Google Renting Compute from SpaceX at $920M/Month
Google has signed a deal to pay SpaceX $920 million per month for GPU access, with a hard deadline of September 30, 2026, after which Google can terminate if SpaceX fails to deliver. The contract includes a one-month grace period but is otherwise unusually punishing. The thread on HN immediately zeroed in on what this says about Google's internal compute situation: if you can't build or buy the DRAM you need, you rent your entire stack from a competitor.
The pattern here is that the GPU shortage has created an entirely new class of infrastructure dependency. Google, one of the most sophisticated infrastructure operators on the planet, is paying a rocket company almost $1 billion a month for compute. That is not a normal vendor relationship. That is a hostage situation dressed up as a contract.
The HN discussion also raised a fair question: what is SpaceX's core business now? Rockets, Starlink, and now wholesale GPU leasing to hyperscalers. The answer is increasingly 'whatever the market needs most right now.'
So what?
If Google is this exposed to compute scarcity, every founder building on cloud infrastructure should be thinking hard about lock-in and supply risk. The cost structure of AI-native products is not stable. Renegotiate your compute contracts before your vendor's vendor runs out of GPUs.