Advanced nuclear, geothermal, and gas with carbon capture are positioned as medium-term entrants as developers seek more scalable and lower-carbon BTM profiles
The System Pressure
The North American data center market has expanded at a 37% compound annual growth rate since 2021. By mid-2026, the region supports roughly 70 GW of operational and pre-leased capacity, with 17 GW under construction and more than 200 GW advancing through planning and development. That pipeline is enormous. The infrastructure to serve it is not.
Vacancy rates tell the compression story cleanly: from approximately 6% in 2021 to 1% in 2026. At 1% vacancy, the market is functionally full. New supply cannot arrive fast enough because the gating factor is no longer capital, land, or compute hardware—it is power, and specifically the certainty and timing of its delivery. Development timelines have stretched to five to ten years as projects stack up against permitting delays, interconnection backlogs, and supply chain constraints. For operators whose AI infrastructure decisions are made in months, not decades, those timelines are structurally incompatible with the business.
FERC’s recent determination that PJM’s existing tariff lacked clear and consistent rules governing co-located generation and large-load facilities added regulatory weight to what operators were already experiencing operationally. PJM projects approximately 85 GW of demand growth over the next 15 years. NERC’s latest long-term reliability assessment forecasts more than 224 GW of summer peak load growth across North America over the next decade—more than 69% above the prior forecast. These are not incremental revisions; they signal a fundamentally different demand environment that front-of-the-meter infrastructure was not designed to serve at this speed.
The Drivers, Dependencies, and Constraints
Behind-the-meter power has moved into the gap. Reciprocating engines, aeroderivative gas turbines, simple-cycle turbines, and conventional generators currently account for most near-term AI data center BTM deployments. Reciprocating engines have gained particular market acceptance as the fastest speed-to-power option, partially displacing aeroderivative turbines that face OEM backlog constraints of their own. Fuel cells are emerging for edge deployments. Advanced nuclear, geothermal, and gas with carbon capture are positioned as medium-term entrants as developers seek more scalable and lower-carbon BTM profiles.
The strategic logic of BTM has shifted away from cost arbitrage. The question operators are now asking is not whether onsite generation is cheaper than grid power—it is whether getting reliable power two to five years sooner justifies the incremental investment. For a growing number of AI developers, the answer is yes. BTM also reduces exposure to transmission and distribution charges, system upgrade costs, and ISO and utility fees, though these savings must be weighed against the capital burden of owning generation assets and managing fuel supply.
The reliability architecture changes as well. The U.S. grid delivers approximately 99.95% average availability—a figure that seems strong until measured against the four-nines and five-nines requirements of mission-critical AI workloads. Traditional data center designs layer UPS systems, battery storage, and diesel backup on top of a single utility feed to close that gap. BTM campuses reframe the architecture: multiple modular generation assets, integrated storage, and advanced microgrid controls create redundancy across the generation fleet itself rather than backing up a single grid connection. The loss of one unit in a distributed BTM campus carries less operational consequence than a single-point grid event. That said, operating large-scale BTM microgrids as integrated utility-scale systems supporting hyperscale AI loads remains a relatively new practice. Achieving consistent three-nines reliability at this scale will require continued operational experience and architectural refinement as deployments accumulate.
The capital structure introduces its own dependencies. BTM generation assets are typically purpose-built for a specific campus under long-term arrangements, which means they carry limited alternative uses if a customer changes site strategy, consolidates workloads, or shifts technology requirements. Stranded-asset risk is real and must be priced. Developers frequently must commit capital to credible powered-land positions before long-term customer commitments are in place, creating a coordination gap between operators who need power certainty and investors who need demand certainty. Contractual mechanisms—delivery commitments, availability guarantees, liquidated damages, and capped remedies—are becoming the standard language for bridging that gap and making BTM financeable at institutional scale.
Open Dependencies
Several material questions remain unresolved and should inform how aggressively operators commit to BTM architectures today. FERC has directed PJM to formalize co-location rules for large-load facilities, but the final rule structure, its timeline, and how other RTOs respond are not yet confirmed. Regulatory alignment across PJM, ERCOT, MISO, and the Western Interconnection would materially affect which BTM structures are permissible, financeable, and operationally advantageous in each jurisdiction.
The emissions profile of gas-based BTM is a second open variable. Most near-term deployments rely on fossil generation. Whether carbon accounting frameworks, state clean energy standards, or operator sustainability commitments will constrain or condition BTM expansion is not answered by current evidence. Operators with 24/7 carbon-free energy targets face a genuine tension: BTM gas generation solves the availability problem while creating a Scope 2 problem.
The long-term relationship between BTM and the grid also remains in flux. Most hyperscalers continue to view utility interconnection as the preferred long-term power source for scale and efficiency, positioning BTM as complementary—a bridge, not a replacement. Whether that framing holds as BTM platforms scale and vertical integration deepens depends on technology maturation, regulatory evolution, and the pace of grid investment, all of which remain uncertain.
The Operating Exposure for Global Heads of Data Center Energy
The shift from energy transactions to capacity transactions is the most important structural change for procurement strategy. Earlier data center power agreements were priced to hedge electricity cost. AI-era agreements are priced for guaranteed availability—gigawatt-scale, twenty-year-plus commitments where the premium is not for cheap electrons but for certain ones. This changes the counterparty risk profile, the contract structure, the capital exposure, and the valuation logic entirely.
For energy heads managing multi-GW portfolios, BTM introduces a new category of ownership responsibility: generation asset management, fuel supply continuity, microgrid operations, and interconnection strategy are now portfolio questions, not just individual site questions. The organizations moving fastest are those building repeatable BTM platforms—standardized technology stacks, commercial structures, and financing templates that can be replicated across sites rather than renegotiated from scratch at each project.
The competitive consequence of moving slowly is concrete. At 1% vacancy and five-to-ten-year grid development timelines, operators who cannot access reliable near-term power face site selection constraints that compound over time. Emerging markets with faster power deployment potential are attracting capacity that established hubs can no longer accommodate. Decisions made now about BTM platform development, IPP relationships, and contract structures will determine interconnection and capacity positioning through the early 2030s.
Signals the System Is Shifting
Watch FERC’s PJM co-location rulemaking for the first formal framework governing BTM large-load arrangements—what it permits, how it prices grid backup service, and whether it becomes the template other RTOs follow. Watch reciprocating engine OEM capacity and lead times as a proxy for BTM deployment momentum; if supply constraints emerge there, the speed-to-power advantage narrows. Watch how major hyperscalers characterize BTM in sustainability reporting—if gas-based onsite generation begins appearing as a Scope 2 liability rather than a deployment enabler, regulatory or board pressure could reshape the technology mix. And watch the financing market: as BTM projects increasingly attract institutional infrastructure capital through long-term contracted cash flows, deal structures will standardize rapidly, reducing the negotiating leverage available to later movers.
Sources
- Fticonsulting — BTM Solutions are Accelerating Data Center Deployment (Link)
