Off-Grid Data Centers: Quick Fix or Grid Cost Bomb?

That is not a fringe workaround; it is becoming a default planning posture in markets where interconnection timelines outrun AI build schedules

Decision Lens

The off-grid debate has moved from theoretical to structural. Roughly 30% of planned U.S. data center capacity is now paired with behind-the-meter power, yet the sector’s most visible operators — Amazon, Google, Schneider Electric — are publicly arguing this path is unsustainable. The core contradiction: interconnection queues are forcing off-grid workarounds at scale, while each defection compounds the grid affordability problem for everyone who remains connected, including the operators most exposed to long-term utility rate risk.

90-Second Brief

As the week closes, at CERAWeek 2026, technology and energy executives debated whether data centers should keep connecting to the grid or build self-sufficient generation. Approximately 30% of planned U.S. Data center capacity is already paired with behind-the-meter power, a share driven by multi-year interconnection delays. Google, Amazon, and Schneider Electric pushed back publicly, arguing that off-grid development worsens grid affordability and is not a durable answer to the interconnection bottleneck.

What’s Actually Happening

The mechanism is straightforward but underappreciated: interconnection queues running three to seven or more years have created a structural incentive to bypass utility hookups entirely. Developers are deploying mobile gas generators and repurposed aircraft turbines to bring capacity online faster. Fermi America is taking the logic furthest — its planned Texas complex will use on-site gas, nuclear, and solar with no transmission lines to the grid at all, according to its CEO at CERAWeek.

The volume effect is significant. Cleanview data cited at the conference puts nearly a third of planned U.S. capacity in behind-the-meter configurations. That is not a fringe workaround; it is becoming a default planning posture in markets where interconnection timelines outrun AI build schedules. The aging transmission and distribution infrastructure — already identified by Lawrence Berkeley National Laboratory as a primary driver of rising electricity prices — does not get upgraded by operators who have disconnected from it. Each new islanded campus removes a potential cost-share contributor from the grid modernization equation.

Why It Matters for Global Heads of Data Center Energy?

For energy leads managing multi-region portfolios, the off-grid trend creates simultaneous pressure on two fronts. First, the operational temptation is real: behind-the-meter power eliminates interconnection queue exposure and gives the team direct control over generation. Second, the systemic risk runs in the opposite direction: as more large consumers defect, remaining grid-connected loads — including your own facilities in other markets — absorb a proportionally higher share of transmission and distribution costs through rising utility rates.

Google’s response is instructive. Rather than exiting the grid, the company announced a 1 GW demand response integration into utility contracts across the South and Midwest, positioning its facilities to curtail during peak periods. That is not altruism; it is a strategy to preserve utility relationships, manage grid costs, and build regulatory goodwill at a moment when FERC is actively considering policy tools to keep data centers grid-connected. For energy heads with multi-jurisdictional interconnection strategies, the choice between Google’s demand-flexibility path and more islanded capacity will carry long-duration PPA, tariff, and regulatory consequences that are difficult to reverse.

The Forward View

FERC Commissioner Judy Chang’s public acknowledgment that price signals alone are insufficient to drive demand-side investment is a signal that regulatory intervention is being considered, not merely discussed. The form that intervention takes — mandatory demand response requirements, new tariff structures, or direct co-location rules — will materially affect site-selection economics for data centers that are currently grid-connected.

EPRI’s active piloting of demand response and flexibility tools for data centers suggests the technical framework for mandatory participation is being developed in parallel with the policy debate. Operators who have already embedded demand response capacity — as Google has — will be better positioned to comply, negotiate, or influence whatever regulatory structure emerges. Operators who have moved off-grid entirely may find themselves outside the conversation, and potentially subject to re-interconnection requirements or stranded-asset risk if grid-defection rules tighten.

What We’re Uncertain About?

Whether FERC will mandate demand response for grid-connected data centers. Commissioner Chang’s comments suggest active deliberation, but no rulemaking timeline has been confirmed. Resolution would come from a formal FERC Notice of Proposed Rulemaking or advanced rulemaking proceeding.
The long-term economics of fully islanded campuses. Fermi America’s Texas model bundles gas, nuclear, and solar on-site, but the capital cost, fuel price exposure, and permitting complexity of that approach at scale are not publicly quantified. Independent cost benchmarking would be needed to assess whether it outperforms grid-connected alternatives over a 15–20-year horizon.
How utilities will respond to load loss from behind-the-meter defections. If a significant fraction of projected load growth does not materialize on the grid, utility rate cases and capacity planning assumptions will shift — potentially accelerating cost recovery pressure on remaining connected customers. The magnitude and timing of that feedback loop are not yet established.
The regulatory treatment of hybrid configurations. Many operators will neither go fully off-grid nor remain purely grid-connected. Whether FERC or state PUCs develop clear rules for hybrid behind-the-meter/grid-parallel configurations — and how those rules affect interconnection rights — remains an open question.

One Question to Bring to Your Team

For each market where we are currently evaluating behind-the-meter capacity to work around interconnection delays, have we modeled the 10-year utility rate trajectory for our grid-connected facilities in adjacent regions — and does that trajectory change our calculus on where we accept stranded interconnection queue positions versus where we exit?