Fidelity just gave institutional air cover to the wildest infrastructure bet in tech: training AI in orbit.
The Summary
- Fidelity, a major SpaceX holder, sees viable economics for Musk's orbital data center concept, contingent on Starship reaching full operational capacity
- The plan: bypass terrestrial grid constraints and NIMBYism by running compute in space with direct solar power
- Signal: When Fidelity validates moon-shot infrastructure, follow the constraint removal, not the science fiction
The Signal
The numbers tell you everything. Training frontier AI models now requires 100+ megawatts per cluster. Google, Meta, and Microsoft are buying nuclear plants because the grid can't scale fast enough. They're negotiating with utilities, communities, and regulators just to plug in more servers. Musk looked at that bottleneck and decided gravity itself was the problem.
Fidelity's backing isn't about believing in orbital romanticism. It's about watching Starship launch costs crater toward $10 million per flight while terrestrial data center permitting stretches past 18 months. The institutional calculus is shifting. If Starship hits projected reusability, lifting compute hardware becomes cheaper than fighting zoning boards.
The cooling problem actually gets easier in space. Radiate heat directly into the void. No chillers, no water rights battles, no worried neighbors. Solar works 24/7 above the atmosphere. The constraint isn't physics anymore, it's whether SpaceX can turn Starship from prototype into cargo truck.
This matters beyond Musk. Once orbital infrastructure gets institutional validation, the race starts. Every hyperscaler with a sovereign AI strategy will run the math. China's already testing space-based computing. The constraint removal thesis is simple: whoever solves compute scarcity first wins the model race. If that means leaving the planet, capital will follow.
The Implication
Watch Starship's flight cadence and payload costs more than the AI headlines. If SpaceX hits 50+ flights in 2026, orbital compute shifts from wild idea to viable hedge against grid-locked competitors. For everyone else building agent infrastructure, the message is clear: traditional constraints are negotiable if you're willing to rethink the entire stack. The companies that win Web4 won't be the ones with better prompts. They'll be the ones who solved the power problem everyone else assumed was fixed.
Source: Bloomberg Tech