Source: NYT > Business (paywall)
Caltech and Sophia Space are betting that space's vacuum and solar power can solve the thermal bottleneck in AI infrastructure. Data centers consume 4-5% of US electricity, and cooling accounts for roughly 30% of that load. If deployed at orbital altitudes, the system could decouple AI compute from terrestrial power grids and water scarcity constraints, potentially shifting where hyperscalers build training facilities and lowering the cost of model development. The bet assumes space infrastructure costs drop enough to compete with incremental grid expansion—contingent on cheap launch economics and orbital real estate becoming viable commodity markets.