Why are we putting TPUs in Sun-Synchronous Orbits?
Project Suncatcher, vacuum lasers, and the physics of orbital data centers.
Orbital compute shifts the engineering governor from terrestrial grid connections and zoning laws to payload launch weight, vacuum blackbody heat radiation limits, and radiation-induced Single Event Upsets.
Project Suncatcher: Racks in Space
Rather than imagining massive terrestrial-like structures floating in space, orbital compute consists of modular, satellite-sized compute racks placed in dawn-dusk sun-synchronous orbits (SSO) where their solar panels are perpetually exposed to sunlight.
This represents the primary architectural blueprint of Google's Project Suncatcher and the Planet Labs 2027 prototype testbed, designed to bypass Earth's critical power, cooling water, and land zoning constraints.
Continuous Solar Power Without Batteries
In a dawn-dusk SSO at ~600km altitude, a satellite flies perpetually along the Earth's terminator line. Because it never passes behind the planetary shadow, the solar arrays receive continuous solar illumination.
This eliminates the heavy battery systems required for eclipse cycles on traditional satellites, allowing every kilogram of launch payload to be dedicated directly to compute logic, active voltage rails, or thermal cooling loops.
The Vacuum Thermal Rejection Bottleneck
The ultimate physical constraint of space-based compute is heat rejection. Without ambient air for convection, heat can only escape via blackbody radiative cooling panels, governed by the Stefan-Boltzmann equation.
How much panel that takes is not a constant — it is set by how hot you run the skin and whether you fly edge-on to the Sun. Radiated power scales with the fourth power of temperature (P/A = 2·ε·σ·T⁴, counting both faces). A cool panel left facing the Sun nets only ~350 W/m² after the solar wash and Earth's infrared; a hot panel (~65 °C skin) flown edge-on, so each face sees deep space, rejects ~1,400 W/m² — roughly 4× more from the same area.
SpaceX's AI1 proves the aggressive end: it cools a 150 kW payload with just 110 m² of double-sided radiator, where a naive ~350 W/m² sizing would demand ~430 m². A 40 kW Suncatcher-class rack therefore lands between ~30 m² (hot, edge-on) and ~115 m² (cool, conservative) of panel — comparable to its own solar array, not the dominant structure earlier estimates implied. The price of the small radiator is thermal margin: silicon running near ~90 °C, just under its ~105 °C throttle, on high-reliability liquid loops.
Hardening Commercial Silicon Against Cosmic Rays
Frontier logic manufactured on sub-10nm nodes is highly vulnerable to Single Event Upsets (SEUs) and latch-ups caused by high-energy solar protons and galactic cosmic rays. Hardening is critical to survival.
Rather than relying on legacy, radiation-hardened defense chips which operate multiple nodes behind commercial performance, Project Suncatcher uses commercial-grade silicon (like Google TPU Trillium / v6e clusters) protected by Triple Modular Redundancy (TMR) at the compiler level and active error scrubbing, allowing the payload to tolerate a Total Ionizing Dose (TID) of up to 15 krad for a 5-year orbital service life.
The Starship Economics Crossover
Space-based compute trades electricity and land costs for launch costs. At historical Falcon 9 launch prices (~$3,000/kg), putting a 40kW cluster into orbit is economically non-viable.
However, with fully reusable heavy-lift launch vehicles (Starship) targeting operational flight costs of under $200 per kilogram, the launch capex for a 1,500 kg space compute satellite falls below $300,000. When amortized over 5 years against terrestrial electricity bills and cooling water permits, space compute establishes a superior total cost of ownership (TCO) curve.