Vermont's Data Center Bill Is Dead. Developers Weren't Lining Up to Build One Here Anyway
ANALYSIS: By the count of the industry tracker Silicon Report, Vermont ranks 50th of 50 states for data center development.
Governor Phil Scott’s veto of H.727 held. On Friday the House fell short of the two-thirds needed to override it and sustained the veto, 83–52, ending the data center bill’s path this session. It had passed both chambers — the House without a recorded roll call, the Senate 26–3 — and drawn a late override push from the advocacy groups that backed it, including the Vermont Natural Resources Council.
Beneath that fight sits a question that drew less attention during the year’s debate: how much demand is there to build a large-scale data center in Vermont in the first place?
The public record reviewed for this story points to very little. There are no active proposals before the state, the one recent attempt collapsed on cost, and Vermont sits at or near the bottom of the commercial data-center and AI-adoption measures reviewed for this story.
By the count of the industry tracker Silicon Report, Vermont ranks 50th of 50 states for data center development — one small operational facility, none under construction, and none announced — one of a handful of states, alongside New Hampshire, Rhode Island, Hawaii and the District of Columbia, with nothing in the pipeline at all. That is a single commercial tally, not an official ranking, but it tracks with what state officials and utilities describe below. The earlier shorthand — that Vermont is “not in the top 30” of states developers are considering — if anything describes a smaller gap than the tally shows.
The regulatory effort
The Legislature took up two measures this session. H.727, sponsored by Rep. Laura Sibilia, I-Dover, would apply enhanced siting, environmental, and ratepayer-protection requirements to facilities drawing 20 megawatts or more. Sibilia has described the target as “factory-sized developments that can strain local electric grids, raise customers’ electric bills, and cause environmental and noise pollution.”
A second bill, S.205 from Sen. Becca White, D-Windsor, would impose a moratorium on AI data centers — defined as facilities above 100 megawatts, a hyperscale threshold no existing Vermont facility approaches — through July 1, 2030, pending a Public Utility Commission study.
The impulse reaches the local level too. In March, Royalton voters approved a five-year municipal moratorium on AI and cryptocurrency data centers — in a town that, as the measure’s own backers acknowledged, has no such facilities and none proposed.
The demand record
What the legislative record does not show is a queue of developers seeking to build here.
Kerrick Johnson, Commissioner of the Department of Public Service, told Vermont Public there are no active large-scale data center proposals before the state. Green Mountain Power, the state’s largest utility, reported “maybe one or two very speculative general inquiries over the years” and no serious proposals, spokesperson Kristin Kelly said.
The one recent large-scale attempt identified in published reporting collapsed on cost. St. Albans pursued a data center proposal last year — a project described at various points as 8 to 12 megawatts or as large as 50, which would have created roughly 1,000 temporary construction jobs and about 20 permanent positions, former Town Manager Sean Adkins told VTDigger. It failed when connecting to the grid was estimated to require about $30 million in infrastructure upgrades.
“It all came down to power,” Adkins said. “It was going to cost just too much money to build a substation that would have allowed the project to move forward.”
Sam Chevalier, an assistant professor of electrical and biomedical engineering at the University of Vermont who studies power systems, told Vermont Public the case reflects a broader pattern: “The infrastructure limits and the price of electricity together just make it kind of a non-starter.”
Vermont’s total peak electricity demand hovers around 900 megawatts, per VELCO figures cited in earlier reporting — a figure that itself constrains how large a single new load the existing system could absorb.
Data-center siting is typically shaped by access to cheap power, grid capacity, and fast permitting — areas where Vermont’s high regional electricity rates and review processes can be constraints, even though, as VELCO general counsel Mark Sciarrotta told lawmakers, electricity rates may not weigh heavily for operators making large profits on these facilities.
Weighing the advocates’ claims
VNRC’s appeal to override the veto rests on four specific harms: higher electricity bills, PFAS water pollution, heavy water use for cooling, and carbon emissions. Each is documented somewhere in the national record. How squarely each applies to a Vermont facility is a separate question, and the answers differ.
Electricity demand and bills. The claim that a large data center “could consume more electricity than the entire state of Vermont” is directionally accurate for the largest AI campuses being built nationally, some of which exceed 1,000 megawatts — more than Vermont’s roughly 900-megawatt peak.
