The Unseen Power Source Driving AI’s Expansion

Artificial intelligence has an energy problem that most people overlook. A single hyperscale data center can consume as much electricity as a small city, yet the spotlight usually falls on the latest GPU clusters rather than the steady power supply keeping them running. That’s changing. Nuclear energy has become an unlikely linchpin for the AI boom, with tech giants signing multi-billion-dollar agreements to secure it. Utilities with existing atomic plants are suddenly in high demand.

NextEra Energy, the largest U.S. utility by market cap, is a case in point. The company operates a regulated electric business in Florida alongside a massive renewable energy portfolio. Its executives project energy demand will jump 60% between 2025 and 2045—a sharp acceleration from the 10% growth seen over the prior two decades. The stock offers a 2.8% dividend yield, with management targeting roughly 6% annual dividend growth. Yahoo Finance highlighted the company in late May as an AI infrastructure play hiding in plain sight.

But renewables alone can’t guarantee the always-on power AI workloads require. That’s where nuclear comes in. Talen Energy owns the Susquehanna nuclear plant in Pennsylvania, a 2.2-gigawatt facility now feeding a dedicated data center campus. Amazon signed a deal for up to 1.92 gigawatts of carbon-free power through 2042, a 17-year agreement that moves Talen from volatile merchant markets to stable, long-term revenue. Seeking Alpha rated the stock a Strong Buy in April, citing forward revenue growth near 30% and potential EBITDA expansion of more than 43%. The company’s 2026 targets call for EBITDA between $1.75 billion and $2.05 billion, with free cash flow reaching $980 million to $1.18 billion.

Power availability has become the binding constraint on AI expansion. Chip shortages and land access once set the pace; now the first question in site selection is where to get firm power fast. Nuclear has reentered the conversation for large campuses needing durable, clean capacity. Jensen Huang of Nvidia noted at a recent forum that while the U.S. leads in models and chips, other nations are quicker to spin up new generation. Microgrids offer one distributed solution, as reported by POWER Magazine in early May.

Public sentiment adds an unexpected twist. Americans dislike new data centers more than nuclear plants. A Gallup poll found 71% of respondents somewhat or strongly oppose local AI data center construction, while only 53% oppose a nuclear facility nearby. Seventy percent worry about environmental effects from the centers, which can consume five million gallons of water daily—equivalent to a town of 10,000 to 50,000 people. Forbes detailed the survey in mid-May, noting that the backlash centers on resource strain and quality of life. Nuclear, surprisingly, polls better.

Tech companies are acting on this reality. Meta announced landmark nuclear agreements in January 2026, unlocking up to 6.6 gigawatts. The pacts include funding for TerraPower reactors and a 1.2-gigawatt campus with Oklo in Ohio, featuring 16 Aurora Powerhouse units of 75 megawatts each across 206 acres. Meta is offering prepayments to speed fuel supply and construction, with first power possible by 2030. The company also signed 20-year extensions with Vistra to keep existing plants running in the PJM grid.

Google struck the first corporate small modular reactor (SMR) power purchase agreement with Kairos Power in 2025. Amazon bought a data center campus next to Talen’s Susquehanna plant for direct power access. Microsoft reached deals to restart portions of Three Mile Island. These moves signal a clear pivot: hyperscalers no longer trust the strained grid alone. They want dedicated, carbon-free, 24/7 power. iRecruit reported in late May that one modern facility now requires 80 megawatts, more than double the old standard of 32 megawatts.

NANO Nuclear Energy signed a memorandum of understanding with Supermicro in early May to explore pairing microreactors with AI servers and data center platforms. The partnership targets clean, scalable solutions for the expanding AI economy. Such collaborations point to deeper vertical integration, where power and compute converge.

Global spending on new AI data centers could exceed $7 trillion by 2030, yet opposition inside the U.S. grows. CNBC noted in early May that communities resist these massive builds, with some projects now receiving permitting treatment akin to military priorities. The energy shortfall persists. The International Energy Agency once projected data centers could consume 1,000 terawatt-hours annually by certain benchmarks, rivaling entire national grids. Natural gas has filled much of the gap so far—40% of data center electricity came from it in recent years—while nuclear supplied 20%. The mix must change to meet carbon goals and reliability demands.

Small modular reactors promise factory-built speed, shipping like modules. Regulators still pose hurdles, construction timelines stretch, and fuel supply chains need work. But momentum builds. Meta’s capital commitments give developers certainty. Oklo, TerraPower, and others gain footing. Retired Navy reactors even draw interest as potential stopgaps. The Atlantic painted a stark picture: xAI’s Colossus cluster, if run full tilt, would match the annual electricity of 200,000 American homes. Elon Musk’s team deployed dozens of natural-gas turbines to accelerate launch, sparking smog concerns. Such interim steps highlight the urgency.

Wall Street senses the shift. Analyses suggest AI data centers could trigger one of the largest power demand expansions in decades. Investors eye utilities with nuclear fleets or nuclear-adjacent assets. Talen proves the model works: co-location cuts transmission losses and bypasses interconnection queues that stretch years. Behind-the-meter setups evolve into formal contracts. AWS and Talen moved to a 17-year power purchase agreement, replacing merchant risk with stability.

Challenges remain: nuclear waste, public perception in some pockets, high upfront capital. Yet the alternative looks worse. Intermittent renewables require massive storage. Gas plants face fuel price swings and emissions rules. The grid itself strains under peak loads. New firm generation must arrive.

Tech firms are voting with their balance sheets. Over $50 billion collectively committed by Meta, Microsoft, Amazon, and Google to nuclear projects as of early 2026, according to industry trackers. Data center power demand could surge 220% globally by 2030, per Goldman Sachs commentary. SMR developers like Oklo and NuScale see renewed interest despite share volatility.

The AI boom no longer hides its appetite. It consumes. And the power sources best suited to feed it without interruption have gained fresh respect. Nuclear plants once seemed relics of another era. Now they anchor the future compute infrastructure. Utilities that own them or can expand around them hold cards few others do. The quiet play grows louder with every terawatt-hour demanded.