Behind-the-meter generation is displacing grid connections as the default architecture for new AI data centers -- fuel cells, recip engines, and modular gas plants bypass the 3-5 year interconnection queue, with 10+ GW in active development

A paradigm shift is underway in how data centers are powered. Oracle's Jupiter campus (2.45 GW, 100% Bloom Energy fuel cells) defines a no-grid, no-diesel, no-battery paradigm. Microsoft partnered with CVX for behind-the-meter generation in West Texas. Applied Digital is building a 1.2 GW gas-fired IPP (Base Electron). Core Scientific is pursuing behind-the-meter power. MARA acquired Long Ridge with an operating gas power plant. Liberty Energy reserved 400 MW for Vantage projects. ProPetro secured a 2.1 GW Caterpillar framework agreement. The driver is simple: grid interconnection queues are 3-5 years, and hyperscalers need power now. Behind-the-meter generation eliminates the queue entirely. Bitcoin miners pivoting to AI colocation (CIFR, IREN, CORZ, RIOT, WULF, CLSK, MARA, BTDR, APLD) are uniquely positioned because they already control multi-GW power positions, often with ERCOT batch-0 interconnection rights that cannot be replicated. The implications cascade through the power equipment chain. Gas turbine and reciprocating engine capacity is sold out through 2028+, creating an oligopolistic bottleneck for CAT, GEV, and CMI. Compression equipment lead times exceed 180 weeks. OEMs are raising long-term growth algorithms 6-12 months after Investor Days, signaling that AI power demand has structurally exceeded their initial estimates. The convergence of oilfield services companies (SLB selected as NVIDIA DSX AI factory modular design partner, targeting $1B data center run rate) and data center operators into the power generation market is a structural, not cyclical, development.