Utilities have disclosed more than 187 gigawatts of data center load commitments across 19 regulated utilities, a figure aggregated from public earnings disclosures that has circulated widely as evidence that demand for artificial intelligence is reshaping the American grid.
But what exactly does a commitment entail? New research from the investment firm Jefferies suggests that a stricter definition of these agreements could drop that total to as low as 69 GW across 19 regulated utilities.
Jefferies analysts separated disclosed data center load into three tiers.
Reported commitments cover everything utilities describe as “committed” or “firm” regardless of whether binding financial terms have been disclosed. The next tier, at 151 GW, covers arrangements where binding financial backing can be reasonably inferred from public materials even if the contract structure is not fully disclosed. Apply a stricter standard of disclosed contractual terms through electric service agreements ESAs or equivalent instruments, and the total falls to 107 GW. And a view limited to formal ESAs alone includes just 69 GW.
This difference in scale has direct consequences for anyone building toward this demand. A developer sizing generation, planning transmission, or negotiating a clean energy offtake needs to know how much of that disclosed load will actually materialize. And as Latitude Intelligence detailed in our recent analysis of 25 utility data center tariffs, utilities have spent the last two years constructing an entirely new contractual framework for this load class — long-term contracts, higher minimum billing demands, collateral requirements — all sized to the disclosed demand numbers.
Jefferies finds that roughly 43% of total reported commitments lack firm disclosure of binding financial obligations. Part of the issue is that utilities are each defining a binding commitment through different contract labels, regulatory frameworks, and financial structures, which makes any sector-level aggregation inherently imprecise. The numbers are not contradictory so much as they answer different questions about what qualifies as signed.
Five utilities — AEP, Dominion, Entergy, Sempra, and Southern — hold roughly 56% of explicit binding commitments in the dataset, and PJM and Southern utilities together account for about 80%. Where the AI buildout lands, where the grid strain concentrates, and where the clean energy procurement will flow is largely going to be determined in those territories.
AEP is the quintessential example of a utility planning around uncertainty. The company reported 56 GW of contracted load in February 2026, the largest disclosed book in the sector and the fastest growth, up from 28 GW just four months earlier. However, 36 GW of that total sits in ERCOT, where the contracting structure is a letter of intent rather than an executed service agreement. That distinction matters as ERCOT’s large-load queue now exceeds 225 GW, and experts widely estimate that speculative requests run several times actual builds.
Meanwhile, Entergy’s 15.5 GW represents the largest single block of load backed by fully executed, disclosed ESA terms in the dataset. Southern Company’s 10 GW carries minimum 15-year durations, take-or-pay provisions, and collateral requirements. And while Duke’s signed book is smaller at 4.5 GW, every one of those agreements is already under construction, which is a level of execution certainty that most of the sector cannot match.
