One proposed data center in Genesee County, New York, would consume 500 megawatts of power and receive approximately $1.4 billion in tax breaks. A single facility. The jobs it would create in return are described by residents and some lawmakers as insufficient to justify that exchange. That political pressure is real and growing. But the more consequential question being worked in Albany right now has nothing to do with tax incentives.
New York state regulators are weighing rules that would require large energy users to fund a greater share of the grid infrastructure their power demands require. If those rules take effect, the economics of every data center in the New York development pipeline change materially. The developers who modeled their projects on current cost-allocation assumptions built in a subsidy that may not survive the regulatory cycle. That is the story.
Five hundred megawatts from a single facility is not unusual by 2026 standards. The Indiana buildout profile showed the same scale in agricultural land being converted to compute infrastructure. What makes the Genesee County project a political flashpoint is the combination: 500 MW of load, $1.4 billion in tax relief, and a job creation profile that residents and lawmakers argue does not come close to justifying the strain on the grid, water systems, and local infrastructure.
The broader scale of the New York pipeline amplifies the political pressure. The collective electricity consumption of proposed AI data centers across New York could approach the consumption of New York City itself. That comparison lands differently than an abstract megawatt figure. New York City draws roughly 50 terawatt-hours per year. A data center pipeline that approaches that number is not a fringe concern for grid planners. It is a load growth scenario that forces hard decisions about where the power comes from and who pays to deliver it.
Governor Kathy Hochul has opposed a statewide moratorium on data center development. That position reflects the economic development argument: data center investment brings construction jobs, tax revenue, and long-term presence from companies that make supply chain decisions based partly on regulatory predictability. A moratorium ends those conversations. Hochul is not wrong about the economic logic. The question is whether the current regulatory framework distributes the costs of that investment fairly, and the answer from grid operators and community advocates is that it does not.
Under current New York utility regulation, when a large new load comes onto the grid, the cost of upgrading the substation, transmission lines, and distribution infrastructure to serve it gets socialized across the existing ratepayer base. Everyone's electricity bill absorbs a small portion of what the new customer required. This is standard utility practice across most of the United States, rooted in the regulatory compact that treats the grid as a shared public resource.
That compact was written when the largest new loads were factories, hospitals, and commercial buildings adding a few megawatts. A data center adding 500 MW is a different category of customer. The substation upgrades, new transmission capacity, and distribution hardening required to serve a facility at that scale can run into hundreds of millions of dollars. When that cost gets distributed across residential and commercial ratepayers who receive no direct benefit from the facility, the political sustainability of the arrangement collapses.
Virginia's experience with electricity cost increases tied to data center load growth shows exactly how quickly that political sustainability erodes. When ratepayers see their bills increase to fund grid upgrades for facilities that employ hundreds rather than thousands of people, the opposition consolidates. New York is watching Virginia's trajectory and the regulatory response it has triggered.
The rules under consideration in Albany would shift some or all of those grid upgrade costs directly to the large energy users who require them. A developer planning a 500 MW facility in Genesee County would need to fund the substation expansion, the transmission upgrades, and the distribution capacity that its load demands, rather than having that cost spread across all ratepayers. The developer still benefits from using the grid. The difference is that the infrastructure investment required to accommodate that load is priced into the project rather than externalized onto the public.
Data center developers underwrite projects with detailed assumptions about electricity costs, grid connection costs, and the timeline from permitting to commercial operation. The electricity cost projection is the largest single operating variable. In states like Virginia and Texas, electricity cost increases tied to grid congestion and infrastructure investment are already moving projects off pro formas that looked viable two years ago.
Direct cost allocation for grid upgrades adds a different line item. The capital cost of a substation expansion required to serve a 500 MW facility in rural New York can run $150 to $400 million depending on the distance from existing high-voltage transmission, the required voltage step-down configuration, and the capacity additions needed on the transmission side. Under current rules, most of that cost lands on the utility and gets recovered through rates. Under a direct cost allocation rule, it lands on the developer's capital budget.
That changes site selection calculus immediately. Locations that appear cheap on a lease-rate basis become more expensive once the developer has to price in the full grid connection cost. Locations with existing high-voltage infrastructure and available substation capacity become more attractive even at higher land and lease costs. The sites in Genesee County, chosen partly for inexpensive land and access to New York Power Authority hydropower allocations, become less competitive if the developer has to fund the transmission capacity to get that power to the facility reliably at 500 MW.
Residents and lawmakers citing insufficient permanent jobs are raising a real concern, but the permanent job count is a proxy for the underlying economic exchange. A data center that creates 50 to 200 permanent jobs while drawing 500 MW of power, consuming water at scale, and absorbing $1.4 billion in tax incentives has transferred a substantial share of the economic surplus it generates to shareholders and corporate tax jurisdictions, not to the community hosting it. The job number is how that transfer gets visible to people who do not follow utility rate cases.
The grid upgrade cost allocation question gets at the same transfer from a different angle. When grid infrastructure costs are socialized, the hosting community subsidizes the facility's operating economics through utility bills, in addition to the tax incentives that reduced its capital cost. The two subsidies compound. A developer who receives $1.4 billion in tax breaks and also avoids funding the grid upgrades its load requires has externalized a substantial portion of its true operating cost onto the public.
The grid impact of the AI buildout is already showing up in utility planning documents across the northeastern United States. New York's regulators are not operating in a vacuum. The question in front of them is whether New York sets a precedent on cost allocation that other states follow, or whether developers route the next wave of projects to states where the regulatory framework still socializes those costs.
If New York's Public Service Commission adopts direct cost allocation rules with meaningful bite, the immediate effect is felt in the projects under active development, where pro formas built on current cost structures need to be revised or abandoned. The longer-term effect is on the site selection analysis for the next generation of projects.
Developers read regulatory decisions across states as signals about where to build the next project. Community opposition and regulatory tightening in one jurisdiction accelerates permitting timelines in competing jurisdictions. Virginia's electricity cost increases sent developers looking at Ohio, Indiana, and the Carolinas. New York adopting strong direct cost allocation would send the same signal. The facilities that are currently in permitting in New York might complete. The ones still in site selection might not choose New York.
The operators building data center infrastructure at 500 MW scale are not small companies making marginal capital allocation decisions. They are hyperscalers and large neoclouds with site selection teams that model regulatory trajectories across dozens of jurisdictions simultaneously. A state that changes the cost allocation rules in a way that adds $200 to $400 million to a project's effective capital cost does not need to pass a moratorium to change the development calculus. It just needs to make the math worse than the alternatives.
New York's decision will be watched carefully. The state has a large existing utility infrastructure, access to significant hydro capacity from NYPA, a sophisticated regulatory apparatus, and a development pipeline that has attracted serious capital. It also has a governor who wants the economic development and a ratepayer base that is already paying more for electricity than almost anywhere else in the country. The outcome of the cost allocation proceeding will determine whether New York captures a meaningful share of the next phase of AI infrastructure buildout, or prices itself out of it.
The cooling infrastructure angle is direct: facilities that internalize grid upgrade costs will be designed with tighter efficiency constraints from day one. A developer who has just written a $300 million check for substation capacity is not going to tolerate a cooling system that runs a PUE of 1.5 and blows through the energy budget. The water-power tradeoff that operators keep getting wrong becomes a first-principles design constraint when every watt carries a fully-loaded cost. Cost allocation reform, if it lands, will do more to accelerate liquid cooling adoption in New York than any industry roadmap.