“Elon Musk was even more blunt: ‘AI will run out of electricity by next year.’” That warning, once dismissed as hyperbole, now reads like a blueprint for the next trillion-dollar market shift. The artificial intelligence boom is colliding head-on with a power infrastructure bottleneck, and the scale of the problem is staggering. Data centers-those sprawling, climate-controlled fortresses where AI models live and learn-already consume more electricity than some nations. Each hyperscale facility can draw as much power as a small city, and according to 451 Research, U.S. grid demand from data centers will rise 22% in 2025 and nearly triple by 2030, hitting 134.4 GW.
The surge is driven by accelerated servers optimized for AI workloads, which the International Energy Agency projects will grow electricity consumption 30% annually through 2030. This is not a gradual uptick-it’s a hyper-acceleration, outpacing overall grid growth fourfold. The U.S. Energy Information Administration forecasts record power consumption of 4,199 billion kWh in 2025, with AI and cryptocurrency data centers being a major contributor. The Department of Energy warns that data centers could account for 12% of US electricity use by 2028, up from 4% today, with AI workloads comprising as much as 40% of that total.
But against that backdrop, utilities are scrambling. Some AEP Ohio is one example are imposing tariffs requiring the largest data centers to pay for at least 85% of subscribed power, used or not. Others are exploring “behind-the-meter” solutions, which means colocating directly with generation assets as a way to avoid transmission bottlenecks altogether. Options run the gamut from natural gas plants to advanced nuclear reactors, but idle baseload is in exceptionally short supply. That scarcity is why nuclear power, with its “always-on” capability, is staging a comeback after three decades of stagnation.
The nuclear revival is not theoretical. President Trump’s AI Action Plan labeled AI data centers as critical defense facilities and called for ten large reactors under construction by 2030. DOE has already set aside sites at Idaho National Laboratory, Oak Ridge, Paducah, and Savannah River for AI-linked nuclear projects. Industry giants are moving quickly: Microsoft signed a 20yr PPA to restart Three Mile Island Unit 1 solely for data center operations, while AWS, Meta, and OpenAI are also investing in nuclear-powered campuses. Brookfield Asset Management, which is the majority owner of Westinghouse, is in talks to build sole-source nuclear plants for hyperscale AI facilities.
Small Modular Reactors are central to that buildout. Under DOE programs, including the Advanced Reactor Demonstration Program, billions are flowing to developers capable of delivering modular, scalable nuclear units with shorter construction timelines. China’s state-driven model-standardized designs, domestic manufacturing, and five-year build cycles-has given it a commanding lead, with 78 reactors operating or under construction and dozens more approved. The U.S., meanwhile, is racing to close the gap, leveraging both reanimated legacy plants and next-gen designs.
Natural gas still supplies 60% of new data centre demand, according to Morningstar, making the fuel the dominant near-term solution. LNG export infrastructure is thus another critical piece of the puzzle-particularly in light of Trump’s “America First” energy doctrine, which is forcing allies to buy U.S. LNG. EPC firms able to deliver both nuclear and LNG projects are uniquely positioned. After all, these companies don’t just build-they own and operate toll-booth assets that collect revenue on every megawatt or cubic foot flowing to market.
One such firm, broadly overlooked by mainstream AI investors, has no debt, is sitting on cash equal to nearly one-third of its market cap, and also holds a significant equity stake in a separate high-growth AI play. It is among a select few global players with both the scale and expertise to execute large EPC projects across nuclear, LNG, and industrial infrastructure. With AI-driven data center demand compelling a massive expansion in baseload power capacity, this firm thanks to its integrated role in engineering everything from LNG export terminals to constructing nuclear plants is poised to enjoy major beneficiary status from the AI megatrend.
Billionaires are already positioning. Bill Gates calls AI “the biggest technological advance in my lifetime,” Larry Ellison is embedding generative AI across Oracle’s cloud while buying up Nvidia chips by the billion, and Warren Buffett sees “hugely beneficial social impact” in the breakthrough. Hedge funds are whispering about this energy-infrastructure stock at closed-door summits, noting it trades at less than seven times earnings excluding cash and investments. In a market obsessed with flashy AI tickers, this is the quiet, underpriced operator building the power backbone for the AI age.
With AI’s electricity appetite accelerating toward a supply crunch, the collision of nuclear capacity expansion and LNG export growth with the march of hyperscale data centers is more than just an energy story; it is an investment thesis. The companies that own and control the infrastructure feeding AI’s voracious demand are the ones minting returns long after the chipmakers peak.
