SpaceX unveiled AI1, a low‑Earth‑orbit satellite that functions as a modular data center, capable of delivering up to 150 kilowatts of peak power for AI workloads.. the craft, revealed in a press briefing this week, is designed to run heavy compute tasks that normally require massive ground‑based facilities, using solar power and a novel cooling systm.

AI1’s 150‑kilowatt peak power challenges Earth grids

According to the launch briefing, AI1 can generate a peak output of 150 kilowatts and sustain an average of 120 kilowatts, powered by solar arrays that produce 250 watts per square meter. This capacity is comparable to a single Nvidia GB300 rack, which on Earth would draw roughly 140 kilowatts, according to Elon Musk’s remarks. by moving such workloads to orbit, SpaceX hopes to relieve local electricity grids that are increasingly strained by AI training demands .

Modular compute payload lets SpaceX swap any chipmaker’s hardware

The satellite’s open‑architecture payload is built to accept interchangeable processors from any vendor, a move designed to sidesteep the semiconductor supply crunch.. SpaceX’s partnership with Tesla’s TeraFab solution provides the manufacturing flexibility needed to upgrade AI1 with newer GPUs or specialized accelerators as they become available, as the company explained in the briefing.

110 m² liquid radiators tackle space‑borne heat

Cooling in vacuum is a major engineering hurdle, and AI1 addresses it with 110 square meters of deployable liquid radiators and redundant pumping loops. Unlike terrestrial servers that rely on air or water cooling, the satellite radiates heat into space, a system described by SpaceX engineers as achieving 70 kilowatts per ton of power efficiency.

Will AI1 ever become cost‑effective versus terrestrial data centers?

Critics, including Nvidia CEO Jensen Huang and leaders at Amazon Web Services, have questioned whether the launch and maintenance costs can ever match the economics of ground‑based clouds. As the report notes, the current AI1 is a draft version meant as a proof of concept, leaving open whether future iterations will achieve cost parity.