Seven Signatures, One Fed Model, and 0.13 Points of Inflation

The White House's Ratepayer Protection Pledge, signed by Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI, dominates today's news cycle, but the Federal Reserve Bank of Dallas quietly dropped the more consequential analysis: data center power demand will double from 40 GW to 80 GW by 2031, pushing annual PCE inflation up 0.04 to 0.13 percentage points by 2030, with the effect nearly doubling if renewable deployment lags. The Dallas Fed's model is the first to quantify datacenter-driven electricity inflation using granular plant and facility data, giving Texas policymakers and grid planners a hard number to work with just as the White House asks industry to self-police on costs.

Seven hyperscalers committed to covering the full cost of new generation and grid upgrades for their datacenters. Entergy jumped in fast, announcing $5 billion in projected customer savings across Arkansas, Louisiana, and Mississippi over 20 years from datacenter PPAs already in place, and explicitly aligning its "Fair Share Plus" framework with the White House pledge.

But the pledge has zero disclosed penalty mechanisms, no verification protocols, and no timeline for implementation. Ed Hirs, University of Houston energy economist, flagged that the seven signatories represent only a fraction of datacenter operators. The Dispatch's analysis cuts deeper: in a capital-intensive network where costs are recovered over decades across layered federal and state institutions, "who pays" has no clean answer. Large new customers can sometimes *lower* average rates, depending on allocation structures.

For Texas specifically, the state's deregulated wholesale market means large datacenters already negotiate directly with generators or sign bilateral PPAs. The political pressure to formalize cost-sharing could reshape how ERCOT-area utilities structure megacustomer deals. Bottom line: treat the pledge as a political signal, not a binding framework, and watch for whether the PUCT moves to codify any of it.

EPRI now projects U.S. datacenters will consume 9-17% of total electricity by 2030, up from 4-5% today and a 60% increase over its own 2024 forecast. Annual consumption could climb from 177-192 TWh to 380-790 TWh. Virginia's already at 25% state share and could hit 41-59%.

Seven states beyond Virginia, including Arizona, Indiana, and Nevada, may exceed 20% datacenter electricity share. That geographic diffusion is reshaping site selection: developers are actively relocating to regions with generation surplus, land, and faster permitting. The Dallas Fed's inflation modeling reinforces EPRI's warning from a different angle: supply chain constraints on turbines, batteries, transformers, and chips could bottleneck both datacenter deployment and generation buildout simultaneously. If new generation can't keep pace, wholesale prices spike and the inflation math gets ugly.

Bottom line: the grid timing mismatch, where datacenters scale in months and power plants take years, is the single biggest execution risk in the sector right now.

Texas generated 58,634 GWh from utility-scale solar in 2025, surpassing California's 53,713 GWh for the first time in a full calendar year. Purdue's Kevin Kircher attributed the milestone to a regulatory environment that lowers barriers to energy project development. Solar grew 35% nationally, but it still delivers one-sixth the output of natural gas, which commands 40.8% of U.S. generation.

On the hyperscaler front, early reports suggest Amazon is advancing an 18-building campus beside Comanche Peak Nuclear Power Plant in Somervell County and a 21-building, 1,265-acre Project Spectrum in Hood County powered by Vistra subsidiary Luminant. Google's $40 billion Texas investment through 2027, including the Wilbarger campus with AES clean power, takes a different approach: third-party generation rather than nuclear co-location.

Meanwhile, Monarch Energy's pivot from green hydrogen to datacenter development tells a structural story. The company holds 4 GW of pre-permitted sites across ERCOT, SPP, PJM, and MISO, and says off-takers for green hydrogen simply wouldn't commit. Lancium, lead developer of Project Stargate's Abilene campus, was Monarch's first investor. About 2 GW of Monarch's portfolio suits hyperscaler needs; the rest sits in Gulf Coast areas too exposed to storm surge for sensitive compute.

Comanche Peak's transmission capacity is the constraint to watch: stacking 18+ buildings near a single nuclear plant creates interconnection bottlenecks if ERCOT's batch study process doesn't clear quickly enough.