Under the first, a third-party generator develops, owns, and operates the generation asset under a long-term PPA with the data center developer
The System Pressure
The foundation of conventional data center power strategy — connect to the grid, negotiate an interconnection agreement, wait — is breaking down under simultaneous pressure from demand growth, aging network infrastructure, and a multi-year queue backlog that shows no near-term resolution.
According to the International Energy Agency, data centers consumed just under 1.5% of global electricity in 2024, a figure that grew by roughly 17% through 2025 and is projected to approximately double by 2030. At the facility level, the trajectory is equally stark: new hyperscale AI facilities routinely exceed 100MW at opening, with projects like OpenAI’s Stargate initiative targeting up to 10GW — a load profile the existing interconnection framework was never designed to absorb at this pace.
Meanwhile, the grid itself is structurally underinvested. The IEA has estimated that global grid investment would need to reach approximately USD 600 billion annually by 2030 — roughly double early-2020s levels — simply to support climate and energy targets. That figure does not account for the additional strain that concentrated, gigawatt-scale data center load places on regional transmission systems designed for a mid-twentieth-century generation model. The result is a widening gap between what the grid can deliver and what AI infrastructure requires.
The Drivers, Dependencies, and Constraints
Behind-the-meter frameworks offer a specific structural response: a dedicated generation asset co-located with the data center, either bypassing the public grid entirely or using it only as backup. The appeal is direct — circumventing congested queues, reducing exposure to volatile wholesale prices, and giving operators meaningful control over their own power supply timeline.
But the model introduces a distinct set of interdependencies that do not appear in a standard grid-connected project. Two primary commercial structures are emerging. Under the first, a third-party generator develops, owns, and operates the generation asset under a long-term PPA with the data center developer. The creditworthiness of the offtaker becomes the financing anchor; lenders focus intensely on PPA performance obligations, curtailment allocation, and force majeure terms — along with stranded asset risk if the data center ceases to operate and the generator has no grid connection or alternative purchaser. Under the second model, the data center developer owns the generation asset directly, either building from the ground up or acquiring through a build-transfer agreement. This concentrates risk but also control, and forces lenders to underwrite a hybrid credit profile that combines data center real estate exposure with operational power generation risk.
Technology choice creates additional dependencies. Renewable-only BTM configurations face the fundamental challenge of reconciling intermittency with the “five-nines” uptime standard that AI GPU clusters demand. Conventional short-duration battery storage cannot yet reliably bridge that gap. Long-duration energy storage is approaching commercial viability but has not yet scaled to the load requirements of a hyperscale facility. Natural gas-fired generation addresses the reliability constraint but introduces fuel-supply infrastructure requirements, volatile commodity exposure, and emissions obligations that complicate sustainability reporting. Small modular reactors represent a longer-horizon option; regulatory maturity and cost competitiveness are unlikely to make them commercially viable before the 2030s in most jurisdictions.
On top of technology constraints, gas turbine procurement has emerged as a hard bottleneck. Manufacturers are reporting soaring order backlogs, with delivery timelines extending years beyond order placement. A secondary market offering reduced contractual protections has emerged in response — a signal of how acute the equipment shortage has become.
Open Dependencies
Several assumptions underlying current BTM planning remain unresolved. The regulatory treatment of BTM frameworks is the most consequential. Most power sector regulatory regimes were built around centralized grid models; BTM arrangements frequently fall into existing gaps or trigger licensing requirements not designed for this use case.
FERC’s December 2025 order — along with subsequent PJM proposals — requires large co-located loads above defined thresholds to pay for transmission services, marking a significant departure from prior practice. The financial implications for projects that assumed cost separation between the generation asset and the broader transmission system are material and, as of this analysis, still working through implementation. Whether other ISOs and RTOs adopt analogous frameworks is an open question that could reshape BTM economics across multiple markets.
The obligation to maintain backup grid connection is also unresolved across jurisdictions. In Ireland, where data centers account for roughly a quarter of national electricity demand, regulators already require on-site backup power or battery solutions. Across Europe, the principle that BTM facilities benefiting from grid backup should bear a greater share of network charges is reportedly gaining traction but has not been uniformly codified. Each jurisdiction requires independent analysis; the absence of a consistent cross-border framework means that legal and regulatory exposure is not yet fully priceable at the portfolio level.
The Operating Exposure for Global Heads of Data Center Energy
For a Global Head of Data Center Energy, the BTM shift creates a fundamental category change in project complexity. What has historically been structured as a real estate and PPA procurement decision is now an integrated energy infrastructure project — with generation asset financing, equipment procurement, fuel supply contracting, and regulatory licensing running in parallel with the data center build itself.
The practical consequence is longer lead times, not shorter. Gas turbine backlogs, regulatory licensing in non-standard BTM jurisdictions, and the construction alignment risk between the generation facility and the data center all extend the timeline to first power. Projects conceived as a speed-to-power alternative to the interconnection queue may encounter their own sequencing delays if equipment and permitting constraints are not addressed at the earliest project stage.
Budget forecasting is also materially affected. A BTM project in which the developer owns the generation asset requires capital allocation models spanning both data center construction and power plant development — with lenders likely demanding separate or coordinated security packages across both asset classes. Energy cost assumptions that held under a grid-connected model, including transmission and distribution cost offsets, are now subject to regulatory reassessment in multiple markets.
The stranded capacity risk already managed on the grid-connected side does not disappear in a BTM structure — it migrates. Instead of stranded data center capacity waiting on an interconnection, the exposure becomes a stranded generation asset if the data center underperforms, is delayed, or ceases to operate without an alternative power purchaser in place.
Signals the System Is Shifting
Several indicators will confirm whether BTM is becoming a durable infrastructure model or a transitional workaround. Watch FERC and PJM cost-allocation decisions as they move from order to implementation — the transmission charge treatment for co-located loads will set a precedent that other RTOs and ISOs are likely to follow or adapt. Watch gas turbine order backlog data; if lead times stabilize or compress, the equipment constraint eases and BTM project timelines become more predictable.
Regulatory movement in key European markets on BTM licensing will indicate whether cross-border portfolio BTM strategies are viable, or whether jurisdiction-by-jurisdiction regulatory negotiation remains the default. And watch whether any long-duration storage technology achieves commercial-scale deployment at a hyperscale facility — that milestone would materially change the reliability calculus for renewable-only BTM configurations and reduce dependence on gas-fired backup in markets where sustainability commitments are binding.
The fundamental question is not whether BTM becomes mainstream. Grid queue dynamics make some form of generation co-location increasingly necessary. The question is which configurations, in which jurisdictions, under which commercial structures, can actually be financed — and on what timeline.
Sources
- Aoshearman — Data centers and the rise of behind (Link)
