The two companies most associated with the AI boom's hardware layer are teaming up to move some of it off the planet.
According to Memeburn, SpaceX and NVIDIA have announced a partnership to build Starmind AI1, a satellite compute payload — essentially a data center designed to run in orbit rather than in a warehouse. Memeburn reports the payload will be powered by NVIDIA's Vera Rubin NVL72 racks, each carrying 72 GPUs, with prototype launches targeted for early 2027.
That is the whole of what has been confirmed so far: a name, a chip platform, a rack configuration, and a launch window roughly a year and a half out. Details on cost, capacity, or how many satellites would follow a successful prototype have not been disclosed in these reports.
Investors noticed quickly. Zacks Investment Research published a piece asking whether the partnership makes Nvidia or SpaceX worth buying, a sign that the announcement is being read as a business story as much as an engineering one.
Why space at all? AI data centers on the ground are running into hard physical limits — electricity, land, and cooling. Orbit offers uninterrupted solar power and the cold of space as a heat sink, though it also introduces launch costs, radiation, and the problem of servicing hardware you cannot physically reach.
It matters because if orbital compute works even partially, it changes where the AI industry's biggest constraint — power — has to be solved, and puts the company that builds the chips alongside the company that owns the cheapest ride to space.