Everything About AI
NextFin News - The congressman whose district contains the fourth-largest concentration of data centers in the world says the United States can no longer let hyperscale AI buildout happen by accident. On Sept. 8, Representative Suhas Subramanyam, a Virginia Democrat whose district spans part of Northern Virginia's data-center corridor, introduced a four-bill National Data Center Plan that would impose federal siting standards, force data centers to pay the full cost of the grid upgrades they trigger, standardize measurement of their energy and water use, and treat data infrastructure as a critical national-security asset. The stakes are concrete: Virginia's commercial electricity sales rose by roughly 30 million megawatt-hours between 2019 and 2025, a jump federal energy data attribute largely to data centers, and the question of who pays for the buildout is moving from local zoning hearings to the House floor.
The package lands at the moment when AI's power appetite has become the binding constraint on the industry itself. Global data-center electricity consumption is projected to reach 565 terawatt-hours in 2026, up 26 percent from 447 TWh in 2025, according to Gartner, with power demand rising to 133 gigawatts from 105 GW. Subramanyam's answer is that the rules have not kept pace with the machines.
"Our region is home to the fourth largest concentration of data centers in the world, and many more are on the way," Subramanyam said in a statement. "Because of this, our utility bills are up and transmission lines are cutting through our schools and backyards. People are concerned about the future of our community and angry about the impact this will have on noise, the environment, and their property. The approach to building data centers has been driven by short-term profit rather than long term planning."
The Four Bills, and What Each One Actually Does
The plan is not a single grand gesture but four separate bills, each aimed at a different failure point in the current ad hoc system.
The Responsible Data Center Siting Act would direct the Secretary of Energy to establish best practices for where data centers go, weighing energy availability, impacts on utility bills, water and electricity use, and the environment, along with community factors such as noise, national security, workforce, and tax revenue. It carries six Democratic co-sponsors.
The Data Center Fair Share Act is the one with teeth. It would require states to adopt large-load tariffs ensuring that data centers and other large energy users cover the full cost of the generation and transmission upgrades needed to serve them. The enforcement lever is blunt: states that fail to adopt the tariffs lose access to federal highway funds. In Northern Virginia, residents have been stuck with rising utility bills to support the enormous energy demands of hundreds of data centers, the bill's backers argue.
The Data Infrastructure Energy Measurement and Standards Act (H.R. 9372) directs the National Institute of Standards and Technology to develop common methods for measuring data-center energy and water use, including the power drawn by AI model training and operation. This bill has already cleared the House Science, Space and Technology Committee on a 34-1 vote and is headed to the House floor.
The Data Infrastructure Risk Reduction Act (H.R. 8711) would require federal officials to identify data centers that should be treated as critical infrastructure and develop a strategy addressing threats to the facilities, nearby communities, and the power and water systems supporting them.
The political momentum behind the cost-allocation argument runs deeper than one congressman's press release. Senator Jon Husted, an Ohio Republican, introduced the Ratepayer Protection Act on July 16, 2026, and the House passed it on Sept. 16 by a vote of 417-3. Husted brought it to the Senate floor the next day seeking unanimous consent, arguing that America is expected to build the equivalent of 1,000 major data centers over the next five years, with more than 2,000 projects currently proposed and tracked. A Senate Democrat objected, blocking the bill from passage — a procedural defeat that signals the fight has only just begun. The bipartisan House margin matters: this is not a red-versus-blue fight over whether data centers should exist, but a widening consensus that the current cost allocation is unsustainable, even as the two parties disagree on how hard Washington should reach.
Why Virginia Is the Canary in the Coal Mine
Northern Virginia is not an abstract case study. It is the world's data-center capital, and the stress is already visible in utility filings.
Dominion Energy Virginia, the commonwealth's largest electric utility, expects to invest nearly $7.6 billion in transmission projects through 2031, with roughly 68 percent of those upgrades tied to serving data centers. In its rate case, the Virginia State Corporation Commission approved a revenue increase of $565.7 million for 2026 and $209.9 million for 2027 — significantly less than the $822 million and $346 million Dominion had requested — while creating a new rate class for large-scale users that takes effect in 2027. Under that class, certain large users must pay at least 85 percent of their distribution and transmission demand charges and 60 percent of generation demand.
The local fiscal picture cuts both ways. Loudoun County says data centers generated $1.2 billion in real and personal property taxes in fiscal 2026, accounting for 39 percent of the county's overall budget. That is the bargain communities have accepted: tax revenue in exchange for power lines, substations, and rising bills. Subramanyam's plan is an attempt to rewrite the terms of that bargain at the federal level.
