The hottest question in the U.S. data center market is no longer whether AI needs more power.
We already know it does.
The question is who will pay to build the energy system required to supply it.
Why the debate has changed
For the past several years, the conversation centered on capacity. Could the grid physically supply enormous new loads fast enough? Interconnection queues, turbine lead times, and transformer shortages dominated the discussion.
That question has not disappeared. But it has been joined by a different one, and in a growing number of state capitals it has become the louder of the two.
Communities and regulators are now asking a more pointed question: when a utility builds new generation, transmission, and substations to serve a single enormous customer, who actually pays for it? The developer? The utility's shareholders? Or the residential and commercial customers who happen to share the same grid?
This is not an abstract policy debate. It is showing up in actual orders. As of mid-2026, roughly two dozen states have approved at least one large-load tariff built specifically to answer this question, with several more pending. Cost allocation has become one of the central factors determining which data center projects earn political, regulatory, and community support.
To be clear, this is not an argument that data centers are driving up everyone's electric bill. That framing is too simple, and it ignores the real economic benefits these projects bring to host communities. The more accurate statement is this: regulators are no longer willing to assume the answer. They want it in writing, and increasingly they are willing to say no to utilities that show up without it.
What "paying your own way" actually means, in practice
Ohio is the clearest example on the books. AEP Ohio spent 14 months in a contested proceeding after proposing a dedicated data center tariff in May 2024, following a moratorium the utility had placed on new large-load connections. Big tech pushed back hard: Amazon, Google, Meta, and Microsoft argued the terms were discriminatory. In July 2025, the Public Utilities Commission of Ohio sided with AEP anyway. New data centers above 25 MW now have to commit to paying for at least 85 percent of their contracted capacity every month, whether they use it or not, for contract terms running up to 12 years. Walk away early, and the exit fee runs three years' worth of those minimum payments.
Georgia has gone the opposite direction on generosity, and it is worth sitting with why. Georgia Power asked the Public Service Commission for approval to add nearly 10 GW of new generation, primarily gas, to serve a data center pipeline it says could reach the mid-30s of gigawatts. The commission's own staff testified that the forecast did not hold up: by their analysis, only about a third of the requested capacity was backed by signed contracts, and if the rest of the forecast doesn't materialize, ratepayers could be on the hook for billions annually in unneeded generation. The commission approved the full request anyway, in a unanimous vote, with a rate freeze through 2028 as the main protection if the numbers turn out to be wrong.
Virginia has landed on a structurally similar answer through a different door. In November 2025, the State Corporation Commission approved a new "GS-5" rate class for Dominion customers above 25 MW with a load factor over 75 percent, the profile that describes most data centers. Those customers now sign 14-year contracts and pay a minimum demand charge of 85 percent for transmission and distribution capacity and 60 percent for generation, whether they use it or not. Arizona has taken a lighter-touch route: Arizona Public Service now requires "load commitment agreements" that guarantee minimum usage before it will build, without Ohio's or Virginia's long contract terms.
Three states, three different answers to the same question. That range is the real story here, and it points to something the tariff mechanics alone do not capture.
The safeguards are only as good as the willingness to enforce them
Here is the part of this debate that gets glossed over: having a large-load tariff on the books does not, by itself, protect ratepayers. Georgia has one of the more detailed cost-allocation frameworks in the country, on paper.
Its commission still voted 5-0 to approve nearly the full 9,885 MW request over its own staff's recommendation to certify roughly a third of that amount as firm, with the remainder conditional on further signed contracts.
What happens to the risk from here is genuinely disputed, and it is worth stating plainly rather than picking a side. Georgia Power and the commission's order point to provisions requiring the utility to bear the cost if the demand does not materialize as forecast. Consumer and environmental advocates who opposed the vote argue those provisions are not strong enough, and that a meaningful share of the exposure will still land on residential and commercial ratepayers if the data center pipeline shrinks. Both things can be roughly true at once: the order has ratepayer-protection language in it, and critics can reasonably doubt whether that language holds up if a multi-billion-dollar forecast miss actually happens in 2028 or 2029.
