The biggest corporate merger of 2026 is not in tech. It is in electricity. And the reason is AI.
On Monday, May 18, NextEra Energy announced it would acquire Dominion Energy in an all-stock transaction valued at roughly $67 billion. By Wednesday morning, analysts were already calling it the largest utility acquisition in US history. I have covered AI infrastructure deals for two years now, and I can tell you that this one is different from anything that has come before. Not because of the dollar amount, although $67 billion is a staggering number. Because of what it reveals about where the AI industry's real bottleneck has moved.
It is no longer chips. It is no longer talent. It is no longer even capital. The bottleneck is electricity. And this merger is what it looks like when the biggest companies on Earth decide to fix that problem by fundamentally restructuring America's power grid.
NextEra CEO John Ketchum did not pretend otherwise on the announcement call. He told investors the combined company exists to "satisfy enormous and fast-growing demand for electricity" driven by hyperscalers, increased electrification, and population growth. Strip away the corporate language and here is what he said: the deal got done because Microsoft, Google, Amazon, and Meta need someone big enough to build the power infrastructure their data centers will demand through 2032 and beyond. A $67 billion all-stock merger is what that someone looks like.
Virginia is not just important. Virginia is the whole point.
To understand why this merger happened, you need to understand one geographical fact that most people outside the data center industry do not know. Northern Virginia, specifically Loudoun County and the corridor along Route 28, is the densest concentration of data centers anywhere on the planet. Roughly 70% of global internet traffic at some point touches a server in Northern Virginia. Every major hyperscaler runs production AI workloads there. Every major sovereign AI program with US ties stages capacity there.
Dominion Energy owns the grid that powers all of it. That is its crown jewel. Not its generation fleet, not its transmission lines, not its customer base. Its crown jewel is that it controls the electrical infrastructure underneath the most important cluster of AI computing power in the world.
The numbers explain why the deal exists. Northern Virginia's data center power demand was 16.6 gigawatts in 2024. By 2030, projections put it at 33 gigawatts or more. To put that in perspective, 16.6 gigawatts is already larger than the entire peak electricity demand of countries like Ireland or Denmark. By 2030, Northern Virginia alone is projected to need roughly the equivalent of Greece's entire current electricity demand. One corridor in one US state, mostly serving AI inference and training workloads, consuming as much power as an entire European country.
Dominion could not finance that buildout alone. Its balance sheet was not big enough, its capital access was not fast enough, and the regulatory complexity of expanding at that rate was beyond what a single mid-sized utility could manage. NextEra could not access the demand without owning the grid. It had the construction expertise, the renewable energy portfolio, and the balance sheet, but it did not have the thing that mattered most: the physical connection to the customers who needed the power.
The merger is the obvious answer to a problem both companies had been working around for two years. Together, they create the only utility on Earth with the scale to match what the hyperscalers need.
130 gigawatts: a number that should stop you in your tracks
The single most important number in the announcement was buried in the supplementary deck, not featured in the press release. A combined construction backlog of 130 gigawatts.
I need to give you context for that number because it is easy to see "130 GW" and not grasp what it means. The total installed electricity generation capacity of the United Kingdom is roughly 75 gigawatts. The combined NextEra-Dominion construction pipeline, meaning power projects already designed, permitted, or under contract, is 1.7 times the size of the UK's entire grid.
That backlog breaks down roughly like this. About 55 gigawatts of renewables: solar, wind, and battery storage. NextEra is already the largest player in renewables globally, and Dominion's offshore wind portfolio plus East Coast solar adds material capacity. About 40 gigawatts of natural gas generation. This is the part that nobody in the clean energy conversation wants to talk about, but the reality is that renewables alone cannot meet 24/7 data center demand. Natural gas plants are getting built at the fastest rate since 2008 because AI data centers need power at 3 AM on a windless night in January, and there is no solar panel or wind turbine that delivers that. About 15 gigawatts of nuclear, including Dominion's existing four nuclear units in Virginia and NextEra's pipeline of partnerships with NuScale and TerraPower for small modular reactors. And about 20 gigawatts equivalent of transmission and grid infrastructure, because the grid itself is the constraint. Building more generation capacity does nothing if you cannot move the electrons to where they are needed.
