← Latest AI News

August 25, 2026 · Daily brief

SpaceX is putting Nvidia chips in orbit by late 2027

Sovereignty angle
Data centers used to be something you could visit, audit, or switch away from. Now they're in orbit, unreachable, designed by two companies, owned by one person. That's not innovation. That's a new landlord you can never evict.

Elon Musk and Nvidia just announced space-based AI data centers using custom Vera Rubin hardware, targeting a Q4 2027 launch with massive scale-up in 2028.

SpaceX and Nvidia announced they're building orbital AI data centers around Nvidia's Vera Rubin NVL72 system. Elon Musk said the first space-optimized version launches in the fourth quarter of 2027, with significant expansion in 2028.

The hardware going up

Each rack combines 72 Rubin GPUs and 36 Vera CPUs in one liquid-cooled system. Nvidia claims each Rubin chip delivers up to 25 times the AI processing power of the older H100. The orbital version is lighter, simpler, and hardened for radiation and the vacuum of space. SpaceX will build the satellites at its Gigasat Factory in Bastrop, Texas.

Musk called Vera Rubin the best AI computer and committed SpaceXAI exclusively to Nvidia hardware. The company targets 2 gigawatts of compute this year and 10 gigawatts by the end of 2027. Space compute costs over four times ground compute today, but Musk claims that gap will flip in a few years as cooling becomes free and solar power is unlimited.

What it means

Sam Altman called space data centers ridiculous earlier this year. Now the timeline is real, the hardware partner is locked, and the first launch is sixteen months away. Ground data centers face land fights, water shortages, and power grid limits. Orbit has none of those. It also has no technicians, no repairs, and one owner.