The hardest problems in space compute have no ticker
June 14, 2026
Putting AI data centers in orbit has gone from sci-fi to a funded race — but the engineering bottlenecks and the investable names point in opposite directions.
Why it matters: Every public "space compute" stock clusters around the problems that are already mostly solved. The two hardest problems — keeping chips cool and keeping them alive in radiation — are where you basically can't put money to work.
The big picture: Orbital compute is really a five-layer stack sitting on one foundation. Get the foundation wrong and nothing above it matters.
Driving the news: Jensen Huang declared at GTC that "space computing, the final frontier, has arrived." SpaceX, Google's Project Suncatcher, Starcloud and Blue Origin have all filed for compute constellations. First real hardware tests hit in 2026–27.
The stack, bottom to top:
Launch — the master gate. Needs to fall below ~$200/kg to work. SPCX and RKLB are the plays. Cheap launch dissolves the problems above it.
Thermal — the hardest engineering problem. No convection in a vacuum; you radiate heat or you cook. A 1MW center needs ~100 tons of radiators. Best names: RDW, RKLB. The real innovators are private.
Power. Near-constant orbital sunlight, but arrays bigger than the ISS. RKLB and RDW. Closest to solved.
Rad-hard compute. Frontier chips degrade in under three years in orbit; hardened versions cost 5–10x and lag generations. No clean pure-play exists.
Connectivity. Optical inter-satellite links. RKLB (via its Mynaric buy) and CIEN. The cleanest picks-and-shovels.
The twist: Investability runs inverse to difficulty. The market is most buyable exactly where the engineering is most finished.
The catch: This whole basket trades on SpaceX sentiment, not orbital-compute earnings — which don't exist yet. When SPCX's valuation target slipped below $2T, the entire group sold off overnight.
The bottom line: The names that have run hardest are levered to the solved parts of the problem, not the binding constraints. If a pure-play thermal or rad-hard-AI-silicon company goes public, that's the asymmetric one to watch.