The combined company will rank number one globally in renewables and battery storage, number one in US gas generation, number one in US total generation, number one in generation buildout, and number two in US nuclear. There has never been a single utility with that portfolio anywhere in the world. This is not consolidation for efficiency. This is the creation of an entirely new kind of energy company, one built specifically to serve the AI economy.
The question nobody at NextEra wants you to ask
Here is where the deal gets politically uncomfortable, and where I think the most important story is hiding.
A $67 billion merger does not happen in a vacuum. NextEra-Dominion has to recover those costs somewhere, and the obvious somewhere is the rate base, which is the regulated framework that lets utilities pass infrastructure costs through to electricity customers. Translation: Virginia ratepayers are going to pay for this. Unless someone stops them.
Virginia ratepayers are already absorbing power bill increases tied to data center buildout. Average residential electricity rates in Dominion's service territory have risen faster than the national average for three consecutive years. Consumer watchdog groups, including Common Dreams, called the merger announcement "absurd" on Monday, not because the strategy is wrong but because the regulatory framework for splitting costs between hyperscalers and households is still being written.
This is the fight I covered in detail in my article on data center ratepayer protection. The short version is that state utility commissions across Virginia, Texas, and Ohio are starting to force hyperscalers into special data center tariff structures so that ordinary residential customers do not subsidize Microsoft and Meta's capacity needs. The merger raises the stakes on those negotiations by an order of magnitude because now the utility on the other side of the table is not a mid-sized regional player. It is the largest utility in the country with the kind of lobbying budget that reshapes regulation.
NextEra's bet is that by the time the deal closes in mid-to-late 2027, the tariff structures will favor utilities over consumer protections. Given the political weight of AI investment in 2026, that bet is probably correct. But "probably correct" is doing a lot of work in a $67 billion transaction, and the regulatory fight between now and the close is going to be brutal.
What this means if you work in AI
Let me be concrete about why this merger matters to people who do not care about utility stocks.
If you run procurement at Microsoft, Google, AWS, Meta, or Oracle, this deal is unambiguously good news. The supply side of the AI power equation just got 50% more concentrated, and the consolidated counterparty has the balance sheet to actually deliver capacity at the timeline you need. You have been fighting for megawatts for two years. You just got a vendor that can deliver gigawatts.
If you are an AI-native company that depends on cloud capacity, like OpenAI, Anthropic, xAI, Mistral, or Cohere, the picture is more nuanced. Hyperscalers will continue passing through power costs to you in the form of higher GPU-hour pricing. The AI cloud capacity crunch I covered in my piece on enterprise AI infrastructure does not get fixed by this merger. It gets fixed by the 130 gigawatt backlog actually getting built over 5 to 7 years. Until then, every quarterly capacity allocation conversation gets harder, not easier.
If you are buying NVIDIA GPU systems, the question is no longer "can I get the chips." The question is "can I get the megawatts to run the chips." A B200 NVL72 rack pulls roughly 132 kilowatts at full utilization. A one-gigawatt deployment needs 7,500 of those racks plus all the cooling, networking, and redundancy. The chip side of the equation scales faster than the power side, and this deal is what the power side scaling up actually looks like when it hits the public markets.
And if you are a consumer who uses ChatGPT, Claude, Gemini, or any AI product, understand that the electricity cost of running your queries is about to get baked into the infrastructure of the largest utility merger in history. You may not see it on your electric bill directly. But you will see it in the subscription price of AI products, in the cost of cloud computing, and in the rates your employer pays for the AI tools your company uses. The cost of AI is fundamentally an energy cost, and this merger is repricing that energy.
The startup landscape just shifted
A wave of well-funded startups emerged in 2024 and 2025 trying to solve the AI power problem outside the regulated utility framework. Companies like Crusoe, Standard Industries, Verrus, and Stack Infrastructure raised billions to build modular data centers near stranded generation assets, gas wells, and nuclear sites. The thesis was that the regulated utilities were too slow and too bureaucratic to serve AI demand, so there was a window for nimble startups to fill the gap.
That thesis just got complicated in two ways.
First, the exit window opened. With the largest utility in the world actively buying capacity-adjacent infrastructure, every neocloud and modular data center startup just got a credible strategic acquirer. Expect several billion-dollar acquisition announcements in the next 18 months from companies that suddenly look attractive to NextEra-Dominion's development team.
