CoreWeave's Partnered Power Strategy

This shift means CoreWeave is no longer trying to buy control of power in one move, it is stitching together a power estate site by site, with long leases, new builds, and anchor partnerships. After the Core Scientific deal failed, the company kept pushing ahead with developments that lock in megawatts directly, from a 250MW first phase in West Texas to a 100MW Pennsylvania site that can expand to 300MW, plus new UK capacity commitments.1 2 3 4 5 6

• The abandoned Core Scientific merger shows what CoreWeave wanted. The deal would have given it about 1.3GW of gross power and more than 1GW of expansion potential, plus direct ownership of powered data center shells instead of relying on leased capacity. That would have made financing easier and lowered infrastructure dependence on third parties.2 3 7
• Project Horizon shows the replacement playbook. Instead of buying a whole operator, CoreWeave is acting as anchor tenant in a giant campus built with partners. Poolside said the project targets 2GW in West Texas, uses on site natural gas generation, and gives CoreWeave the first 250MW with another 500MW reserved under a 15 year lease.1 4 8
• This is also how CoreWeave is differentiating from other GPU clouds. Crusoe is built around cheap stranded gas power, while CoreWeave is assembling capacity across multiple models, including leased sites, regional buildouts, and partner campuses. That matters because in AI cloud, the scarce input is often not GPUs, but ready power with cooling and interconnects attached.4 5 9 10

The next phase is likely a more utility like CoreWeave, with growth determined by how fast it can convert reserved megawatts into live clusters. If it keeps landing anchor customer demand while bringing Pennsylvania, UK, and West Texas capacity online, power procurement becomes the moat that is hardest for smaller GPU clouds to copy.1 4 5 6