On March 4, Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI stood in the White House and signed a pledge that their data centers won’t raise electricity bills for American households. It includes commitments to build, bring, or buy their own power, cover the cost of grid infrastructure upgrades, and pay for the electricity capacity dedicated to their facilities, whether they actually use it or not.
It looked like a breakthrough: the tech industry finally accepting accountability for its energy footprint.
In reality, though, it was a voluntary commitment to principles that state utility commissions have spent two years trying to make enforceable and haven’t finished yet. The pledge has no audit mechanism, no defined terms — and ultimately no teeth.
The state-level tariffs do. But as of today, the tariffs don’t cover most of the country.
Eighteen of the 25 data center-specific utility tariffs now on the books were filed or approved in 2024 and 2025 alone, with several still pending. A new Latitude Intelligence analysis of all 25 tariffs in 19 states is the first systematic primary-sourced review of the category; it documents what these enforceable structures actually look like, how aggressive they already are — and what some of the binding frameworks got wrong.
The legally binding version is not gentle
The White House pledge commits hyperscalers to pay for power “whether they use the electricity or not.” In tariff terms, that is a billing floor — a mechanism that 15 of the 25 state-level tariffs have already embedded through specific numeric thresholds, averaging around 80% of contracted demand. For utilities like Idaho Power and FPL, which previously billed on actual consumption alone, these tariffs created a revenue floor that had never existed for any customer class.
Stack that floor against long contract terms and upfront collateral, and the terms look nothing like a press-event commitment. AEP Ohio requires an 85% billing floor over an eight-year minimum term. FPL locks customers in for 20 years. Dominion Energy Virginia requires collateral of $1.5 million per megawatt; for a 100 MW facility, that’s $150 million posted upfront before a single kilowatt-hour flows.
None of this is accidental. Residential electricity prices have risen more than 36% since 2020, and in PJM, capacity prices surged 833% in just two consecutive delivery years. These hikes are driven substantially by data center load growth.
The first data center-specific tariff dates to 2018, when the New York Municipal Power Agency created a dedicated rate class for high-density loads. However, of the 25 tariffs now on the books, 18 were filed or approved in 2025 and 2025.
And tariffs are just one track of the states’ four-track response. The map understates regulatory activity in the densest markets: California, Georgia, Illinois, and Texas technically have no published tariffs but have each pursued alternative frameworks through legislation, individual contracts, or regulatory proceedings.
This work began long before protecting ratepayers from data center costs became the national political issue of the day. The Ratepayer Protection Pledge demonstrates that the White House has arrived at the same conclusion that New York’s municipal utilities did, just seven years later and without the enforcement mechanism that makes the effort mean anything.
A voluntary commitment with no audit mechanism and no defined standard for what “negotiating a separate rate structure” means is not the same thing as a tariff that’s binding in a rate case.
The part nobody has made enforceable yet
The pledge’s fifth commitment is the one that got the least attention. The hyperscalers also agreed to “coordinate with grid operators to make backup generation capacity available during periods of electricity scarcity.” It’s a flexibility mechanism, which is the one thing that none of the 25 state-level tariffs have included a framework for.
A 2024 DOE study found that utilities can typically accommodate data center demand approximately 350 days per year, with only the remaining 15 days (roughly 360 hours) actually straining the grid. A GridLab analysis of NV Energy found that curtailing just one gigawatt of data center load during 500 to 880 summer peak hours could yield significant system cost savings and accelerate interconnection timelines. An ImpactECI analysis of PJM’s Dominion Zone found that 50 to 350 hours of targeted curtailment would free 6% to 17% of total system capacity (several GW of headroom) without building a single new power plant.
None of the 25 tariffs capture any of it. Not one offers a pricing incentive for load reduction during peak hours, and not one creates a pathway for data centers to provide the service the pledge now gestures toward.
The grid already has markets that pay for exactly that service. PG&E’s Emergency Load Reduction Program pays for dispatched curtailment, and PJM’s capacity market compensates resources that deliver during emergencies. Meanwhile, companies are developing new load flexibility approaches; Oracle and Nvidia demonstrated a 25% reduction in data center power consumption during a live grid stress event at a Phoenix facility in 2025.
The capability exists, and is developing fast. But no tariff in the dataset connects data centers to the compensation for using it.
Texas got there through legislation rather than tariff design. SB 6, signed last June, requires new large loads to install curtailment equipment and directs ERCOT to build a voluntary demand response structure, making it the third venue (after state legislatures and now the White House) to endorse the same flexibility logic without embedding it in anything enforceable.
So now, major tech companies have voluntarily committed to grid coordination that regulators, with binding legal authority, have not yet managed to design into law.
What the next 12 months will actually settle
The Ratepayer Protection Pledge is not the conclusion of this regulatory debate. It is merely a signal that the principle of full cost assignment for data center load is no longer contested at the highest level — which matters more for the state-level proceedings now underway, where the actual terms are still being set.
Pennsylvania is finalizing the first model tariff template, with an 80% billing floor and 15-year term. Texas has five active SB 6 rulemakings underway, with final rules expected throughout 2026. DTE must file Michigan’s first general data center tariff within 90 days of its Oracle contract approval. And Virginia’s legislature has introduced approximately 30 data center reform bills this session, covering cost allocation, demand response, and the conditions on the state’s $1.9 billion annual tax exemption.
These proceedings will determine whether the enforceable framework catches up to what the voluntary one promised, or stays narrowly focused on cost protection while the flexibility that regulators, legislators, and now the White House have all endorsed remains unbuilt.
The companies publicly agreed, at a presidential press event, that they should pay for what they consume. The harder work is designing the version that holds when the cameras are gone.
The full analysis, covering all 25 tariffs across affordability, flexibility, and clean energy, is available now in the Latitude Intelligence report, “The terms of power: Inside the new utility rates for data centers” is available here for download.