Second, the competitive moat narrowed. The exact playbook these startups were running, "let us go around the slow utilities," gets much harder when one utility owns the rate base across four states and 10 million customers. Permitting, interconnection queue priority, and power purchase agreement financing all favor the combined entity. The startups that survive will be the ones with truly differentiated approaches: stranded methane conversion, on-site generation, specialized immersion cooling. Generic "data center developer" pitches are going to face a much rougher fundraising environment in 2027.
The part most US energy reporting is missing
There is a geopolitical layer to this deal that I think is going to matter more than the domestic energy angle over the next decade.
The UK, Saudi Arabia, UAE, France, Japan, and Korea are all running sovereign AI infrastructure programs where the bottleneck is not chips. It is grid capacity. None of those countries has a utility merger of this scale on the table. Most are still operating with fragmented regional utilities that cannot move at hyperscaler speed. Europe's regulatory environment for utility consolidation makes a deal like this essentially impossible. The Middle East has the capital but not the construction pipeline.
The NextEra-Dominion deal effectively gives US-based AI workloads a structural cost-of-capacity advantage over Europe and the Middle East for the next decade. If you are an AI startup choosing where to deploy your training cluster, "Northern Virginia under the new combined utility" just became the most predictable power supply you can get anywhere in the world.
That is a moat that compounds, and it is one of the under-discussed reasons US AI valuations keep diverging from the rest of the global AI sector. It is not just that the US has more AI companies. It is that the US is building the physical infrastructure to sustain those companies at a scale and speed that no other country can match.
Three things to watch
The deal is announced but not closed. Three things determine whether it delivers on the promise.
First, the regulatory approval timeline. FERC and state utility commissions will scrutinize a deal of this size in a politically charged industry. Watch the Virginia State Corporation Commission, the North Carolina Utilities Commission, and FERC for early signals on whether the mid-to-late 2027 close holds. Any delay past 2027 changes the economics significantly because the hyperscaler demand is not waiting.
Second, the first major hyperscaler power purchase agreement post-merger. Within 6 to 9 months, expect a multi-gigawatt PPA announcement involving Microsoft, Google, or AWS. The pricing of that contract becomes the benchmark for every subsequent data center power deal in the US. It will tell you how much of the merger premium gets passed to AI companies and, through them, to AI consumers.
Third, whether the 130 gigawatt backlog actually accelerates. Combining two companies often slows execution before it speeds up. Integration takes time, cultures clash, decision-making stalls. If the new entity is still shipping 6 gigawatts per year by the third quarter of 2027 instead of the implied 18 to 25 gigawatts per year, the bull case takes a serious hit.
What I actually think about this deal
NextEra paid $67 billion to buy access to the largest concentration of AI compute demand on Earth. Dominion shareholders got a 25.5% stake in what will be the most strategically positioned utility of the AI era. The deal makes sense for both sides. The construction backlog is real. The demand is real. The competitive position is extraordinary.
The losers, if there are any, will be Virginia ratepayers and the consumer advocates fighting to keep household power bills separate from the cost of training Claude and ChatGPT. That fight matters. It matters for fairness. It matters for the political sustainability of the AI buildout. And it matters because if residential ratepayers end up subsidizing hyperscaler data centers, the backlash could slow down the very infrastructure expansion the deal was designed to accelerate.
This deal is a leading indicator that AI infrastructure has become too important and too capital-intensive to leave to fragmented regional utilities. Expect at least two more mega-mergers in US utilities within 24 months. The trillion-dollar AI race and the massive equity bets flowing into AI are rounding errors compared to what will actually be spent building the grid that runs the next decade of artificial intelligence.
The capital is moving. NextEra just fired the starting gun.
---
Keep Reading
- AI Startup Ideas for 2026 -- the opportunities created by this infrastructure buildout
- How AI Is Killing the Real Estate Industry -- another traditional industry being reshaped by AI economics
- Lovable vs Bolt vs Cursor vs Claude Code -- the AI developer tools that are driving demand for all this compute
- How Midjourney Hit $500M with Zero Funding -- a company that found a way to grow without raising billions
