Oluwadi
Investigating the Overlooked
Region
What
Who For
Analysis
Oluwadi
Investigating the Overlooked
Oracle invoked force majeure on a data center it has not finished building. The clause is the one contracts reserve for hurricanes, wars, and acts of God; Oracle reached for it at its New Mexico Stargate site over energy and permit delays, with reporting tracing the failure to a 17-mile gas pipeline that could push the launch past 2028.[1] That is the state of the AI buildout on the ground: the most aggressive infrastructure program in a generation, stalled not by capital or chips but by pipe, substations, water, and county paperwork. So one of the largest builders in the business has decided to stop fighting the ground entirely. On October 1, Google puts a data center in orbit.
The vehicle is a SpaceX Falcon 9, and the payload is four of Google's own Tensor Processing Units — the AI chips that normally sit in a hyperscale hall in Oregon or Iowa — flown as the first live test of a program Google calls Project Suncatcher.[2] It is not a stunt. Google's stated vision is kilometer-scale arrays of 81 satellites flying in tight formation, each carrying TPUs, wired into orbital compute clusters and run on sunlight that never sets.[3] The October launch is the stress test: can the silicon survive radiation and run in vacuum before anyone builds the constellation.
Read the logic and it stops sounding like science fiction and starts sounding like a real-estate decision. In low Earth orbit there is no interconnection queue, because there is no grid. There is no noise ordinance, because there are no neighbors. There is no water table to draw down for cooling, because you reject heat by radiating it into vacuum. And the sun delivers nearly round-the-clock, full-intensity power to a panel that never sees a cloud or a night. Every hard constraint that has turned Data Center Alley into a permitting war — power, cooling, water, land, and the people who live beside it — weakens or vanishes the moment the rack leaves the planet.
I have stood at the wall this is trying to climb over. As a director at a managed-services provider operating inside RagingWire — later NTT — in Ashburn, the heart of Data Center Alley, I watched some of the most advanced colocation space in the world fail to deliver the water cooling or hold the temperatures a full GPU rack demands. We ended up leaving four rack units unpopulated between GPU machines just to give them enough air — capacity the customer had paid for and could not use. When the densest, best-resourced halls in the country cannot cool the newest chips, "put it where cooling is free" stops being a joke and starts being a line on a spreadsheet.
The catch is real and worth stating plainly. Lifting mass to orbit is expensive, orbit is far from every user so latency rules out the real-time inference that most AI actually serves, radiation degrades silicon over time, and no technician is ever going to fly up to reseat a failed board. The economics pencil only for the enormous, latency-tolerant training runs — which happen to be exactly the workloads breaking the grid today. Orbit does not replace the ground. It escapes one specific, power-hungry corner of it.
Which is the asymmetry worth sitting with. Space did not suddenly get cheap. The ground got impossible. When the marginal data center pencils out better above the atmosphere than in Virginia, that is not a triumph of the space economy — it is an indictment of how thoroughly we have run out of room, power, and water down here. Google launching TPUs on a rocket is the loudest possible signal that the terrestrial buildout has hit a physical ceiling, and that serious money is now willing to pay launch costs to get above it.