With every passing day, it seems like a slightly less crazy idea to put data centres into space. Progress in artificial intelligence (AI) is increasing demand for these facilities, but opposition to building them on Earth is growing.

On July 14th, Kathy Hochul, governor of New York, imposed a one-year moratorium on new data centres with an energy consumption of 50 megawatts or larger, the first state-level ban of its kind in America. She cited concerns about data centres’ environmental impact and their potential to raise electricity bills (one survey found that 71% of Americans would oppose one in their neighbourhood, up from 42% last year). Other jurisdictions are considering similar moves.

Starlink, his firm’s satellite-internet system, has already demonstrated much of the needed technology, he said. But how feasible are such claims? In the video, Mr Musk and other SpaceX executives outlined the design of AI1, the satellite that would be the building-block of Starmind, the company’s proposed orbital data centre. Unlike its communications sibling, an AI satellite does not need fancy radio antennas to communicate with the ground. What it does need is a lot of solar cells to power the AI chips at its core, and a large radiator to dissipate the resulting heat.

SpaceX’s AI1 satellite will have two wing-like solar arrays on either side of the core, each of them 30 metres by 10 metres in size, delivering a peak power of 150 kilowatts (kW). This is a bit larger than the solar arrays on the forthcoming Starlink v3 satellite, which are 19 metres long. The solar cells will deliver 250 watts per square metre, a plausible figure, and will be manufactured by SpaceX at its facility in Bastrop, Texas. SpaceX plans to put its AI1 satellites into special orbits that keep them in daylight 98% of the time.

While the solar arrays are positioned “face on” to the Sun, to maximise their exposure to sunlight, the radiator is positioned “knife-edge” to it, to minimise incoming light. The radiator will be 20 metres by three metres, with a circulating fluid carrying heat from the processors to be radiated away into space at a rate of 154kW. Like the solar panels, the radiator will be unfolded in orbit. These are ambitious goals. Existing Starlink satellites, including the v3 design, do not have dedicated external radiators, but use the chassis of the satellite itself, covered with special coatings, to radiate heat.

The AI1 is “a reasonable extension” of the technology in the Starlink v3 satellite, says Bruce Cameron of Technology Strategy Partners, a consulting firm based in Boston. It is, he suggests, best regarded as a proof-of-concept design to show that the various components can be operated together as a system. Subsequent versions will deal with the challenges of proving out the economics and scaling up the design. To test some of the new technologies needed for AI1, SpaceX intends to launch modified versions of its Starlink v2 satellites, which the company calls “canary sats”, in the coming months.

Similar experiments are being carried out by Starcloud, a startup that has its own plans to build an orbital data centre. In November it launched Starcloud-1, a 60kg test satellite containing a single Nvidia chip, to see how it performed in orbit. In January 2027 it plans to launch Starcloud-2, with eight times the mass and 100 times the processing power, to test its design for a deployable radiator. (Philip Johnston, Starcloud’s boss, says two-thirds of his engineers are working on cooling and radiators.) The firm raised $170m in March at a valuation of $1.1bn.

Showing that AI satellites can work in orbit is a necessary step towards building an orbital data centre, but it is not enough. The biggest question remains that of launch costs. SpaceX currently delivers payloads to orbit at a price of around $1,500 per kilogram with its Falcon Heavy rocket, or $3,400/kg with its Falcon 9. The actual cost to SpaceX is around $600/kg, according to Bain, another consulting firm. To compete with terrestrial data centres, launch costs would have to be much lower: around $50-100/kg, Bain reckons.

SpaceX’s enormous Starship rocket, which is designed to be fully reusable, could cut costs to that level. But Starship does not work yet; its next test flight (its 13th) is expected in the coming days. And Starlink v3 satellites, let alone AI1 satellites, are too large to launch with the existing Falcon 9. So SpaceX really is betting everything on getting Starship to work. Mr Musk talks optimistically about launching the first AI1 satellites during 2027, but he is infamous for his over-optimistic timelines.

The prospects for orbital data centres also depend on more down-to-Earth factors. Will demand for AI computing capacity continue to grow? An industry crash or a technical breakthrough could slam on the brakes. And will opposition to terrestrial data centres intensify or subside? Orbital data centres may not violate the laws of physics. But the economics and politics are less clear-cut. ■