SpaceX seeks $40 billion for Nvidia chips as orbital AI race heats up

Vera gives us the CPU performance and memory bandwidth to run enormous amounts of orchestration, code and data processing while keeping GPUs doing what they do best.

— Mike Nicolls, president of SpaceXAI, in Nvidia’s announcement

SpaceX is not alone off the planet

According to SpaceX, the Starmind satellites would harness solar energy in sun-synchronous orbit and let off heat into space. Each satellite has a peak processing capacity of 250 kilowatts.

A similar experiment is being conducted by Google as reported by NPR. The prototype of the Project Suncatcher carried four TPUs to see how the chips with AI capability would withstand the radiation and the extreme temperatures of outer space.

SpaceX and Google are not the only companies venturing in this direction. Starcloud has successfully launched an Nvidia H100 into orbit. This makes orbital AI appear less like a science experiment and more like a budding competition for computing power.

The economics and physics are still open

Gartner projects that worldwide spending on AI is likely to reach $2.7 trillion in 2026, reflecting a 49.5% increase, which will be attributed to infrastructure costs.

Currently, the economics of placing that infrastructure into orbit are still very complicated, as BCG estimates that orbital data facilities will cost 2.5 to 3 times more than ground facilities.

Cooling is another issue. Brookings has estimated that one orbital data center would require 2.15 million square feet of radiators.

However, heat is not the only engineering problem. Chips in space must also withstand radiation. This has led semiconductor companies to put more effort into designing chips that are radiation-hardened (rad-hard).

We have seen a major uptick in activity for rad-hard fab processes.

— Dana Neustadter, senior director at Synopsys, via Semiconductor Engineering

Radiation, temperature fluctuations, and complicated repairs all make space computing much more difficult than using the same equipment on Earth.

What the deal does to the field

The financing strengthens Nvidia’s position while making the race more expensive for smaller rivals. Reuters cites Morgan Stanley’s estimate that AI infrastructure will need $1.5 trillion in outside financing by 2028. Those pressures are already showing up as power constraints complicate the broader AI buildout.

SpaceX is trying to raise around $40 billion with the help of Apollo Global Management to acquire AI chips from Nvidia, as per Financial Times. The importance of this deal lies in the fact that these chips can be used to power SpaceXAI’s data centers on the planet, as well as its planned AI infrastructure in orbit.

Media reports suggest that the financing will consist of approximately $10 billion in bank loans and $30 billion in the form of investment-grade debt. The transaction is expected to conclude in 2027, with PIMCO being one of the lenders that have been approached. Following the news, SpaceX stock dropped by 1% after-hours trading, while Nvidia gained 0.5%.

Why the financing points past Earth

Nvidia says SpaceXAI will use Vera CPUs and its Vera Rubin platform as Grok’s infrastructure grows toward gigawatts of computing capacity. The first-generation Starmind satellite is also set to use an optimized Vera Rubin NVL72 system.

Vera is Nvidia’s first CPU built for AI agents. It has 88 Olympus cores and up to 1.2 TB/s of memory bandwidth. Nvidia says it can complete some tasks up to 1.8 times faster than x86 processors.

Vera gives us the CPU performance and memory bandwidth to run enormous amounts of orchestration, code and data processing while keeping GPUs doing what they do best.

— Mike Nicolls, president of SpaceXAI, in Nvidia’s announcement

SpaceX is not alone off the planet

According to SpaceX, the Starmind satellites would harness solar energy in sun-synchronous orbit and let off heat into space. Each satellite has a peak processing capacity of 250 kilowatts.

A similar experiment is being conducted by Google as reported by NPR. The prototype of the Project Suncatcher carried four TPUs to see how the chips with AI capability would withstand the radiation and the extreme temperatures of outer space.

SpaceX and Google are not the only companies venturing in this direction. Starcloud has successfully launched an Nvidia H100 into orbit. This makes orbital AI appear less like a science experiment and more like a budding competition for computing power.

The economics and physics are still open

Gartner projects that worldwide spending on AI is likely to reach $2.7 trillion in 2026, reflecting a 49.5% increase, which will be attributed to infrastructure costs.

Currently, the economics of placing that infrastructure into orbit are still very complicated, as BCG estimates that orbital data facilities will cost 2.5 to 3 times more than ground facilities.

Cooling is another issue. Brookings has estimated that one orbital data center would require 2.15 million square feet of radiators.

However, heat is not the only engineering problem. Chips in space must also withstand radiation. This has led semiconductor companies to put more effort into designing chips that are radiation-hardened (rad-hard).

We have seen a major uptick in activity for rad-hard fab processes.

— Dana Neustadter, senior director at Synopsys, via Semiconductor Engineering

Radiation, temperature fluctuations, and complicated repairs all make space computing much more difficult than using the same equipment on Earth.

What the deal does to the field

The financing strengthens Nvidia’s position while making the race more expensive for smaller rivals. Reuters cites Morgan Stanley’s estimate that AI infrastructure will need $1.5 trillion in outside financing by 2028. Those pressures are already showing up as power constraints complicate the broader AI buildout.