SemiAnalysis reports that behind-the-meter onsite power has gone from fringe experiment to mainstream practice for every major AI lab and hyperscaler, with 75GW of firm binding OEM orders (about 20GW added in Q2 2026 alone) and roughly 3GW of operational US datacenter capacity expected by year end. The driver is grid interconnection timelines of five-plus years versus the economics of inference revenue, with Microsoft, Google, OpenAI, Anthropic, and Meta all signing multi-GW onsite gas, turbine, and fuel-cell deals. The report also catalogs emerging execution risks: permitting delays, fuel pipeline shortfalls, financing bottlenecks for developers, labor shortages, and a secondary market forming for turbines from failed projects. For financial institutions, this confirms power procurement as a binding constraint on AI deployment timelines and a new multi-billion-dollar-per-gigawatt capital requirement embedded in project financing.
150 other tracked developments touching the same claims.