On March 4, executives from Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI gathered at the White House to sign the Ratepayer Protection Pledge. Under the agreement, America’s largest technology companies must “build, bring or buy” the power their data centers need, pay for grid upgrades their facilities require and negotiate separate rate structures with utilities and state governments.
The pledge is voluntary. It has no enforcement mechanism. And it says little about the small and midsize businesses caught between rising commercial electricity rates and a compute market increasingly designed for companies that can build their own power plants.
Voluntary climate and infrastructure commitments from the same group of companies have a mixed track record — Microsoft and Google have both publicly walked back earlier carbon pledges as AI compute demand has grown — and the pledge’s “build, bring or buy” language leaves significant room for interpretation.
“Big Tech signing up to self-power their data centers tells you one thing — the electricity needed to run AI is now as valuable as the AI itself,” says David Sherman, head of brand strategy at distributed compute network io.net. “Companies like Google, Meta and Microsoft can afford to build their own power stations, but that option simply doesn’t exist for the thousands of startups, scale-ups and midsized businesses that also need serious computing power to compete.”
The risk, he warns, is “a two-tier tech economy where only the richest firms get reliable access to the infrastructure AI demands, while everyone else faces rising costs, longer wait times and fewer options.”
The bill, in numbers
US data center construction spending reached $41 billion in 2025, up 32% from 2024 and more than quadrupling 2020 spending, according to data from the US Census Bureau. The buildout has reshaped local power markets. A fall 2025 Bloomberg analysis found that wholesale energy prices rose by as much as 267% over five years in areas near significant data center activity.
In PJM Interconnection — the country’s largest grid operator, covering 13 states from Illinois to Virginia — capacity prices for the 2026–27 delivery year rose to $329.17 per megawatt-day, more than 11 times the rate two years earlier. PJM’s independent market monitor attributed 45% of the December 2025 auction’s $16.4 billion cost to data center demand. The price cap that softened the last two auctions expires after the next one, scheduled for May or June.
Behind the numbers is a scale issue which may explain why hyperscalers are racing to self-power in the first place.
“A single large facility can consume as much electricity as a medium-sized town, which means clusters of data centers can place significant pressure on local grids,” says John Haw, CEO of UK energy procurement firm Fidelity Energy.
For companies that want to scale AI reliably, he argues, “securing dedicated generation capacity — through power purchase agreements, on-site generation or private energy projects — gives operators more control over costs and avoids long delays for grid connections.”
This option isn’t on the table for everyone. Small and midsize businesses on commercial tariffs feel the rate pressure directly. The pledge’s protections, where they hold, are aimed at residential bills.
The physics
Even if every company wanted to follow the hyperscaler playbook, the supply chain may not be able to support it. Adnan Masood, chief AI architect at technology services company UST, argues that the mainstream image of AI — code, models, chips — conceals what the industry actually builds.
“In the trenches of the industry builds, the real picture is substations, transformers, cooling and water,” he says. Once AI campuses start drawing the load of a small city, “the question is no longer just who has the best model; it is who can secure power without asking ordinary customers to underwrite private compute expansion.”
A signature on a pledge, he adds, “would not magically clear an interconnection queue, produce transformers, solve water constraints or settle local permitting fights. AI has entered the phase where steel, copper and permits matter as much as code. You cannot prompt your way around a transformer shortage.”
Harry Sudock, chief business officer at bitcoin miner CleanSpark, sees those same physical constraints — but reads them as evidence the pledge is well-timed rather than naive. “The pledge is admirable and highlights a key need in our power system: the time to build additional generation is now,” he says. He notes that bitcoin operations can locate wherever power is cheapest, but AI workloads can’t: “AI and HPC data centers are not necessarily able to follow this playbook, as they’re tied to fiber and latency requirements, customer uptime needs and other cooling form factors.”
What the pledge gets right, in his view, is putting the burden where it belongs. “Scalable, responsible AI infrastructure will bring additional generation online faster than utilities were able to over the prior decades — the benefits are clear, and the burden is on the developers, and their tenants, to scale responsibly.”
“Self-powered is not the same as clean”
The pledge is silent on what kind of generation hyperscalers should build. That has alarmed analysts who note the path of least resistance — and shortest lead time — runs through natural gas. His view is that by 2030, the strongest AI companies “will not just run bigger clusters. They will run a better energy stack, and that will be the real competitive advantage.”
Fidelity Energy’s Haw echoes the concern from a procurement perspective: “If AI demand outpaces renewable deployment and grid reinforcement, companies may turn to faster-to-build gas generation to guarantee supply. That creates a real tension between digital growth and climate targets.”
Working with the grid
For companies that can’t build their way out — which is most of the AI economy — alternatives exist. Ayse Coskun, chief scientist at Emerald AI, argues that flexibility is faster to deploy than generation. Her company’s partnership with InfraPartners, announced in March 2026, introduced what they call Flex-Ready data centers — facilities designed to dynamically adjust load in response to grid conditions.
Building new generation takes years of planning and permitting. By dynamically adjusting power consumption, “data centers become grid-aware power consumers that are more controllable and predictable in how they use electricity.” The result, she argues, is “a mutually beneficial model where AI infrastructure can scale while supporting a more reliable, affordable and sustainable power system.”
Sherman makes a related case from the demand side. Distributed compute, he says, “uses computing hardware that already exists and is already plugged in, avoiding the need for costly new power plants while reducing strain on overstretched grids.”
A pledge with commercial logic
Whatever its shortcomings, Masood argues the shift is both necessary and commercially self-reinforcing. “If hyperscalers bring additive generation, pay for delivery upgrades and accept take-or-pay commitments, AI growth becomes more credible with regulators, communities and customers,” he says.
The end state is better for consumers because it reduces cost shifting and better for businesses because power procurement becomes part “of the engineering plan instead of a late surprise.”
The demand picture may be softer than it looks
Some analysts caution that the headline numbers driving the pledge debate may be overstated. An April analysis from the Information Technology and Innovation Foundation, a Washington-based tech policy think tank, noted that new data center deals underway fell more than 40% between the third and fourth quarters of 2025. Of that 240 GW of planned construction, only one-third is actually being built, the report’s author argues, and OpenAI’s $500 billion Stargate project in Texas appears to have stalled.
Even if demand holds, Haw argues the picture isn’t entirely zero-sum. The scale of hyperscaler investment, he says, could ultimately accelerate broader energy infrastructure: “When companies of this scale invest directly in power infrastructure, it can bring forward new renewable capacity, storage and grid upgrades.”
He closes by highlighitng what defines the SMB squeeze: “The companies that secure reliable, low-carbon electricity first will have a major competitive advantage.”