Orbital Data Centers: The Honest Case

‍

The pitch for orbital data centers is seductive: near-limitless solar power, an infinite heat sink, and not a single county permit. As terrestrial AI runs into power shortages and water fights, the idea of moving compute above the grid has gone from fringe to boardroom fast. Elon Musk says all of it is going up. “It may very well be that Elon’s right, and in three years all of our data centers will be in space,” Jonny Dyer told the room. “You may not be right about that timeline. But in some ways it doesn’t matter, because we’re on this secular trajectory where that’s where we’re headed.”

Dyer, the Muon Space CEO, sat down with Space Capital partner Tom Ingersoll at the 2026 Space Capital Summit at Nasdaq MarketSite to separate the physics from the hype. His read: the physics works, the economics do not yet, and almost every obstacle in between is an engineering problem rather than a physics one. And orbital data centers, he argued, are the least of it. “It’s sort of the ice above the water,” he said, “and there’s an iceberg below the surface that really needs to be looked at to even understand why we’re talking about data centers in space.”

Why we are talking about this now

The conversation is only possible because of launch, and Dyer has watched the cost collapse firsthand. “When Tom and I were at Skybox fifteen years ago, we were buying converted Russian ICBMs to launch our satellites,” he said, “because that was the best, lowest-cost, most available way to put a small satellite in orbit, which is kind of crazy in hindsight. Now we can buy spots on SpaceX launches almost like a plane ticket.” Starship, he expects, will reset the economics again.

The second shift is how spacecraft are built. Space has stopped being bespoke and started borrowing from the industries that already manufacture at volume. “The Starlink satellites look a lot more like a rack that goes in a data center than they look like a satellite,” Dyer said. Cheaper launch and commodity-style manufacturing feed each other, and that loop is what puts gigawatts of orbital compute in the realm of the possible at all.

Why orbit is genuinely attractive

Start with what is real, because a lot of it is. On power, the sun is stronger above the atmosphere, and in the right orbit it never sets. “You can pick orbits where the sun is baseload power, where it’s always on in a way that you fundamentally can’t on Earth,” Dyer said, which means always-on solar without the storage a ground array needs to survive night and weather.

Cooling is the second advantage. Deep space is a free heat sink, an almost-absolute-zero background a data center can radiate into, with none of the water, real estate, or permitting fights that cooling triggers on Earth. The third is regulatory: there are no county councils in orbit. “If you come up with a regulatory approach, it’s a global approach, literally,” he said. The catch, as he put it, is that capturing any of this is “not physics, it’s engineering.”

The demand is real. The economics are not there yet.

Demand is not the issue. The compute shortage is real and, in Dyer’s words, “potentially unbounded in size as we look forward to a future where AI is growing explosively,” while the grid, especially in the United States, is falling behind.

Supply is where he gets honest. At today’s launch prices, a large orbital data center runs roughly eight to ten times more expensive than its terrestrial equivalent. Closing that gap means getting launch cost down about five times and cadence up about ten. Falcon 9 flies something like 100 to 200 times a year; the math wants 500 to 1,000. “Luckily,” he noted, “that’s also the scale of launch you need to put gigawatts of data centers in space,” so the two numbers move together. It all comes back to the cost of launch.

Engineering problems, not physics problems

This was his refrain, and it is the most useful lens on the debate. Communications is the objection people reach for first. Moving data between spacecraft is easy; the throttle is the regulated, bandwidth-limited link down to the ground. But for AI workloads, most of the heavy traffic stays machine to machine. Starlink’s inter-satellite links already run about 100 gigabits per second, roughly an order of magnitude short of the terabit fabrics inside a data center. A gap, not a wall.

Radiation, the next objection, has quietly inverted. Terrestrial chips are already hardened against cosmic-ray bit flips, because data centers hit them at scale on the ground. Dyer’s favorite proof is that Google’s fleet effectively became “the world’s largest cosmic ray telescope.” “These problems are getting easier over time,” he said, “not harder over time.”

What he will not wave away is that space is still hard, for reasons the easy-button narrative skips. “It’s still a hostile environment,” he said. “These are autonomous robots that we’re sending in space that have to take care of themselves. If you don’t do that right, they will fail, and we can’t go up and fix them.” The deeper challenge is integration, stitching hardware, sensing, compute, and software into one system that works in orbit. “Integration is one of the hardest problems in technology,” he said, “and we just haven’t done it enough in space to get it to the point where it’s muscle memory.”

Not a switch, a progression

So none of this flips overnight. “We’re not going to flip a switch tomorrow and suddenly there’s gigawatts of data centers in space,” Dyer said. “It’s going to be a progression.” National security workloads and data that is simply better kept in orbit will pull the first systems up early, and from there the flywheel of cheaper launch and rack-like satellites compounds. Every hyperscaler will have to run the experiment regardless. “If you’re a hyperscaler looking towards the future and how you stay relevant, you have to be thinking about this,” he said. Google alone is spending something like $100 billion on data center capex this year, he noted, and a very small fraction of that goes a very long way above the atmosphere.

What it means

For all the airtime the data center got, Dyer kept pointing past it. For seventy years, he said, space behaved like “the Galapagos,” a separate ecosystem “detached from the mainland” of the economy. That isolation is ending. “Those things will be so deeply integrated,” he said of space and terrestrial technology, “that you won’t even know the difference anymore.” The corollary is a warning: “Those that don’t have a strategy there are going to get left behind.”

Muon is already building that convergence in the open. Its FireSat constellation pairs orbital sensing with on-board AI compute and SpaceX optical links that, in Dyer’s words, take a satellite “from dial-up to broadband.” Built with Cal Fire, with backing from Google and the Environmental Defense Fund, it is meant to give crews something they have never had: “full situational awareness, overhead views of where the fire is, how hot it’s burning, how fast it’s moving,” updated hourly, day and night. That is Kepler meeting Turing in a single spacecraft.

The same pattern runs through the other industries Dyer rattled off: broadband reaching where fiber never will, a resiliency layer overhead the day someone cuts an undersea cable, supply chains tracked to the individual parcel from orbit, energy infrastructure and markets read from space. The data center is just the piece that made the news. “It feels like we’re at this inflection point where there’s this massive opportunity in front of us in space in a way there’s never been,” he closed, “in a way that I’ve never seen in my career.”

Further reading

Jonny Dyer’s essay that sparked this conversation, When Kepler Meets Turing. 

Bloomberg: Muon Space Raises $250 Million With Google, Salesforce Backing