The most quietly significant deal of the week had nothing to do with rockets. SpaceX signed a $150 million per month compute contract with Reflection AI, an open-source lab, for access to Nvidia GB300 chips starting July 1. That number — $1.8 billion annually — is larger than most nation-state space budgets. The new space race is not about orbit. It is about who controls the substrate.

Infrastructure as the New Territory

Read alongside federal regulators fast-tracking AI data center grid connections, the picture sharpens. The US government is essentially subsidizing the energy infrastructure for private AI compute, while SpaceX monetizes its satellite and computing assets to fund AI labs. The loop is getting tighter: AI needs energy, energy needs grid upgrades, grid upgrades get fast-tracked for AI, and the companies that own the physical layer — satellites, chips, power connections — are becoming the new landlords of the digital economy. A 2024 paper in Energy Research and Social Science by Brevini and Pasquale found that AI infrastructure investment is concentrating in a handful of vertically integrated players who simultaneously produce, consume, and sell compute — a structural dynamic that looks less like a market and more like a utility monopoly in formation.

Open Source as Cover Story

Reflection AI's open-source positioning is worth interrogating here. Openness at the model layer does not offset closure at the infrastructure layer. If your training runs on SpaceX's compute at $150 million a month, the open-source label is a marketing gesture, not a structural commitment to access. that is making this dynamic increasingly visible to early-stage founders trying to navigate the AI stack. The question is not whether Reflection's models will be open. It is whether open models running on closed infrastructure constitute openness in any meaningful sense — a tension that Brewster Kahle has been flagging for years in the context of public AI and the global brain.