AINEWS
Sophia Space and Caltech just patented a chip cooling tile for orbital data centers, but no mass budget, data round trip time from orbit to ground, or on orbit test exists before 2030.
Artificial intelligence keeps asking for more compute, and the places hosting that compute are pushing back. In Henrico County, Virginia, residents have watched 37 data centers come online while their electrical bills climbed; state legislators in Virginia, Texas, and Georgia have started writing siting and water-use rules. Power bills are climbing, water tables are dropping, and the friction is moving into state law. That pressure is what is pushing a long-running idea, compute in orbit, back into the engineering conversation.
The hard part was never launch. It was heat.
A satellite in low Earth orbit sits in vacuum. There is no air to blow across a heatsink, no water to run through a cold plate. The only way a chip can dump waste heat is to convert it to infrared radiation and shine it into space. That requires a radiator: a large, flat surface tuned to the infrared band, oriented away from the Sun, and shielded from the Earth's own thermal glow. Bulky radiators are the reason orbital data center concepts have, until now, read more like pitch deck than product.
A patent filed jointly by the California Institute of Technology (Caltech) and Sophia Space, a California startup developing orbital computing, tries to collapse that radiator into the same physical unit as the solar cell and the processor. The product is called Sophia TILE. Each tile carries its own solar collector on one side, its own compute on the other, and radiates waste heat from the edges into deep space. Thousands could be clustered into a large orbital data center; smaller constellations could serve edge or regional workloads. The whole stack runs on continuous solar power with no grid tie.
The design move is coherent. If every tile carries its own radiator, the system does not need a single huge radiator structure. Heat is rejected at the source, where the chip is, instead of piped through a loop to a remote panel. Mass and volume shrink. The deployment model, a swarm of identical self-contained tiles, scales with cheap launch and standardized manufacturing.
The press release around the patent is, however, where the engineering meets the marketing. Sophia Space's stated vision is to test the design "by 2030." No flight hardware exists. No on-orbit thermal performance number has been published. No mass budget per tile, no launch cost, no beaming latency from orbit to ground, and no peer-reviewed validation of the radiator's effectiveness in vacuum has been released. Sophia Space's founder and CTO, Leon Alkalai, frames the patent as the first step in a longer program, and the team is still weighing the radiator mass budget and the rest of the system tradeoffs.
Every orbital data center claim, from this one on, can be tested against a small set of questions: does it name a heat-rejection mechanism, or only a slogan? Is the radiator on the same structure as the compute, or somewhere else in the diagram? What is the per-tile mass, and does the launch cost fit? What is the round-trip latency to a ground user, and is the downlink infrastructure assumed or specified? The Sophia TILE patent is the first proposal to clearly answer the first two. The rest are open.
If Sophia Space and Caltech put a tile in orbit before 2030 and publish a measured thermal rejection number, the orbital data center story moves from concept to engineering. If the 2030 test produces no on-orbit performance data, the field is still vapor. Watch the patent's continuation filings and any Caltech thermal-vacuum chamber test results in the next 12 to 24 months. Then the radiator claim, and not the press release, will be tested.