The Nuclear Stack: Why AI's Hunger for Power Is Rewriting Energy Investing

Sector Thesis, Multi-Ticker Pillar Scores (out of 10): Value 7.20 Growth 7.80 Quality 8.50 Momentum 8.90 Safety 6.80 Composite 8.70 Sector composite weighted toward downstream fleet operators (CEG, VST, TLN) and upstream fuel-services (CCJ, LEU). Midstream SMR developers dilute quality and safety pillars due to pre-commercial profiles. The Grid Was Never Built for This Somewhere in central Texas, a 200-megawatt data center hums at full load around the clock. It draws as much electricity as a small city, and it is one of dozens either operating or under construction across ERCOT territory alone. Across the country in PJM's mid-Atlantic corridor, server farms stretch across former farmland while thousands of GPUs train the models that will define the next decade of technology. In January 2026, the U.S. Energy Information Administration forecast the strongest four-year stretch of electricity demand growth since 2000, driven explicitly by "large computing facilities, including data centers." Not "gradually rising." Not "modest uptick." The most aggressive demand trajectory in a quarter century. Here's what makes the nuclear thesis click: this demand isn't spread evenly across the grid. It's concentrating in specific corridors where data center clusters are creating localized capacity crunches that generic renewable additions alone can't solve. The International Energy Agency adds an important nuance: globally, data centers may account for less than 10% of electricity demand growth, but their geographic concentration makes grid integration far harder than that headline number suggests. Think of it like plumbing. You can have plenty of water in the system, but if everyone on one block turns on their taps at once, the pressure drops to zero on that block. That's what's happening in ERCOT and PJM. Power procurement is shifting from "buy whatever the grid clears at" to "contract and co-locate with generation." PJM's February 2026 proposal set even includes a "bring-your-own-generation" framework for large loads, an institutional admission that the old model of "plug in and draw from the pool" is breaking under AI's appetite. That plumbing problem is where the nuclear investment thesis lives. And it plays out across three distinct segments of the supply chain, each with different risk profiles, different timelines to cash flow, and very different implications for your portfolio. Upstream: Where the Periodic Table Meets Geopolitics Every nuclear reactor begins with uranium, and uranium's supply chain is one of the most concentrated commodity chains on Earth. In 2024, over 60% of global output came from just ten mines across four countries. That concentration alone creates periodic supply squeezes. But the current cycle is compounded by a decade of underinvestment. After Fukushima in 2011, uranium prices cratered, dropping more than 70% to roughly $20 per pound by 2016. Producers responded rationally: they suspended operations, drew down inventories, and stopped investing in exploration. The World Nuclear Association documents that the exploration-to-production lag pushes cash flows "at least 10-15 years into the future." You can't turn mines on and off like a light switch. But here's the counter-intuitive insight: the real upstream chokepoint isn't mining. It's the fuel chain's hidden bottlenecks, conversion and enrichment. Conversion, transforming yellowcake into uranium hexafluoride, runs through only five large-scale facilities worldwide. A 2026 Stanford-led roundtable flagged conversion as a major bottleneck, noting that contracting reluctance has blocked capacity expansion even when the macro need is obvious. Enrichment is even more constrained, particularly for HALEU (high-assay low-enriched uranium, enriched between 5% and 20% U-235), the fuel most advanced reactor designs require. Scale HALEU infrastructure is concentrated in Russia and China. The U.S. import ban on certain Russian uranium products too