Over the past three years, market discussions about AI have consistently revolved around models, chips, and computing power: in 2023, the worry was insufficient GPUs; in 2024, the focus shifted to data center construction speed, followed by advanced packaging, land, and infrastructure access. The technological narrative keeps advancing upstream, but a more fundamental issue is emerging: what determines the speed of AI expansion may no longer be just chips, but electricity. For a long time, electricity has been seen as a naturally existing, readily available basic resource, and the U.S. utility industry has thus been labeled with tags like low growth and stable dividends, rarely standing at the center of the tech revolution stage. But when the power consumption of a large AI data center approaches that of a city, when the speed of data center construction outpaces grid expansion, energy is stepping from the backstage to the forefront for the first time, becoming a key variable affecting the pace of the AI industry.
The approximately $66.8 billion acquisition of Dominion Energy by NextEra Energy appears in this context. On the surface, it's a record-breaking utility merger; looking deeper, it reflects how AI is restructuring the growth logic, capital flow, and competitive order of the energy industry. In the past, tech companies sought power resources; now, it's starting to shift towards power companies proactively expanding scale to meet the computing power demand of the coming decades. When the growth rate of computing power exceeds the speed of energy infrastructure construction, AI competition is no longer just about model capability and chip performance; it increasingly resembles a new infrastructure war centered on power generation, transmission, grid interconnection, and energy dispatch capabilities.
What $67 Billion Buys
NextEra Energy's all-stock acquisition of Dominion Energy for about $66.8 billion, on the surface, is a major consolidation in the U.S. utility industry, but it truly points to a deeper change: AI infrastructure expansion is re-positioning power companies from "stable dividend assets" back towards "strategic growth assets."
Post-transaction, the combined company will become one of the world's largest regulated electric utilities, with an enterprise value of around $420 billion. More importantly, NextEra will enter the most critical data center hub in the U.S. via Dominion – Virginia, specifically Northern Virginia's Data Center Alley. This is one of the world's densest data center clusters and a core node for cloud computing and AI infrastructure. Dominion doesn't just sell electricity to ordinary residents and businesses; it holds large load demand from tech giants like Alphabet, Amazon, Microsoft, and Meta, with related signed data center capacity nearing 51GW.
This means NextEra isn't buying a traditional grid balance sheet, but a ticket to the epicenter of power demand in the AI era.
AI Rewrites the Power Growth Logic
For over two decades, U.S. electricity demand growth has been relatively flat, and the investment logic for utility companies has also been stable: regulated returns, regional monopolies, dividends, low volatility. But the explosion of AI data centers has broken this model.
The U.S. Energy Information Administration expects U.S. electricity consumption to rise from a record 4195 billion kilowatt-hours in 2025 to 4248 billion kWh in 2026 and 4379 billion kWh in 2027. Behind this growth, AI data centers, cryptocurrency mining, electrification, and industrial expansion are the main drivers.
The bigger issue is that AI loads are not ordinary commercial electricity in the traditional sense. The power demand of a large AI data center campus can reach hundreds of megawatts, even approaching 1GW, equivalent to the electricity consumption of a medium-sized city. In the past, utilities served residents, commercial buildings, and manufacturing plants; now, they must face city-scale, continuous, high-reliability power demands presented by cloud providers and AI companies.
The International Energy Agency expects global data center electricity consumption to grow from about 485 TWh to nearly 950 TWh by 2030, almost doubling. AI data centers will be the fastest-growing segment.
The problem isn't the increase in electricity demand, but the change in its structure. Past new demand came from population and industrial growth, a relatively smooth increase; now, new demand comes from mega data center clusters, a concentrated, sudden, and highly dense growth.
This change means the operational logic of the power industry formed over the past few decades is being rewritten.
What's Truly Scarce is Interconnection Rights
Over the past two years, when the market discussed AI infrastructure, the focus was mainly on GPUs, advanced manufacturing processes, HBM, servers, and data center construction. But the bottleneck is migrating further upstream: chips can be bought, data centers can be built; what's truly difficult is whether you can quickly interconnect with enough stable, cheap, and continuously available power.
Morgan Stanley Research predicts that by 2028, U.S. data center power demand could reach 74GW, but there could be a gap of about 49GW in available interconnection capacity.
This number reveals a key contradiction: the construction cycle for computing power is far faster than for energy infrastructure.
Data center projects typically land in 18 to 24 months, while power generation projects, transmission networks, substation expansions, and grid interconnections usually take years. Some transmission projects require even longer cycles.
Thus, an unprecedented phenomenon is emerging in the AI industry: it's not a lack of servers, but a lack of electricity.
In the past, tech companies' core competency was procuring GPUs; in the future, they may need to lock in long-term power purchase agreements, nuclear projects, natural gas assets, and grid interconnection qualifications in advance.
The center of gravity for resource scarcity is shifting.
Virginia: The AI Power Frontline
Dominion's greatest value lies not in asset scale, but in geographic location.
Northern Virginia has long been one of the densest regions for U.S. internet infrastructure and home to the world's largest data center cluster. It aggregates cloud services, fiber networks, internet exchange points, and numerous enterprise clients.
Once data centers form an agglomeration effect, they attract more customers to enter, as the advantages of low latency, network connectivity, and ecosystem synergy become stronger.
Dominion has long held the power supply rights in this region, thus naturally occupying a critical node for AI expansion.
