Server farms in low Earth orbit could run on continuous sunlight and radiate heat directly into space, skipping the multi year wait for a grid hookup. The harder question is whether the physics and economics actually close.
A developer in northern Virginia has the land, the capital, and a customer willing to pay for a new AI training cluster. What it does not have, after three years of paperwork, is a wire to the grid. The wait is not unusual: some U.S. interconnection queues now stretch past a decade, and the cost of bridging that wait is one of the quieter drivers behind a fresh pitch from a small but vocal group of space startups. Move the data center to orbit.
That pitch is gaining visibility because SpaceX, the company that proved the case with Starlink, has begun talking publicly about putting compute in space, too. In a recent SpaceNews essay, the founder of orbital-energy startup Overview Energy argues that an "orbital energy economy" is the next step beyond orbital data centers, with continuous sunlight in space and the ability to radiate waste heat directly into the vacuum as the new enabling physics. He points to Starlink's 10,000-satellite fleet as proof that hardware can be launched, upgraded, and replenished on orbit at a pace that resembles a rolling infrastructure program rather than a series of flagship missions. He also points to falling launch costs and runaway AI compute demand as the economic engine.
The essay is an opinion piece from a startup CEO with a vested interest in the answer, and it is worth naming that openly. The honest question is not whether the boosters are excited. It is what the math and the physics actually require for orbital compute, or for the older idea of space-based solar power beaming energy down to a receiving station, to beat the ground-plus-grid-plus-storage alternative in this decade.
Two constraints show up immediately. The first is heat. A server farm in low Earth orbit does not have air to blow across its chips. It has to radiate heat from radiator panels into deep space, and the area of those panels, not the solar arrays on top, tends to set the size of the system. No independent thermal engineer has, on the record, named the radiator floor for a multi-megawatt orbital data center. Without that number, the megawatts that advocates cite are an aspiration, not a bill of materials.
The second is money. The argument from launch-cost decline is real. SpaceX has compressed launch costs per kilogram to a level that lets Starlink replace satellites regularly, and it has talked about orbital data centers as a natural follow-on. But the comparison that matters is not launch cost per kilogram. It is the levelized cost of a delivered megawatt-hour of compute, including the radiator, the structure, the launch, the operations, the deorbit insurance, and the ground backup for eclipse and storm. That number has not been published by an independent analyst. Without it, the case for orbit is a slope on a chart, not a budget.
Grid queue, the third leg, is the one that gives the orbital idea its opening. Hyperscalers and utility planners both acknowledge that some U.S. markets now have interconnection waits in the seven-to-ten-year range, and that storage-plus-onsite generation is the current workaround, not a final answer. If those waits persist, the option value of a compute platform that bypasses the grid entirely goes up. If queue reform actually lands in the next few years, the case for orbit shrinks. Which one a developer should bet on is a question that depends on state-level regulation the orbital startups do not control.
What to watch next: any published, peer-reviewed or independently modeled levelized cost for orbital compute at the megawatt scale, and any on-record statement from a utility planner or thermal engineer naming the radiator or capex floor below which the orbital option cannot pencil out. Until those appear, "orbital data centers" is a marketing term for a question the industry has not yet answered in numbers a CFO can sign.