Georgia residential electricity bills climbed 43% between 2023 and 2025. Not because Georgians started mining Bitcoin in their garages. Because data centers moved in, drew enormous load, and the grid had to expand to serve them. Senator Jon Ossoff is now investigating exactly how that happened. The answer is not complicated: utilities filed rate cases, regulators approved them, and households paid.
Meanwhile, in February 2026, the White House announced that Meta, Google, and OpenAI had pledged to cover a larger share of data center energy and grid infrastructure costs. In March, a formal "Ratepayer Protection Pledge" followed. These announcements read like a fix. They are not. Both pledges are voluntary, with no published enforcement mechanism. A pledge without a penalty is a press release with extra steps.
The Gap Between the README and the Running Code
I spend a lot of time reading documentation that does not match what the software actually does. This situation has the same smell. The documentation says tech companies are stepping up. The running code, which is the actual rate structure, says residential customers absorb the cost of grid upgrades driven by data center demand.
The Belfer Center analysis lays out the core dispute clearly. Utilities argue that whoever caused the upgrade should pay 100% of it. Data centers argue the grid is shared infrastructure, so costs should spread broadly. That second argument is not crazy on its face: shared infrastructure does get shared costs in most utility frameworks. But it breaks down when the load growth is this concentrated. PJM Interconnection's congestion costs rose 64% in 2024. Texas is planning $30 billion in transmission upgrades. Individual AI data centers now exceed 1 gigawatt of demand. This is not diffuse load growth. This is a specific, traceable spike.
Utilities plan more than $1.4 trillion in grid investments over the next five years, much of it tied to data center demand. One recent report found that utility load forecasts may be overestimating data center growth by 40%. That is the part that should make you angry. Ratepayers could end up funding infrastructure for demand that never arrives, because the hyperscalers revised their buildout plans or found cheaper power elsewhere. Stranded costs in utility accounting do not disappear. They get recovered from whoever is left on the grid.
What Actually Needs to Ship
Some states are moving in the right direction. Several regulatory agencies have proposed large-load tariffs requiring data centers to pay higher upfront fees and commit to long-term contracts. Dominion Energy requires $1.5 million in collateral per megawatt of contracted capacity. That is a real mechanism with real teeth. Southern Company's current large-load rate class, according to an April 2026 shareholder proxy memo, does not fully protect residential and small-business customers. That gap is not a policy nuance. It is a transfer of wealth from households to hyperscalers.
The fair point for the other side: data centers do create jobs and tax revenue in the states that host them, and some of that flows back to residents. Fine. That does not justify making a family in Marietta pay higher electric bills so that a server farm in Douglasville can train the next foundation model at subsidized rates.
State utility commissions need to stop treating voluntary pledges as a substitute for mandatory cost-allocation rules. The model exists: large-load tariffs with collateral requirements and contract commitments. The Dominion structure is not perfect, but it is enforceable. Regulators in every PJM and Southeast state should be implementing something equivalent before the next rate case lands. The $25 billion in environmental and grid costs that data centers imposed on the economy last year did not distribute themselves voluntarily either.
