SpaceX Partners With NVIDIA to Put Datacentre-Class AI Computing Into Orbit

There’s a moment when a technology shifts from something you read about in science fiction to something engineers are actually building. SpaceX’s announcement of its Starmind AI1 project, developed in partnership with NVIDIA, feels like one of those moments.

SpaceX posted details of the collaboration, describing a plan to place a fully capable AI compute payload into orbit — not a communications satellite in the conventional sense, but something closer to a data centre floating above the atmosphere.

What Starmind AI1 Actually Is

The concept is straightforward enough, even if the engineering behind it is anything but. SpaceX’s Starmind materials describe an orbital satellite that uses available power and cooling in space to run localised computing workloads, then beams the results back down via high-bandwidth laser links to the Starlink constellation. From there, data reaches users on the ground through the existing Starlink network.

Think of it as a server farm in low Earth orbit, using the cold vacuum of space to help manage heat, and NVIDIA’s latest hardware to do the heavy lifting.

NVIDIA’s own space-computing page confirms the arrangement, stating that the Starmind AI1 payload is powered by the NVIDIA Vera Rubin NVL72 — a system built around NVIDIA’s Rubin GPUs and Vera CPUs. The Vera Rubin NVL72 represents NVIDIA’s current generation of high-density AI compute architecture, the same class of hardware that’s driving demand in ground-based AI data centres around the world.

The Power Figures Tell the Story

SpaceX’s published Starmind materials list a compute payload drawing 150 kilowatts at peak and 120 kilowatts on average. To put that in context, a typical family home in the UK uses somewhere around 3 to 4 kilowatts at peak. A single Starmind satellite, on these figures, would draw roughly the equivalent of 40 homes running flat out — all packed into a spacecraft.

Some secondary reporting, drawing on SpaceX’s Starmind site, has cited higher figures of up to 250 kilowatts peak and 175 kilowatts average. Those numbers haven’t been independently verified in SpaceX’s or NVIDIA’s official published materials, so treat them with some caution for now.

Either way, the scale of what’s being proposed is striking.

An Orbital Data Centre, Not Just a Satellite

Jensen Huang, NVIDIA’s chief executive, has spoken repeatedly about the company’s ambitions beyond the ground. But the Starmind partnership is a concrete step rather than a vague aspiration. SpaceX and NVIDIA are publicly naming the hardware, describing the architecture, and framing this as a real engineering programme.

The project’s positioning is deliberate. SpaceX is not describing Starmind AI1 as a communications upgrade to Starlink. It’s presenting it as AI infrastructure — a platform for running AI workloads in orbit, with Starlink serving as the return pipe for data.

That’s a meaningful distinction. It suggests SpaceX sees a commercial opportunity in selling orbital compute time, not just bandwidth.

What Comes Next — and What’s Still Unclear

Prototype testing for Starmind AI1 is reported in secondary coverage to be planned for early 2027, though that detail hasn’t been confirmed in SpaceX’s or NVIDIA’s official published materials. The exact commercial model — who buys orbital compute time, at what price, and for what use cases — also remains unspecified in what’s been made public so far.

And some observers are cautious. The power figures in circulation have shifted across different reports, which suggests the design may still be evolving. It would be fair to say this is an ambitious programme that’s been announced publicly at what appears to be a fairly early stage.

But SpaceX has a track record of moving faster than sceptics expect. And NVIDIA’s involvement, with named hardware rather than a vague partnership announcement, gives the project more substance than many early-stage space ventures.

The wider context matters too. Both companies have strong commercial incentives to make orbital AI compute work. NVIDIA wants its hardware everywhere. SpaceX wants Starlink to be more than a broadband service. Starmind AI1 is where those two ambitions meet.

What This Means for Kent Residents

There’s no direct impact on Kent from this announcement in the near term. But for anyone working in the UK’s growing space technology sector — including research programmes at universities across the South East — the emergence of orbital AI infrastructure as a serious commercial category is worth watching. More broadly, if projects like Starmind AI1 succeed, they could eventually expand the kinds of high-bandwidth, low-latency services available through satellite networks like Starlink, which already serves rural parts of Kent where fixed-line broadband remains limited.