While the entire tech industry was busy watching Anthropic close a $900 billion valuation and OpenAI prep its IPO, the deal that actually tells you where AI is headed got almost no attention. NextEra Energy just agreed to acquire Dominion Energy for $67 billion in the largest utility merger in American history — and the reason has nothing to do with windmills or natural gas. It has everything to do with the fact that AI data centres are about to consume more electricity than some countries, and the current grid cannot handle it.
The combined company will have an enterprise value of roughly $420 billion and a market capitalisation of approximately $249 billion. NextEra shareholders get 74.5% of the new entity; Dominion shareholders get 25.5% plus a pro rata share of $360 million in cash. Management is projecting 9%+ adjusted earnings-per-share growth and $59 billion in annual capital expenditure through 2032. Those numbers don’t make sense for a traditional utility deal. They only make sense if you understand what’s actually being built.
The AI Industry’s Dirtiest Secret: It Needs More Power Than It Can Buy
Every frontier AI lab — Anthropic, OpenAI, Google DeepMind, xAI — is now constrained not by the availability of Nvidia GPUs but by the availability of electrical power. Training a single frontier model can consume the equivalent of powering 30,000 homes for a year. Inference at scale — the actual running of these models billions of times a day — is even worse. Industry projections suggest AI data centres will consume 15–25% of all U.S. electricity by 2030, up from roughly 4% today.
That’s not a gradual ramp. That’s a quadrupling of power demand in four years, on a grid that was already struggling with summer brownouts in Texas and rolling blackouts in California. The physics are brutal: you can design a new chip in 18 months, but building a new power plant takes 5–7 years. Building transmission lines to connect that plant to a data centre takes even longer, thanks to permitting battles that can drag on for a decade.
This is the bottleneck nobody in Silicon Valley wants to talk about. It’s not sexy. It doesn’t fit on a slide deck. But it’s the reason NextEra just wrote a $67 billion cheque.
Why NextEra Specifically — And Why Dominion Specifically
NextEra operates the largest renewable energy portfolio in North America. It already generates more wind and solar power than any other company on the continent. Dominion, meanwhile, controls critical transmission infrastructure across Virginia, North Carolina, and South Carolina — the exact states where Amazon, Microsoft, and Google have been building their largest data centre clusters.
Virginia alone hosts more data centres than any other state in the world. Dominion’s grid literally powers the servers that run a significant chunk of the global internet. The problem? Dominion’s generation capacity was planned for a world where data centres were a small fraction of load. That world is gone. The hyperscalers need tens of gigawatts of new capacity, and they need it within five years.
NextEra is betting it can combine its renewable generation expertise with Dominion’s transmission footprint to build the power infrastructure the AI industry desperately needs. The $59 billion in projected annual capex through 2032 tells you the scale of what they’re planning: new solar farms, new battery storage, potentially new nuclear capacity, and — critically — the transmission lines to connect all of it to the data centre clusters.
Follow the Money: Who Actually Benefits
The merger still needs shareholder approval from both companies, Hart-Scott-Rodino antitrust clearance, FERC approval, NRC sign-off, and utility commission approvals in three states. That’s a minefield. Utility mergers are politically sensitive — regulators in Virginia and the Carolinas will want guarantees that residential rates won’t spike to subsidise data centre buildouts.
But the real winners here aren’t the utility shareholders. The real winners are the hyperscalers themselves. Amazon, Microsoft, and Google have been quietly signing power purchase agreements worth billions with individual utilities. A combined NextEra-Dominion entity gives them a single counterparty with the generation capacity, transmission infrastructure, and regulatory relationships to deliver power at the scale they need. It simplifies a process that was becoming unmanageably fragmented.
The losers? Smaller, independent utilities that were hoping to cash in on the AI power boom piecemeal. A $420 billion utility giant will have pricing power, lobbying muscle, and economies of scale that mid-tier players simply can’t match. Expect more utility consolidation in the next 18 months — this deal sets the template.
The Second-Order Effect Nobody’s Pricing In
Here’s what makes this deal more consequential than it appears: it permanently changes who controls the AI supply chain. For the past three years, the AI power hierarchy was clear — Nvidia controlled compute, TSMC controlled fabrication, and the labs controlled the models. Power was an afterthought, something you solved by writing a cheque to your local utility.
That’s over. Power is now a strategic asset on par with GPU supply. The company that controls power generation and transmission to data centre clusters has leverage over every AI lab, every cloud provider, and every enterprise customer running inference at scale. NextEra just positioned itself as the gatekeeper.
Consider the numbers: Anthropic is paying SpaceX $1.25 billion per month for GPU compute. But the electricity to run those GPUs costs nearly as much — and unlike GPUs, you can’t shop for electricity on a global market. It comes from the local grid, or it doesn’t come at all. When a single company controls the grid that powers the majority of U.S. data centres, it wields a form of leverage that even Nvidia doesn’t have.
The Verdict: The AI Boom’s Infrastructure Layer Just Got Its First Monopolist
Everyone is watching AI labs race to build smarter models. The smarter play is watching who’s building the infrastructure those models need to run. NextEra just made the most expensive bet in utility history that power — not intelligence — is the scarce resource of the AI era. And if the projections about AI electricity demand are even half right, the $67 billion price tag will look like a bargain within three years.
The AI supply chain has a new link at the bottom, and it’s made of copper wire and concrete, not silicon. Every investor fixated on model benchmarks and API pricing should spend an afternoon reading NextEra’s merger filing instead. It tells you more about where AI is actually going than any earnings call from a lab that still can’t reliably keep its GPUs powered through a Virginia summer.
