Florida has turned the data center boom into a cost allocation test. The state is telling major power users that growth is welcome, but ordinary customers should not be left paying for the grid upgrades that make it possible.
Gov. Ron DeSantis signed SB 484 on May 7, putting Florida among the states moving fastest to draw a line between AI infrastructure investment and household electric bills. The law requires large power users, including major data centers, to cover the full cost of service tied to their projects, from connections and new transmission to incremental generation and operations expenses.
The threshold is clear. A large load customer is any customer with an anticipated monthly peak load of 50 megawatts or more at a single location, measured as the highest average load over a 15-minute interval. A large-scale data center uses the same 50-megawatt test. The law does not let a company stitch together smaller connections at one site to dodge the classification, and it counts colocated customers operating at the same location when the combined site meets the threshold.
That matters because modern AI campuses are no longer ordinary commercial buildings with unusually high server bills. A 50-megawatt site can demand power on the scale of a small city, and the biggest proposed campuses can multiply that several times over. Florida's law does not name artificial intelligence as the target, but the practical audience is obvious: hyperscalers, cloud providers, colocation operators and AI infrastructure companies whose projects can force utilities to build expensive capacity years before the economic payoff is proven.
The Florida Senate's enrolled text requires public utilities to file compliant tariffs with the Florida Public Service Commission by October 1, 2026. Those tariffs must ensure that large load customers bear their own full cost of service and that nonpayment risk is not pushed onto the general body of ratepayers. The utilities covered are public electric utilities under Florida's utility statute, which generally means investor-owned utilities regulated by the PSC rather than municipal electric systems or rural electric cooperatives.
The tools available to regulators are not symbolic. The PSC may approve tariffs with construction contributions, required customer infrastructure investments, demand charges, incremental generation charges, financial guarantees, minimum load factors, take-or-pay provisions and minimum service terms with early termination fees. In plain English, a data center that asks the grid to expand may have to put real money and binding commitments behind that request.
This is a different posture from the incentive-heavy economic development model that helped data centers spread across Virginia, Georgia, Texas and other power-rich markets. For years, communities competed for facilities by offering tax breaks, fast permitting and quiet utility arrangements. The new question is whether the public gets stuck with the downside if load forecasts are wrong, projects are delayed, or a large customer walks away after infrastructure has already been built.
Florida also preserved local zoning and land-use authority, which is important in a state where water pressure can be as politically sensitive as electricity costs. Large-scale data centers seeking water allocations face a separate permitting framework, including hearings for permits. Projects requesting at least 100,000 gallons per day must also submit conservation plans. If reclaimed water is feasible and available at the property boundary, regulators can require its use instead of surface or groundwater.
Maryland Shows The Pressure Point
The timing is not accidental. The Maryland Office of People's Counsel has been warning that data center growth in the PJM power market is already showing up in customer bills through capacity, transmission and energy costs. The agency says PJM advanced almost $12 billion in new transmission infrastructure between 2024 and 2025, much of it driven by data center growth concentrated in Northern Virginia, with Maryland customers assigned more than $1.3 billion plus future utility profits.
That is the fight Florida is trying to avoid before it fully arrives. Maryland's problem is regional and tangled in PJM cost allocation rules. Florida's law is more direct. If a massive new load requires infrastructure, the customer causing that need should pay for it through a tariff designed before the project becomes a political emergency.
For startups, the policy cuts both ways. On one hand, clear tariffs can make site planning more predictable and reduce backlash against new AI infrastructure. A founder building an inference cloud or specialized training cluster can at least understand what the state expects before negotiating with a utility or county commission. Predictability has value when power availability has become one of the hardest constraints in AI.
On the other hand, the rules may favor cash-rich incumbents. Amazon, Microsoft, Google and Meta can absorb deposits, long-term commitments and take-or-pay contracts more easily than a young AI infrastructure startup still proving demand. If more states adopt Florida-style requirements, access to cheap capital may become as important as access to GPUs. Smaller players could be pushed toward leasing capacity from larger platforms instead of building their own campuses.
The larger market signal is that AI infrastructure is leaving its first phase. The story is no longer only about chips, land and speed to deployment. It is about who pays for the wires, substations, water systems and standby capacity that make the compute economy possible. Florida has offered one answer, and other states watching power bills rise will have little reason to ignore it.
Also read: OnlyFans shows how creator platforms can be valuable and hard to buy • Canvas restored after a hack exposes the risk inside free SaaS accounts • Wagyu allegations show Monero DeFi has a trust problem
