Poolside Integrates Power and Compute

Project Horizon turns Poolside from a buyer of scarce AI infrastructure into a company helping control the supply. For a model lab selling custom coding systems into banks, defense contractors, and other tightly controlled environments, locking in 15 years of power and a first tranche of 40,000 GB300 NVL72 GPUs matters because frontier training is constrained as much by electricity and site access as by model talent.

• Poolside already sells a highly controlled product, on premises, in private VPCs, or through Bedrock, with forward deployed engineers helping customers wire the system into internal repos and CI/CD. Owning more of the compute stack fits that same pitch, tighter control over where models are built and how they are run.
  1 sacra
• CoreWeave is the kind of partner built for this move. It had roughly 850MW of active power capacity and 3.1GW of contracted power by Q4 2025, plus a $66.8B backlog that shows how quickly large AI buyers are trying to reserve long dated capacity before competitors do.
  2 sacra
• This also shows Poolside moving up the compute market curve. Smaller GPU clouds and brokers help startups get online fast, but the biggest labs eventually graduate to dedicated clusters, long contracts, and utility style infrastructure. Fluidstack, CoreWeave, and Crusoe are all converging on that model as power becomes the bottleneck.
  3 sacra 4 sacra

The next phase of the AI stack is likely to look more like energy infrastructure than ordinary software. Labs that can pair proprietary models with guaranteed megawatts, reserved GPU fleets, and controlled deployment environments will be in a stronger position to win large enterprise and government workloads, and Poolside is building toward that end state early.