"Since its founding, in 2014, the Coalition to Protect Prince William County has been warning residents, elected leaders, and the SCC of the unprecedented power demands of the data center industry. Prince William County residents are already paying a high cost with our rising utility rates and energy infrastructure expansion threatening our homes, our schools, our parks, our private property. Congressman Subramanyam's National Data Center Plan is a needed step in the right direction to protect residents, our environment, and our grid from irresponsible development. Without action, the risk to our grid is no longer a question of IF we will suffer a catastrophic failure, but when."
That quote came from Elena Schlossberg, executive director of the Coalition to Protect Prince William County, a local group that has opposed data-center expansion for more than a decade.
The Mechanism: Who Pays Is the Whole Fight
The surface story is that data centers consume a lot of electricity. The mechanism that matters is how that consumption is translated into bills for everyone else.
Under the traditional utility regulatory compact, a utility earns a regulated return on capital it invests. When a hyperscale campus needs a new substation and miles of transmission, the utility builds it, adds it to its rate base, and spreads the cost across all customers. Residential and small-business ratepayers then subsidize infrastructure built for a single tenant. The Fair Share Act attacks exactly that flow: make the large-load customer pay the incremental cost, and the subsidy disappears.
There is a second channel that is less visible but more corrosive. When grid capacity is scarce, utilities must procure power years in advance, often signing long-term contracts or building new generation on the expectation of future load. If a data-center project is delayed, downsized, or abandoned after the utility has already committed capital, the stranded cost lands on ratepayers. Husted's Ratepayer Protection Act would require state regulators to consider standards ensuring large-load customers pay the full incremental cost, and would allow financial assurances so families are not left paying for stranded infrastructure. That is procurement discipline, not anti-technology policy — and it is the distinction that determines whether the legislation accelerates or strangles the buildout.
The measurement bill targets a different failure: nobody actually knows how much power and water AI is using. Without standardized metrics, forecasts are guesses, and siting decisions are made blind. NIST standards would create the technical foundation for better electricity forecasting and resource management. You cannot price what you cannot measure.
The empirical case for intervention is real but contested. A Federal Reserve Bank of Dallas working paper published in March 2026 found that existing data centers have already pushed average wholesale electricity prices up by 2 to 6 percent nationwide, with substantially larger effects in regions hosting major data-center corridors, and projected that if proposed construction proceeds under high-utilization scenarios wholesale prices could rise by as much as 50 percent by 2028. But the Dallas Fed study measures wholesale prices, not retail bills, and it does not isolate how much of the increase stems from data-center demand versus broader supply constraints. An Electric Power Research Institute study released the same year reached the opposite conclusion on retail rates, finding that a doubling of data-center capacity actually caused residential electricity prices to fall by about 3.5 percent for a fixed level of demand, as the added load spread fixed grid costs across more kilowatt-hours. Both findings can be true at once: wholesale prices rise at the margin where capacity is tight, while retail bills in some jurisdictions fall because large users absorb fixed costs. Which effect dominates is a local question — and that is precisely why a single federal rule fits awkwardly.
Cyclical Squeeze or Structural Regime Change?
Is the data-center power crunch a cyclical supply-demand imbalance that will ease as capacity catches up, or a structural regime change that permanently rewrites how infrastructure is financed? The answer is both, and the distinction determines which parts of Subramanyam's plan survive.
The cyclical leg is real and mean-reverting. Gartner's own forecast explicitly builds in parts shortages, delayed or canceled projects, and geopolitical disruption — all factors that can slow demand growth. If AI inference becomes cheaper and more efficient, or if model developers shift workloads to off-peak hours and cooler climates, the peak-load pressure eases. History offers a parallel: the cloud migration wave of the early 2010s produced similar forecasts of runaway data-center power growth, and efficiency gains absorbed much of it. A cyclical reading says today's grid congestion is a construction lag, and that by 2030 new generation and transmission will have caught up with load.
But the structural leg is stronger, and it will not self-correct. Three pieces of evidence point to a regime shift rather than a cycle. First, the scale is different: Deloitte projects U.S. power demand from AI data centers alone could surge from 4 GW in 2024 to 123 GW by 2035, a thirty-fold increase in a single country, with AI's share of total data-center power demand rising from 12 percent to 70 percent. Second, the load profile is different — AI clusters demand firm, always-on power in specific geographies, not the diversified, flexible load the grid was designed for. Third, and most important, the regulatory response is itself structural: once states and Congress create large-load tariff classes and federal siting standards, those rules do not unwind when the cycle turns. The International Energy Agency's base case has global data-center electricity consumption doubling to around 945 TWh by 2030, just under 3 percent of total global consumption — more than four times faster than growth in all other sectors combined.
Subramanyam's plan is a bet that the structural leg dominates. If he is wrong and the crunch is cyclical, the highway-fund enforcement lever is a sledgehammer aimed at a speed bump, and it could push data-center investment into jurisdictions with weaker environmental and labor standards rather than reducing costs for ratepayers.
