Deep Dive
The Core Problem: Cooling a Supercomputer in Vacuum
Real Engineering breaks down why Starcloud's 5-gigawatt orbital data center concept is fundamentally flawed by starting with physics. An Nvidia NVL-72 server rack draws 120 kilowatts of power — equivalent to 60 suburban homes — but in orbit, that energy becomes heat with nowhere to go. In a vacuum, convection is impossible; the only solution is thermal radiation using Stefan-Boltzmann equations. To keep servers at 20 degrees Celsius while radiating 5 gigawatts requires panels 4 kilometers tall and 840 meters wide. Starcloud's white paper doesn't just underestimate this problem — it explicitly ignores it, claiming passive cooling is possible without heat pumps. The actual solution would require circulating 68,870 kilograms of coolant per second, equivalent to emptying an Olympic pool every 40 seconds, demanding pumping power equivalent to 134 rocket turbopumps. None of this appears in their engineering analysis.
The Mass Fantasy: Starcloud's Numbers Don't Add Up
The creator dismantles Starcloud's cost estimates by checking their claims against real hardware. Starcloud asserts 400 watts per kilogram of compute, but the exact Nvidia servers they reference achieve only 88 watts per kilogram — a factor of 4.5 error. Solar panels fare worse: they quote 1000 watts per kilogram when the best lab prototypes manage 300, and operational systems like ROSA arrays deliver just 100. Using realistic figures, their 5-gigawatt array would weigh 50,000 tons instead of 5,000. The radiator system alone — 4 kilometers by 840 meters — weighs 13,440 to 33,600 tons depending on material. Excluding pumps, coolant, radiation shielding, fuel, inertia wheels, and structural components, the total mass reaches 113 million kilograms: more tonnage than an aircraft carrier, exceeding six times all mass humanity has ever launched into orbit combined.
Launch Costs and the Radiation Nightmare
Even if the hardware existed, launch costs destroy the economics entirely. Starcloud quotes $30 per kilogram, but an actual Starship contract with Voyager Technologies costs $900 per kilogram. At realistic rates, launching 113 million kilograms costs 102 billion dollars — not the implied 3 billion. Beyond cost, space degrades everything aggressively. Atomic oxygen chemically attacks surfaces, Van Allen belt particles damage radiators and solar panels, ultraviolet and cosmic rays degrade performance. The white coating AZ-93 that Starcloud specifies drops from 0.92 to 0.90 emissivity over time. More critical: ionizing radiation forces satellites to run three simultaneous calculations on different processors to detect bit flips and corrupted data, tripling power consumption and mass. The HP Edge servers aboard the ISS already do this; it's a brute-force workaround proving that space computing is inherently fragile and inefficient compared to ground processing.
Orbital Mechanics and Structural Chaos
Starcloud's design ignores orbital mechanics entirely. A satellite with massive flat solar panels and radiators creates aerodynamic drag forcing constant boost burns just to maintain orbit — the ISS does this repeatedly. The satellite's moment of inertia would be enormous, making attitude control via inertia wheels impractical; they scale with rotational mass, and this station carries hundreds of thousands of tons of fluid flowing to distant radiators. Earth's lumpy gravity field creates uneven gravitational pulls on far-flung edges, adding forces that pull the satellite off-course. Starcloud glosses over this as still being in development, but Google's Suncatcher concept addresses it differently: they propose a constellation of 81 smaller satellites in a bounded orbit that naturally resonate together without heavy structural supports. Even then, Google must precisely nail initial conditions. The trade-off between one massive station and a networked constellation remains unsolved, and adding more satellites to busy sun-synchronous orbits increases collision avoidance complexity — SpaceX reported 300,000 collision avoidance maneuvers in 2025 just for Starlink.
Why Space Data Centers Might Actually Matter
Despite evisceration of Starcloud's execution, Real Engineering acknowledges the real strategic case for orbital computing. Terabytes of satellite imagery and synthetic aperture radar data are being collected constantly — when Iran was struck, countries worldwide scrambled to image the event, generating enormous datasets. Processing this intelligence in orbit before downlinking only relevant results reduces bandwidth bottlenecks. More critically, autonomous warfare is already here: Ukraine recently deployed fully autonomous drones using satellite intelligence for real-time battlefield decisions without human guidance. Military applications demand speed; whoever processes geopolitical data fastest wins. Orbital data centers with uninterruptible solar power become nearly untargetable strategic assets during conflict. The military has deep funding pockets and views this as essential infrastructure, not a profitable commercial venture. This changes the calculus entirely — not from venture capital returns, but from national security perspective, making some form of orbital processing inevitable despite current engineering immaturity.