Trump Targets AI Power Costs // Exxon + NextEra’s 1-GW Build // INL Picks Its First Reactor Experiments

The political world is finally waking up to the fact that America’s power grid is being reshaped not by climate targets or regulators, but by the raw physics of AI and the speed at which hyperscalers can hurl billions at concrete and copper. This week brought a presidential foray into rate policy, an oil-and-renewables alliance to build a gigawatt-scale compute complex, and a federal push to accelerate next-generation reactors. Taken together, they sketch a simple picture: the demand curve is now moving faster than the institutional machinery built to serve it.

Trump Steps Into AI Power Pricing, Turning a Regulatory Fight Into a National Issue

Donald Trump has begun framing skyrocketing AI-driven electricity demand as a pocketbook issue, telling supporters he would stop utilities from “jacking up rates” on ordinary families to subsidize hyperscaler data-center buildout. The line lands because commissions in several states — most notably Florida — are already weighing how much of the cost of new gas plants, transmission expansions, and interconnection upgrades should fall on residential ratepayers. Florida Gov. Ron DeSantis echoed the theme this week, accusing utilities of using data centers as an excuse to shift billions in future infrastructure costs onto customers.

But Trump’s comments arrived the same week he previewed an AI executive order aimed at preempting state AI laws, not utility regulations — raising questions about whether a federal attempt to police electricity rates is even legally plausible. Electricity pricing is overwhelmingly a state function, and folding rate design into an AI rulebook would be a regulatory reach of historic proportions. Still, the politics are unmistakable: AI load is expanding faster than utilities can build capacity, and the question of who pays is quickly becoming a national wedge issue rather than a regulatory footnote.

Grid Take: Trump’s instinct to protect ratepayers is understandable, but turning a structural supply problem into a price-control crusade is the wrong lever. The grid doesn’t need political caps — it needs more steel in the ground, more competition, and a serious supply-side agenda. If he focused there, ratepayers wouldn’t need protection in the first place.

Exxon and NextEra Build a 1-GW Compute-Energy Complex for a Hyperscaler

In a move that shows how fast the energy–AI industrial ecosystem is evolving, ExxonMobil and NextEra Energy are teaming up to build a 1-gigawatt data-center complex for an unnamed hyperscaler. The project blends Exxon’s vast industrial land and infrastructure footprint with NextEra’s generation portfolio and development muscle, creating what is effectively a privately engineered grid for a single customer.

A gigawatt-scale project of this kind isn’t just another data center — it’s a sovereign grid node. It requires firm generation, dispatchable backup, high-voltage interconnection, and on-site storage. And by partnering directly with energy majors, the hyperscaler bypasses traditional utility timelines, capital constraints, and regulatory uncertainty. Exxon has signaled this may be a repeatable template: converting existing industrial sites into vertically integrated power-compute hubs that co-locate hydrogen, carbon management, and AI workloads.

Grid Take: This is evolutionary — hyperscalers acting like sovereign grid actors, building private energy islands at national scale. It’s likely good for innovation, but the challenge is ensuring these breakthroughs feed back into the broader grid rather than creating hyper-efficient silos walled off from everyone else.

INL Selects Its First NRIC Reactor Experiments, Accelerating the Advanced Nuclear Pipeline

The Idaho National Laboratory has named the first two projects to operate under the National Reactor Innovation Center (NRIC), giving private developers access to real reactor conditions — something the U.S. nuclear industry has lacked for decades. Kairos Power will conduct molten-salt loop experiments to validate thermal-hydraulic behavior and component integrity for its Hermes low-power demonstration reactor, and Ultra Safe Nuclear Corporation (USNC) will undergo irradiation testing for its proprietary TRISO fuel forms and high-temperature materials.

This infrastructure matters. Most advanced reactor companies have been trapped in a cycle where designs improve on paper but never confront the messy physics of high-temperature materials, neutron environments, salt chemistry, or thermal transients. NRIC provides that missing rung: a bridge between modeling and licensable hardware. With SMRs finally getting federal cost-share dollars and utilities beginning to name sites for next-generation reactors, NRIC’s timing is unusually aligned with the broader political and financial momentum behind new nuclear technologies.

Grid Take: With SMRs fashionable again, NRIC risks becoming the next buzzword — but it’s a necessary one. If we want real innovation instead of PowerPoint reactors, this is exactly the infrastructure the U.S. must push hard to bring into full use.

Upgrade to Grid Brief Premium to get extra deep dives into energy issues all over the world.

Conversation Starters

• Fortune — “IBM’s CEO says hyperscalers can’t profit from the data-center buildout.”
  A strong explainer on the economics of infinite capex: revenue growth may track compute demand, but power, land, and chips scale faster than margins.

• Futurism — “Bitcoin miners are stealing billions in electricity.”
  A reminder that the grid has freeloaders too—just not the ones lawmakers tend to yell about.

• The Hill — “Gas prices and power bills rise as affordability becomes the new political wedge.”
  Energy costs are now polling as a top-three voter concern heading into 2026; both parties know it.

Chart of the Day: Hyperscaler capex vs. cloud profits

Amazon, Alphabet, and Microsoft together spent well over $180 billion on capital expenditures in 2024, yet their combined cloud-segment operating income was under $84 billion. Amazon alone plowed about $84B into capex against $40B in AWS operating income, while Google Cloud earned just $6.1B on top of more than $52B in Alphabet capex. The IBM CEO’s line that there’s “no way” hyperscalers make real money on this kind of data-center arms race doesn’t look crazy when you remember these are long-lived assets with multi-year depreciation tails, not one-and-done marketing spends.

Good Bet / Bad Bet

Good Bet: The power landlords, not the AI landlords
If this capex chart is even roughly right, the cleanest balance sheets in the AI boom may belong to the people selling electrons and real estate, not the folks torching tens of billions a year on GPUs. Regulated utilities and transmission owners with long-term contracts to data centers get growing, relatively predictable cash flows without having to guess which LLM wins. The hyperscalers compete on models and features; the grid just sends them the bill.

Bad Bet: “AI capex always pays”
This is the late-’90s fiber bubble with better hoodies. Telcos once believed that if they buried enough glass in the ground, profits were inevitable; instead they spent a decade writing down “dark fiber” no one used. Today’s hyperscalers are on track to spend hundreds of billions of dollars on AI and data-center capex by 2025, with one estimate putting Big Tech’s 2025 capex at $441 billion, up 184% from 2023. If you assume that kind of spend must translate into fat, durable margins, you’re not investing—you’re praying the laws of capital intensity suddenly stopped applying to this one sector.

Share Grid Brief

We rely on word of mouth to grow. If you're enjoying this, don't forget to forward Grid Brief to your friends and ask them to subscribe!