurpaFur]

It relies on BE-7 engines which are still under development, while Starship's Raptor 3 engines have been used (even in space) many times successfully.

Probably just a typo, but Raptor 3 engines have never flown. They are installed in the vehicles for the upcoming flight 12, which will be the first time Raptor 3 engines are used in a launch. Raptor 2 has flown a bunch.

 

[Plutoisaplanet]

Probably just a typo, but Raptor 3 engines have never flown. They are installed in the vehicles for the upcoming flight 12, which will be the first time Raptor 3 engines are used in a launch. Raptor 2 has flown a bunch.

Thanks for the correction, it's been tested only on the ground so far. Right now the holdup before a potential launch is a failure in the water deluge system (part of the launch pad), though surely there are other unknown issues to address down the line (whether before or after V3's maiden launch).

 

[PurpaFur]

Thanks for the correction, it's been tested only on the ground so far. Right now the holdup before a potential launch is a failure in the water deluge system (part of the launch pad), though surely there are other unknown issues to address down the line (whether before or after V3's maiden launch).

Here's to hoping no delays on the 12th!