That is not a criticism of any single vote. It is a structural point. Regulators are elected or appointed officials weighing real jobs, real tax revenue, and real political pressure against a forecast that may or may not hold up years from now. When the numbers are uncertain and the upside is large, the incentive to approve now and reconcile later is strong, tariff language notwithstanding. AEP Ohio's fight took 14 months and produced hard numbers specifically because Ohio Consumers' Counsel and industrial ratepayer groups had standing and pushed. Georgia's own staff raised a comparable objection and still lost 5-0.
The lesson for developers is not "avoid Georgia" or "prefer Ohio." It is that the presence of a tariff tells you less than the track record of how that state's regulators actually rule when a forecast is contested. That history is now diligence-able, and it should be part of site selection the same way tax abatements and water rights already are.
Why developers need an energy strategy before announcing a campus
The practical consequence of all this is timing. A developer who shows up with an announced load and no answer to the cost allocation question is now at a disadvantage, even where the announced capacity is impressive. Georgia Power's own filings show that since mid-2024, 19 large-load projects that had expressed interest in the state were later canceled or withdrawn, accounting for the large majority of the megawatts that have dropped out of its queue. Utilities and regulators have started pricing in the expectation that a meaningful share of announced demand will not show up as planned, and they are writing tariffs that assume it.
A developer that arrives with a defined mix of grid power and onsite generation, a realistic ramp schedule, and financial commitments that match the size of the ask is negotiating from a different position than one asking to be served entirely from the existing system on day one, on the utility's forecast, on the utility's risk. Onsite generation stops being a hedge here and starts being a negotiating asset, particularly for load that needs to come online before new grid infrastructure is complete.
What this means for site selection and project execution
Site selection criteria are shifting accordingly. Available land and cheap power are no longer sufficient. The questions now include: What is this state's large-load tariff structure? What minimum-demand or credit terms will be required? Is there a realistic self-generation path if grid timelines slip? How has the local utility handled cost allocation disputes with other large customers?
States and utilities that provide clear, published rules on these questions are becoming more attractive, not less, to serious developers. Ambiguity is a cost. A predictable, if strict, framework is easier to underwrite than an open question that could be decided differently project by project.
The winning sites will answer a simpler question
The next generation of data center projects will not earn support simply by proving that demand exists.
They will need to demonstrate that their power plan is deliverable, that their financial commitments are credible, and that the communities surrounding them are not being asked to carry an unfair share of the risk.
In the next phase of this market, the winning sites may not be the ones with the largest announced power capacity.
They may be the ones with the clearest answer to a much simpler question.
Who pays?⚡
The Strategic Imperative
The organizations that ask better questions early are the ones that actually get built.
Right now, there are projects with signed LOIs, committed capital, and serious teams behind them that will not get built. Not because the economics are wrong. Not because the demand isn't there. Because nobody coordinated the energy side early enough.
Most organizations still treat power, natural gas, and energy infrastructure as separate decisions. That separation is where hidden cost and long-term constraint take hold.
Interconnection timelines, not land readiness, now determine feasibility. The median time to commercial operation is approaching five years, with some markets stretching to seven or more. More than 35 GW of data center power is projected to be self-generated by 2030, not because developers prefer it, but because the grid cannot deliver on the timelines that projects require.
A decade ago, developers optimized for the lowest delivered cost. In 2026, they are optimizing for earliest energization. That is the difference between projects that look viable and those that actually get built.
The organizations solving this are not approaching energy as a procurement cycle. They are approaching it as a coordinated system across power markets, natural gas strategy, utility and infrastructure pathways, and real-time operational performance. That level of integration does not happen by accident.
Legend Energy Advisors has already brought 2.5+ GW to market and is currently advising on an additional 6.5+ GW in the data center space alone. A repeatable system built across every layer of energy strategy, working together at scale. Not across one market. Not on one project type. Across the full complexity of what it actually takes to get a project operational.
If your project is next, the conversation starts here.
Don't Just Use Better Energy. Use Energy Better®
Originally published in Energy Ninja Chronicles (LinkedIn newsletter).