But nothing of that scale has been proposed here, and the St. Albans case suggests the grid could not host it without prohibitive upgrades. Green Mountain Power, meanwhile, told lawmakers a properly sized facility — engineer Cam Twarog cited a 50-to-200-megawatt range — “can be kind of a net benefit for costs on the electric system,” and said it had tested several Vermont sites, including locations near a VELCO substation in New Haven, a Williston site near the GlobalFoundries campus, a former quarry in West Rutland, and the former Vermont Yankee site in Vernon. Whether a data center raises or lowers ratepayer costs depends on the contract terms — which is the gap H.727 is written to address.
Water use. Of the four, this is among the most concretely documented. Large evaporative-cooled facilities can consume 1 to 5 million gallons a day, and U.S. data centers directly used an estimated 17.4 billion gallons in 2023, according to Lawrence Berkeley National Laboratory research cited by EESI. But the figure is a function of cooling design, not an inherent property: closed-loop and immersion systems use little or no water. The Maine facility profiled in earlier Compass reporting, which submerges its servers in oil rather than using water cooling, consumes none.
PFAS “forever chemicals.” Of the four, this is the most attenuated as applied to a facility Vermont might realistically see. PFAS enters data centers mainly through two-phase immersion cooling fluids, along with semiconductor components and some fire-suppression systems.
That cooling method is in retreat: after 3M announced its exit from PFAS manufacturing, its Novec and Fluorinert fluids were discontinued by the end of 2025 — a withdrawal that DataCenterDynamics reported has collapsed the two-phase supply chain. Single-phase immersion, the alternative, uses non-PFAS oil-based fluids. The water-contamination concern is documented and pressed by environmental groups such as the Sierra Club and EESI, though it has not been the subject of a specific regulatory finding for data centers, and PFAS’s persistence means fluids already in use remain an environmental concern wherever they were deployed.
As a general claim about how data centers operate, however, it describes a shrinking slice of the industry — and one with no presence in Vermont.
Carbon emissions. Here the local context cuts against the claim more than for it. Vermont’s electricity is among the cleanest in the country — as of 2020 it posted the nation’s lowest power-sector carbon intensity, with almost all of its in-state generation coming from renewables — so a data center drawing on Vermont supply would emit far less than one on a fossil-heavy grid.
The wrinkle is regional: Vermont sits inside ISO-New England, where natural gas is often the marginal generator, so a large new load can pull more gas onto the regional system even when a facility’s own contract is nominally clean. The carbon concern is real, but it turns on sourcing and scale rather than on the data center as such.
The pattern across the claims assessed here is the same: the harms are real where they occur, but each is design- and scale-dependent. And on the demand question itself, the officials and engineers closest to it — the Department of Public Service, Green Mountain Power, and UVM’s Chevalier — describe a state that is not now drawing the large facilities those harms depend on.
The competing reading
Supporters of the bill can answer the demand question without conceding it. The absence of current proposals, they argue, is the window to set rules — not the reason to skip them. A moratorium-and-study approach like S.205, and the proactive parameters in H.727, are built on the premise that it is easier to regulate before a wave arrives than after. On that reading, “no proposals today” is the case for acting, not against it.
The adoption data points the other direction on a different metric. A Microsoft AI Economy Institute report released in May, which measures how many working-age residents use AI tools — not where companies site facilities — ranked Vermont 46th of 50 states, at 23.3 percent in its state-level table. That gauges consumer adoption rather than developer demand, but it places Vermont near the bottom of both measures the technology has generated this year.
What the failed override leaves unresolved
Scott’s veto letter argues that existing law — Act 250, Public Utility Commission oversight, environmental permitting, energy siting rules, and municipal zoning — already supplies “substantial regulatory authority to prevent harmful impacts,” and that new requirements risk “driving current or future jobs and investment to other states.” Supporters counter that those frameworks were not built for facilities of this scale and leave specific gaps on ratepayer protection and water use.
A federal wildcard hangs over the entire question. A December 2025 executive order directs the Justice Department to challenge state AI laws in court. Its instructions for a future federal framework expressly exclude state rules on “AI compute and data center infrastructure, other than generally applicable permitting reforms” — language that suggests H.727’s siting and ratepayer provisions are meant to stay with the states, though the order’s litigation arm is not explicitly bound by that carve-out. A measure built around “AI data centers,” such as S.205, could prove more exposed.
The failed override settles H.727’s fate for this session; barring a return to the issue, the bill is dead. It does not settle the question underneath it: whether the development the bill sought to govern is one Vermont’s grid, geography, and price structure are likely to attract at all.
On that, the record so far is thin — near the bottom of the states, by the available count — though present demand is a weak predictor of where a national AI buildout could push next. That tension, as much as the statutory argument, is what the session’s fight leaves behind.