President Trump announced a “ratepayer protection pledge” during his State of the Union address on February 24, requiring tech companies to build, bring, or buy their own power supply for new AI data centers. Energy Secretary Chris Wright confirmed all major hyperscalers—Google, Meta, Amazon, Microsoft—have signed on. The White House will formalize the policy at a March 2026 meeting. The goal: prevent American electricity bills from rising as AI infrastructure explodes. The reality: this shifts $50-100 billion in power plant costs from ratepayers to Big Tech, assuming enforcement happens.
The Grid Crisis Forcing Trump’s Hand
Data center power consumption doubled between 2018 and 2024, according to Lawrence Berkeley National Laboratory, and is projected to triple again by 2028. Data centers consumed 4.4% of total US electricity in 2023; by 2028, that jumps to 6.7-12%. Individual facilities now require 50-300 megawatts—equivalent to mid-sized cities.
Utility customers in seven PJM states paid $4.4 billion in 2024 for transmission upgrades to bring data centers online. Utilities were upgrading grids for Big Tech, then passing costs to residential ratepayers. Trump’s policy aims to stop that subsidy.
Whether it succeeds depends on enforcement details no one has explained yet.
What Building a Power Plant Actually Costs
Trump says “build, bring, or buy.” Here’s the price tag.
Nuclear: $8-12 billion, 10-20 years from application to operation. Small modular reactors (SMRs) promise faster timelines, but NuScale’s pilot ballooned from $3 billion to $9.3 billion before cancellation in 2023. NRC licensing still takes 3-5 years minimum.
Natural gas: $1-2 billion, 3-5 years total. Fastest path to operational power. Also fossil fuel, contradicting every clean energy commitment these companies made.
Solar/wind: $500 million to $2 billion, 2-4 years. Problem: intermittent power. Data centers run 24/7. Renewables require massive battery storage, adding billions more.
Tech companies spent $650 billion on AI infrastructure in 2026. Adding $1-2 billion per gas plant is manageable. Adding $8-12 billion per nuclear plant would nearly double datacenter costs. For ten facilities, that’s $10-120 billion depending on technology choice.
Economics push toward natural gas. Politics push toward nuclear or renewables. Something has to give.
Can Tech Companies Even Operate Power Plants?
Microsoft’s Three Mile Island deal with Constellation is a power purchase agreement (PPA)—Microsoft buys power, Constellation operates the plant. Google’s Kairos Power agreement for 500 MW of SMR capacity by 2035 is also a PPA. Amazon’s $500 million X-energy investment targeting 5+ gigawatts by 2039—again, PPAs and investment, not ownership.
None of these companies operate power plants themselves. Operating nuclear requires NRC licenses and continuous safety compliance. Gas plants need EPA air quality compliance and grid management. Tech companies would need to build entire energy divisions.
Trump’s policy says “build, bring, or buy.” Does “buy” include PPAs—the current strategy—or require outright ownership? If PPAs count, the policy changes nothing. If they don’t, forced ownership is an unprecedented mandate.
No one has clarified this. That ambiguity makes the policy either toothless or transformative.
The Real Winners: Energy Construction and SMR Startups
If enforced with ownership requirements, energy construction firms (Fluor, Bechtel, Westinghouse, Black & Veatch) could see $50-200 billion in contracts over the next decade.
SMR startups are positioned for a windfall: NuScale Power (only NRC-approved SMR design), Oklo (Sam Altman-backed microreactors), Kairos Power (Google’s partner), X-energy (Amazon’s $500M investment), TerraPower (Bill Gates’ company).
If natural gas wins, fossil fuel companies and gas turbine manufacturers (GE Vernova, Siemens Energy) capture the market. If solar-plus-storage becomes competitive, renewable developers benefit.
Losers: Traditional utilities lose large customers and ratepayer-subsidized revenue. Smaller AI companies can’t afford billion-dollar power plants. If enforcement is weak, residential ratepayers still subsidize Big Tech.
The Unanswered Questions
Is this voluntary or regulatory? What penalties exist? Who enforces it—DOE, EPA, state utility commissions? Does it apply retroactively or only to new facilities? Do smaller AI companies have the same obligation? Can companies jointly fund shared plants?
Critically: what counts as “buy”? If PPAs qualify, nothing changes. If “buy” means purchasing existing plants, supply is limited. If “build” is the only compliant path, 3-20 year delays could slow the AI buildout happening now.
The March 2026 White House meeting will presumably clarify enforcement. Until then, this is political theater.
Power Is the New AI Bottleneck
After chips and memory, energy is now the limiting factor for AI infrastructure. Trump’s policy correctly identifies the problem: ratepayers shouldn’t subsidize Big Tech’s trillion-dollar buildout. The $50-100 billion cost transfer makes economic and political sense.
But implementation will be messy. Natural gas is the only realistic 2026-2030 option, contradicting clean energy commitments. Nuclear takes 10-20 years despite SMR promises. Renewables can’t provide 24/7 baseload without battery infrastructure that doesn’t exist.
The real winners are energy construction firms and SMR startups. The real question is whether this policy has enforcement mechanisms or remains a voluntary pledge.
We’ll find out in March.