NextEra already possessed the largest renewable energy development capability in the U.S. and regulated utility business in Florida, but if it wants to enter the region with the fastest-growing AI electricity demand, directly acquiring Dominion is clearly more efficient than rebuilding from scratch.
Therefore, what the transaction truly buys is the most scarce infrastructure gateway for the next decade.
Who controls the gateway, controls the increment.
The Logic of Economies of Scale is Being Rewritten
Bloomberg defines this transaction as the beginning of the super-utility M&A era driven by AI, behind which lies a change in the minimum efficient scale of the power industry.
In the past, utility scale advantages mainly came from regional monopolies and operational efficiency.
But AI has changed the rules.
First is the scale of capital expenditure. AI infrastructure requires simultaneous construction of natural gas, nuclear, energy storage, transmission, and renewable energy projects, which small power companies struggle to undertake.
Second is financing capability. Large companies find it easier to obtain low-cost funding.
Third is customer capability. Facing super-customers like Microsoft and Amazon requires sufficient size for long-term contract negotiations.
Fourth is regulatory coordination capability. Interstate transmission, electricity pricing mechanisms, residential subsidies, and project approvals all require more complex political and regulatory coordination.
NextEra's proposal of about $2.25 billion in customer bill credits is essentially preemptively reducing regulatory resistance.
Because the real difficulty isn't completing the acquisition, but gaining public acceptance.
The issues are becoming increasingly sensitive:
Who bears the cost if AI companies consume large amounts of electricity?
Will residents pay higher electricity bills?
These issues could become regulatory focal points in the future.
It's Not Just Power Companies Being Dragged In
AI's electricity consumption growth will ripple along the energy supply chain in multiple directions.
First is natural gas.
Data centers need 24/7 stable power supply. Although renewable energy costs are falling, their variability means they still require peaking capacity. Natural gas, due to its fast construction speed and strong dispatchability, is likely to become the main supplement in the medium term.
Second is nuclear power.
Large tech companies are increasingly actively engaging in nuclear power collaborations, as AI needs stable, low-carbon, and continuous baseload power.
Third is energy storage.
Energy storage not only serves peak shaving and valley filling functions but may also help data centers reduce reliance on backup diesel systems.
Fourth is transmission equipment.
Often, the problem isn't a lack of generation, but an inability to deliver the electricity.
Transformers, high-voltage lines, and substations could become the most overlooked bottlenecks in the coming years.
Therefore, the AI power arms race won't only benefit a single industry; it will revalue the entire energy infrastructure chain.
Energy Redefines Competition
Recently, a Ryanair executive's warning about a jet fuel crisis seems unrelated to AI, but the underlying logic is the same.
Energy supply is redefining industry competitiveness.
Airlines lock in fuel in advance.
Tech companies lock in electricity in advance.
Manufacturers are starting to compete for grid capacity.
Future competition is no longer just about capital; it increasingly resembles energy competition.
In the past, the core of corporate competition was technology, scale, and channels.
In the future, a new key metric may be added:
Energy securing capability.
The strong secure resources early.
The weak are exposed to price volatility and supply tightness risks.
Energy is returning to the center of industrial competition.
The Real Variable is Regulation
The NextEra-Dominion deal is expected to have an approval cycle of 12 to 18 months, involving state public utility commissions, federal approvals, and antitrust reviews.
Regulators will focus on several issues:
Will the merger raise residential electricity rates?
Will data centers crowd out resources for residents?
Who bears the cost of grid upgrades?
Will clean energy goals be affected?
Will competition diminish after scale increases?
The greatest complexity of the AI power cycle lies in this: the market wants faster expansion, while utilities are naturally constrained by regulation.
Tech companies want fast grid interconnection.
Residents don't want to pay higher bills.
State governments want to attract AI investment.
Communities worry about water resources, land, and noise.
All forces will simultaneously pull at the industry.
Therefore, AI infrastructure expansion is not just a capital problem, but a regulatory and social one.
What the Market Really Needs to Reassess
If this transaction is viewed merely as a traditional M&A deal, then the analytical framework is nothing more than valuation, debt, and synergies.
But if it signifies the opening of the AI energy era, then the market needs to rethink:
First, whether AI electricity demand can be consistently realized.
Second, which companies possess the most scarce interconnection nodes.
Third, whether new investments can translate into regulated returns.
Fourth, whether the energy structure meets tech companies' needs.
Fifth, whether social resistance will continue to increase.
In the future, the most valuable assets might not necessarily be generation assets, but integrated infrastructure platforms with interconnection capability, regulatory resources, and key nodes.
Conclusion
Over the past decade, the AI story has always unfolded in the world of chips, cloud computing, and software.
But the NextEra-Dominion deal shows that the next phase of competition is shifting to another level: power plants, grids, transmission lines, regulatory commissions, and infrastructure construction.
The bottleneck for AI is shifting from chips to energy, from servers to the grid, from competition among tech companies to cross-industry resource competition.
Whoever has stable electricity, whoever controls grid interconnection capability, controls the future pace of computing power expansion.
Therefore, the truly important aspect of this deal is not its $66.8 billion scale, but the signal it releases: the AI arms race of the next decade, superficially a model competition, is increasingly becoming an infrastructure war centered on energy, land, grids, and regulatory capability at its core.
