Should Data Centers Be Required to Generate Some of Their Own Power?

Photo: Aaron Hall, Wikimedia Commons, CC BY-SA 2.0

01 The question the grid is forcing

No legislature has passed a law requiring data centers to generate their own power. The grid is writing the mandate anyway. Dominion's data center customers in Virginia — the densest computing corridor on earth — average a seven-year wait for grid connection. Large-load interconnection in PJM's data center growth zones runs 36 to 48 months; generation projects that entered service in 2025 spent about eight years in queue on average. A data center shell takes 18 to 24 months to build. Buildings now outrun wires by half a decade.

Into that gap has rushed the market's answer. Global Energy Monitor counted 189 gigawatts of natural gas proposed for direct connection to U.S. data centers in the first half of 2026 — double the volume of six months earlier. The question for policy is no longer whether data centers self-generate. It is whether self-generation happens under rules that protect everyone else — or in the regulatory gaps it currently occupies.

Analysis — not prediction. N43 and Hermes AI grounds every scenario in the documented record and verified reporting as of September 19, 2026; where evidence is incomplete we say so.

02 The loophole problem: laws that stop at the meter

The case for requiring some self-supply is strongest where self-generation currently escapes the rules everyone else follows. In Virginia, gas turbines owned by data centers and serving only their own facilities are not beholden to the Virginia Clean Economy Act — the law obliges utilities to retire carbon-emitting sources by 2050, and a private generation island is not a utility. Data-center-owned generation also skips the State Corporation Commission's public-interest finding that utility projects must clear.

Air rules have seams too: gas units above 25 megawatts must participate in the Regional Greenhouse Gas Initiative's carbon allowance auction — but, as Virginia's conservation legislature noted, a 25-megawatt site can spread output across four or five turbines that individually duck the stationary-source threshold. Environmental groups are especially alarmed by proposals to lean harder on backup diesel during grid strain — the same emissions concern that shaped Virginia's HB 897 debate, where one senator summarized the goal as using as little energy as possible and “weaning themselves off of diesel generation.”

03 What the mandate version looks like in practice

A formal self-generation mandate — x percent of load from on-site resources — exists nowhere in U.S. state law as of September 2026. But the requirement is materializing in pieces. Ohio's approved AEP tariff forces large loads to pay for at least 85 percent of subscribed capacity for up to 12 years whether or not they draw it: if you will not build your own power, you will carry the cost of power built for you. Virginia's new GS-5 class imposes 14-year contracts with 85 percent minimum T&D payments. Both are mandates in economic form — self-generation becomes rational the moment grid service is priced at its true long-run cost.

The engineering stack is arriving to meet it. Behind-the-meter generation — on-site gas turbines, batteries, renewable microgrids — now ships as a product: modular offerings pair turbines with skidded electrical infrastructure so power and IT go up in parallel, cutting time-to-power by up to two years. Caterpillar announced an expanded push into data center power generation in March 2026, spanning gensets to 39-megawatt turbines. When the equipment is catalog merchandise, the mandate debate is really about who pays for the wires nobody built.

04 The federal referee: FERC's 50-megawatt line

The decisive rules so far came not from a mandate but from a tariff case. On December 18, 2025, FERC found PJM's tariff unjust and unreasonable: its co-location and behind-the-meter rules let large loads use on-site generation to “net” peak demand, offloading transmission costs onto everyone else — cost shifting contrary to the cost-causation principle. The June 18, 2026 order set the boundary: behind-the-meter netting survives, but capped — PJM's proposed 50-megawatt threshold was accepted (measured by netted load at a single electrical location, not nameplate), with a three-year transition and grandfathered existing contracts.

FERC paired the restriction with permission: new transmission service classes for co-located load, including a firm contract-demand service with an anti-toggling rule and interruptible options for loads that can limit withdrawals. The message is coherent — self-generate if you want, but the grid's costs do not disappear when you do, and backup units that only run during emergencies stay exempt. Five further show-cause orders pushed every other RTO toward the same framework, with the nationwide large-load interconnection rulemaking pending at the Commission.

05 The arguments against — and their weight

Three objections carry real weight. Emissions: a mandated fleet of gas turbines and longer-running diesel is worse for air quality than grid power from an increasingly clean mix — the strongest environmental argument is for mandating what counts as self-generation, not whether. Reliability: an islanded data center that stops paying for grid service still benefits from the grid's existence during its own outages and fuel-supply disruptions; FERC's answer is to price those backup services rather than forbid the island. Cost socialization, inverted: if a mandate exempts data centers from transmission charges, the remaining ratepayers hold more of the system — which is exactly the outcome the netting rules were reformed to prevent.

There is also a quieter failure mode: a mandate ratchets in the wrong direction could lock in combustion assets for two decades just as SMRs, fuel cells and long-duration storage reach maturity. The policy that ages best is the one that prices grid service honestly and lets each operator choose its own mix — which is, not coincidentally, what FERC actually did.

06 The verdict: price it, do not command it

The verified facts: seven-year average grid waits in Virginia; 189 GW of gas proposed for direct data center connection as of mid-2026; FERC's 50 MW netting threshold with a three-year transition; no state mandate anywhere. The analysis: a hard self-generation requirement would be blunter and slower than the tariff instruments already working. Long minimum-take contracts, honest transmission pricing and closed netting loopholes achieve the mandate's goal — data centers carry their own costs and increasingly their own power — without forbidding the grid purchases that still make sense.