NextEra's Dominion Acquisition: A $6.7 Billion Bet on AI's Insatiable Power Appetite

The True Cost of AI Compute: Power, Not Petabytes

NextEra Energy’s $67 billion pursuit of Dominion Energy isn’t a play for natural gas futures; it’s a calculated bid for the planet’s most voracious energy consumer: AI. While the press release trumpets scale and synergy, engineers tasked with deploying the next multi-megawatt training cluster face a stark reality: the grid itself is the fundamental bottleneck. The combined entity might promise more gigawatts, but the question remains: can the underlying infrastructure—transmission, interconnection, and supply chains—actually deliver them at the speed and reliability AI demands?

Scale is Necessary, But Insufficient for AI’s Thirst

The headline figures are staggering. NextEra, already a renewable energy titan, sees Dominion’s extensive transmission network and critical presence in Northern Virginia’s “Data Center Alley” as the missing piece for serving AI’s exponentially growing power appetite. CEO John Ketchum’s assertion that “scale matters more than ever” rings true when considering the sheer magnitude of demand. Dominion’s Virginia territory already saw data centers consume 24% of its electricity in 2023, and the utility is fielding connection requests for 70 GW, a figure nearly triple its historical peak system load of 24.7 GW. NextEra aims to develop between 15 GW and 30 GW of new generation capacity specifically for data centers by 2035, a build-out that necessitates integrating renewables, natural gas, and potentially nuclear sources. Dominion’s existing $65 billion five-year capital investment plan, with a $17 billion slice dedicated to data center support and grid upgrades, underscores the commitment already underway.

However, this pursuit of scale glosses over the systemic friction inherent in grid modernization. The merger, subject to regulatory approvals from three state and two federal commissions with an estimated 12-18 month close, doesn’t magically untangle existing transmission constraints. Dominion itself has previously flagged severe transmission limitations in eastern Loudoun County, Virginia, a core AI hub, demonstrating that generation capacity alone is not the answer. Nationally, the story is even more dire: only 888 miles of high-voltage transmission were completed in 2024 against an estimated need of 5,000 miles. This structural deficit means that even if NextEra brings terawatts of new generation online, the physical pathways to deliver that power to distributed AI compute hubs remain a critical choke point.

The Interconnection Queue: A Time Sink Measured in Decades, Not Months

The sheer volume of planned AI infrastructure clashes violently with the reality of the U.S. grid’s interconnection queue. Over 2,060 GW of projects are awaiting connection by the end of 2025. This gargantuan backlog is not merely a bureaucratic hurdle; it’s a profound indicator of the grid’s inability to absorb new, large-scale loads. Historical data reveals a grim truth: only 13% of projects entering these queues between 2000 and 2019 reached commercial operation by 2024. For data center operators, this translates into a multi-year delay that directly impacts their ability to deploy compute resources and capitalize on AI market opportunities. This is not a problem solved by utility mergers; it’s a symptom of decades of underinvestment in transmission infrastructure, exacerbated by the sudden, immense demand from AI workloads.

Furthermore, the “time-to-power” discrepancy—where utility estimates for delivering power to new data centers are frequently 1.5-2 years longer than hyperscalers anticipate—highlights a fundamental misalignment in planning and expectation. This gap forces AI infrastructure planners into difficult architectural decisions, often leading to the consideration of 100% onsite generation solutions, such as natural gas microgrids, as a pragmatic approach to bypass utility-side bottlenecks. This trend, driven by the pervasive grid and interconnection challenges, represents a potential fragmentation of the centralized power model, introducing its own set of complexities around fuel sourcing, maintenance, and load balancing.

Under the Hood: Supply Chain Shockwaves for the AI Power Grid

Beyond transmission and interconnection, the physical components required to build out new generation and distribution infrastructure are also facing unprecedented strain. Critical equipment like power and distribution transformers, essential for managing and stepping down electricity to usable levels, are experiencing significant supply shortfalls. Projections indicate deficits of 30% and 10% for these transformers in 2025, respectively. Lead times for large transformers have ballooned to over 100 weeks. This scarcity exists independently of any utility merger, meaning that even with increased demand and capital investment, the raw materials and manufacturing capacity to deliver the necessary electrical hardware are not scaling commensurately.

This supply chain crunch has direct implications for the cost and timeline of AI infrastructure deployment. The escalating costs associated with grid upgrades and new generation capacity are already manifesting. The PJM capacity market, which ensures sufficient generation to meet peak demand, saw clearing prices for 2026-2027 increase tenfold, a surge directly linked to the projected power needs of AI data centers. While NextEra has pledged $2.25 billion in bill credits for Dominion customers, these are short-term palliative measures. The underlying economic reality is that the immense, concentrated power demand of AI is driving up the cost of electricity for all grid consumers, potentially creating new financial risks for hyperscalers and their investors, as demonstrated by the financing challenges faced by entities like Nscale when building out their data center capacity. The energy bill is AI’s next major bottleneck, far more impactful than raw compute power or model size, as we explored in The Power Bill is AI’s Next Big Bottleneck.

Bonus Perspective: The Paradox of Centralized Power in a Distributed AI Future

The NextEra-Dominion merger, while aiming for centralized efficiency, arrives at a moment when the fundamental architecture of AI deployment is shifting. The persistent grid limitations are increasingly pushing large AI operators toward distributed, self-contained power solutions. This is not merely a workaround; it’s a strategic pivot to mitigate risk associated with centralized utility dependency.

Consider the implications: If AI workloads increasingly rely on localized generation (e.g., dedicated natural gas plants or even on-site renewable microgrids coupled with battery storage), the envisioned scale of grid upgrades by utilities like the combined NextEra-Dominion might face a demand shortfall. The projected 9-17% of U.S. data center electricity consumption by 2030, while substantial, carries a wide range of uncertainty due to these architectural shifts. This creates a genuine risk of stranded utility investments—billions spent on transmission and generation capacity that may not be fully utilized by the very AI boom it was meant to fuel. This isn’t about whether AI needs more power, but how and where that power will be sourced, and whether current infrastructure investments align with the emergent architectural realities of AI compute deployment. The challenges seen in Microsoft’s geothermal AI data center proposal in Kenya, facing power hurdles, serve as a microcosm of the complex interplay between AI demand and localized energy solutions.

An Opinionated Verdict: The Grid Needs an API, Not Just a Merger

NextEra’s acquisition of Dominion is a significant move, demonstrating a clear understanding of AI’s massive energy requirements. However, for engineers on the ground, this merger is a symptom, not a cure, for the AI power crisis. The true challenge lies not in combining existing utility assets, but in fundamentally re-architecting how power is provisioned and delivered to the edge.

The existing grid operates on a 20th-century model ill-suited for the 21st-century demands of AI. What’s needed is not just more kilowatt-hours, but a more dynamic, responsive, and transparent power infrastructure. This implies a need for better grid intelligence—an “API” for power that allows AI infrastructure planners to accurately forecast availability, negotiate delivery timelines with greater certainty, and integrate diverse energy sources more effectively. Until such systemic improvements are made, even a $67 billion merger will be navigating a landscape where the fundamental resource—reliable, scalable electricity—remains the most formidable bottleneck. The question for practitioners is no longer “can we get enough compute?”, but “can we get enough power to run it at the edge, on our timeline, and within our budget?” The answer, today, is far from guaranteed.