The economics make no sense yet, but Google's betting that gravity costs more than rocket fuel in the long run.
The Summary
- Google and SpaceX are negotiating to launch orbital data centers, starting with test products despite costs far exceeding terrestrial infrastructure today
- The talks focus on using SpaceX's launch capacity to position space as future compute real estate for AI workloads
- This signals Google's long-term bet that AI training bottlenecks, energy constraints, and cooling costs will eventually make orbit economically viable
The Signal
Google's pivot from building yet another Iowa server farm to negotiating rocket launches tells you everything about where AI infrastructure is headed. The company's in talks with SpaceX to put data centers in orbit, according to people familiar with the matter. Both companies are framing this as positioning space for future AI compute, even though today's launch costs dwarf what you'd pay to drop servers in a warehouse.
The math seems insane until you map it forward. Every AI model generation needs more chips running hotter for longer. Data centers already consume 1-2% of global electricity. Cooling alone accounts for 40% of a typical facility's power draw. Space offers infinite cooling, zero real estate cost, and solar power 24/7 with no atmospheric loss.
"Costs today remain far higher than on the ground, but Google's making the bet anyway."
Here's what makes this more than vaporware:
- SpaceX's Starship just dropped launch costs to roughly $10M per flight with 100+ ton payload capacity
- Google's already burning billions on custom TPU chips and proprietary infrastructure
- The company's proven it will build radically custom hardware (subsea cables, custom ASICs) when scale demands it
The timeline matters. These are test products, not production systems. Google's not moving Gmail to orbit next quarter. They're buying optionality on infrastructure that makes no sense in 2026 but might be the only way to scale foundation models by 2030. When training runs start requiring gigawatt-hours and you're fighting nation-states for power grid access, paying SpaceX to launch your compute starts looking reasonable.
The Implication
Watch for other hyperscalers to follow. If Google's exploring orbital compute, Microsoft and Amazon aren't far behind. The question isn't whether space data centers happen, it's who locks in launch capacity first. SpaceX now has negotiating leverage with every AI company facing power and cooling constraints.
For builders in the agent economy, this is your signal that infrastructure assumptions are all in play. The companies training the models your agents run on are already planning for a world where compute happens in orbit. Plan accordingly.