AI's Next Bottleneck After Memory: Energy
The six nuclear stocks positioned across power generation, uranium supply, and reactor components that I believe will absolutely print shareholders money in the years to come
The 2026 AI trade ran through memory. Nvidia’s GPUs sat on order books measured in years, and the real constraint was no longer the chip - It was high bandwidth memory. SanDisk and Micron turned into some of the hottest AI names overnight. Without their memory stacks, none of the Nvidia compute reached the desired effect of the end user goal. That bottleneck will slowly start to clear and a new one is rapidly tightening underneath it, through the electric grid.
Goldman Sachs Research puts US data center power demand at 31 gigawatts in 2025, climbing to 41 gigawatts this year, and doubling again to 66 gigawatts by 2027. S&P Global expects capacity to cross 100 gigawatts by 2028. Gartner projects global data center electricity consumption growing from 448 terawatt-hours in 2025 to 980 terawatt-hours by 2030, with AI-optimized servers alone jumping from 93 to 432 terawatt-hours over that stretch. A single next-generation AI campus can draw 1 to 5 gigawatts. To put it in simpler terms: that’s the output of an entire power plant for a single customer. Solar and wind can’t cover that load alone. Nuclear plants run above a 92.5% capacity factor, producing close to their maximum output almost around the clock. Solar averages 23 to 25%. Wind averages 34 to 35%, per EIA data. One gigawatt of solar capacity delivers about 250 megawatts of usable output. One gigawatt of nuclear delivers 925 megawatts, day and night, regardless of weather. For a data center serving live inference requests across time zones, that reliability gap is the key focus and metric for success.
Big Tech has already voted with its balance sheet. As of May 2026, 13 announced projects have committed more than 9.8 gigawatts of nuclear capacity to AI infrastructure. Microsoft is paying $16 billion over 20 years to restart Three Mile Island, now rebranded the Crane Clean Energy Center, and will take 100% of its output starting in 2027. Meta pledged up to 6.6 gigawatts combined across deals with Vistra, Oklo, and TerraPower, the largest hyperscaler nuclear commitment on record, on top of a separate agreement with Constellation for its Illinois data centers. Amazon expanded its Talen Energy agreement in June to secure 1,920 megawatts from the Susquehanna plant through 2042, on top of an X-energy small modular reactor partnership targeting up to 5 gigawatts. Google signed with Kairos Power for 500 megawatts.
Nvidia CEO Jensen Huang described the next generation of data centers as “gigawatt factories” on the Joe Rogan podcast in December, and predicted small nuclear reactors would power data centers within six to seven years. The IAEA held its first symposium on AI and nuclear energy this year, with Director General Rafael Grossi framing the pairing as a structural alliance he’s calling Atoms for Algorithms. Washington is leaning the same direction. The Department of Energy already put $400 million each behind the Tennessee Valley Authority and Holtec to accelerate small modular reactor deployment, and in June issued a conditional commitment for $17.5 billion in loans to build ten new Westinghouse AP1000 reactors. Four executive orders signed in 2025 target growth from about 100 gigawatts of US nuclear capacity today to 400 gigawatts by 2050. A nuclear-heavy National Defense Authorization Act added momentum on top of that.
None of this resolves into a single trade. Nuclear splits into three layers with different risk profiles. Operators run plants today and sign the multi-decade power purchase agreements hyperscalers are lining up for, generating real cash flow right now. The fuel chain supplies the uranium those plants burn, in a market where 2025 global production of about 173 million pounds ran well short of primary demand near 204 million pounds, a structural deficit that firms up as more reactors restart and extend their licenses. The future-builders are the small modular reactor developers building the next generation of plants, betting on faster deployment closer to the data centers themselves. Most of these companies are pre-revenue, funded on backlog and promise rather than output. That third layer carries the sector’s real risk. Nuclear construction has run late and over budget for as long as the industry has existed, and even with regulatory tailwinds, Wood Mackenzie still expects US nuclear generation to stay close to flat until 2035 before accelerating. Most of the hyperscaler SMR commitments deliver power in the 2030 to 2035 window, a long runway for companies still burning cash today. Uranium prices carry their own volatility. Spot uranium touched $106.75 a pound in early 2024, fell into the low $70s through most of 2025, and has recovered to $84 to $86 as of June, with Citi projecting a move toward $100 to $125 by year end. A 30% swing in the spot price can move uranium miner equities by far more.
I’m now going to run you through six of my favorite ticker names across all three layers of this stack: two operators already cashing hyperscaler checks, one components manufacturer and one fuel supplier feeding the buildout, one pure-play uranium producer, and one small modular reactor developer carrying the sector’s highest risk and highest ceiling. These are all share plays for me due to the long horizon but I believe they will all have their day. The full breakdown, current prices, analyst targets, and the specific “why” behind each pick, is for subscribers. Lets get into it.
Subscribe to Hidden Gems Research to see the full breakdown on all six names. We caught $ONDS before it ran +2600%, $LPTH before it ran +440%, and $UMAC before it ran +340%. Subscribe so you don't miss the next one.