The Second-Order Question the Market Is Not Asking
The first-order effect of the plan is obvious: data centers pay more, ratepayers pay less. The second-order effect is where the real risk lies — and it cuts against the bill's supporters.
If large-load tariffs and federal siting standards raise the cost and slow the permitting of U.S. data centers, the marginal project does not disappear. It moves. Texas is projected to exceed 40 GW of data-center capacity by 2028, nearly 30 percent of total U.S. demand, in part because its grid and regulatory environment are more permissive. A federal framework that tightens Virginia while leaving Texas untouched simply reallocates buildout within the United States. A framework that tightens the United States as a whole reallocates it offshore — and that collides directly with the national-security rationale in Subramanyam's own fourth bill.
Here is the contradiction the legislation has not resolved: the Risk Reduction Act treats data infrastructure as critical national-security assets that must be protected from adversaries, while the Fair Share Act and Siting Act make it harder and more expensive to build that infrastructure at home. If the net effect is to slow U.S. AI capacity relative to China's, the plan protects American ratepayers' bills while ceding American compute leadership. That is the trade the bill asks voters to accept, and it is not stated plainly in the press release.
The market has not fully priced this tension. Data-center REITs and their power suppliers have rallied on the assumption that buildout continues largely uninterrupted. Equinix reported second-quarter 2026 revenue of $2.625 billion, up 16.4 percent year over year, and Digital Realty posted first-quarter revenue of $1.64 billion, up 16.2 percent. Those numbers assume permitting pipelines stay open. A federal siting regime that adds review layers is a headwind to the very growth thesis underpinning those valuations — even if the policy is economically sensible.
The Strongest Case Against the Bill
The most serious objection is not that the bill is wrong on the economics — it is that it is wrong on the venue and the enforcement mechanism.
Utility regulators, not Congress, are the appropriate venue for rate design. The Virginia SCC has already moved in exactly the direction Subramanyam wants: it created a new large-user rate class, imposed minimum charges on data centers, and approved a smaller rate increase than Dominion requested. The Southern Environmental Law Center, which is hardly a tech-industry lobby, called the order a step toward ensuring data centers pay their fair share. If state commissions are already correcting the cost allocation, a federal highway-fund mandate is redundant at best and coercive at worst.
The counter-argument from the bill's supporters is that state action is patchy and slow, and that ratepayers in states without aggressive regulators are still subsidizing hyperscale buildout. That is a fair point — but it does not answer why the enforcement lever is federal highway money, a funding stream with nothing to do with energy policy. Conditioning transportation funds on utility-rate design sets a precedent that could be turned against any number of state policies, and it invites litigation that could delay the very protections the bill promises.
Karim D. Marshall, director of climate and energy policy at the Consumer Federation of America, framed the stakes differently: "Tech billionaires say that they will responsibly share our power and water. They say that they will be responsible without data and information. They say 'trust us, we know better.' The moment demands that consumers are protected from the abuses of the rich and powerful." The measurement bill directly answers the "trust us" problem. The highway-fund lever answers a different question — how much federal coercion is acceptable to make states comply.
What to Watch Next
The immediate catalyst is the House floor schedule for H.R. 9372, the measurement bill, which has the clearest path because it passed committee 34-1 and does not redistribute costs. The Fair Share Act and Siting Act face a harder road: they would have to clear the House Energy and Commerce and Transportation and Infrastructure committees, and the highway-fund penalty will draw opposition from states' rights conservatives as well as from tech and utility lobbies.
The falsifying signal for the structural-crunch thesis is specific and observable: if U.S. data-center power demand fails to grow toward the Deloitte trajectory — if it stalls well below the path from 4 GW in 2024 toward triple digits of gigawatts by the early 2030s — then the crunch is cyclical, the regulatory response is overbuilt, and the bill's coercive elements become net drag. Conversely, if Dominion's $7.6 billion transmission buildout through 2031 proves insufficient and Virginia's large-user rate class becomes the national template, the structural reading is confirmed and Subramanyam's framework moves from proposal to blueprint.
Short term, the market will watch for any sign that permitting timelines lengthen in Northern Virginia — that is the first place federal siting standards would bite. Medium term, the question is whether other states copy Virginia's large-load tariff model voluntarily, which would undercut the case for federal coercion. Long term, the real test is whether the United States can add generation and transmission fast enough to keep AI buildout domestic while protecting ratepayers — because a policy that succeeds on the second goal but fails on the first protects bills and loses the race.
The central judgment: Subramanyam is right that the ad hoc era is over, but the bill's success depends on whether Washington can design cost allocation without strangling the buildout it is trying to govern. The ratepayer-protection logic is sound; the highway-fund hammer is not. The market should watch Virginia's permitting pipeline, not the press release, for the first real signal of which way this lands.
Explore more exclusive insights at nextfin.ai.
