Nuclear Energy Stocks and the AI Data Center Power Boom

The Power Problem: Why AI Needs Nuclear

AI has a power problem that most investors still underestimate. A single Nvidia GB200 NVL72 rack consumes approximately 120 kW of power. A large-scale AI training cluster with 100,000 GPUs draws 150–200 MW continuously — enough electricity to power 150,000 homes. And the hyperscalers are not building one cluster. They are building dozens, simultaneously, across multiple continents.

The numbers are stark. Goldman Sachs projects US data center power consumption will grow from 19 GW in 2024 to 35 GW by 2030 — an 84% increase in six years. The Electric Power Research Institute (EPRI) has an even more aggressive estimate of 40 GW by 2030. For context, 35 GW is roughly equivalent to the entire electricity generation capacity of the Netherlands.

Where does this power come from? Natural gas can scale but contradicts the carbon-neutral commitments that Microsoft, Google, Meta, and Amazon have all made. Solar and wind are cheap but intermittent — AI training clusters cannot pause when the wind stops blowing. Battery storage helps with intermittency but remains too expensive at the scale required. Geothermal is promising but geographically limited.

That leaves nuclear. It runs 24/7 with capacity factors above 92% (versus 25% for solar and 35% for wind). It is carbon-free. A single 1 GW nuclear plant produces more annual electricity than a solar farm covering 10,000 acres. And unlike new gas plants, nuclear does not expose hyperscalers to volatile commodity pricing or carbon tax risk.

This is not theory. The hyperscalers are voting with their checkbooks.

The Hyperscaler Nuclear Deals: Microsoft, Meta, and Google

Microsoft fired the starting gun in September 2024 with its 20-year power purchase agreement to restart Three Mile Island Unit 1. The deal was symbolically powerful — the name Three Mile Island carried decades of negative connotation from the 1979 Unit 2 accident — and commercially significant. Constellation Energy will invest $1.6 billion to bring the 837 MW reactor back online by 2028, with all output contracted to Microsoft. At estimated electricity prices of $100+/MWh (versus $40–60/MWh for wholesale power), the premium Microsoft is paying signals just how desperate hyperscalers are for reliable clean baseload.

Meta's 1.2 GW Nuclear Ambition

Meta followed with an RFP for 1–4 GW of new nuclear capacity, the largest single nuclear procurement request from a private company in history. Meta's specifications require delivery between 2030 and 2035 — timeline feasibility that effectively limits responses to SMR developers and existing reactor restarts. The 1.2 GW initial target would require the equivalent of four NuScale VOYGR plants or two large conventional reactors. Meta has explicitly stated that nuclear is “a necessary component of a reliable clean energy portfolio” for its data center expansion.

Google and Kairos Power SMRs

Google took a different approach, signing a first-of-its-kind deal with Kairos Power, a startup developing molten-salt-cooled small modular reactors. The agreement targets 500 MW of SMR capacity by 2035, with the first reactor expected online by 2030. Google's bet is on next-generation technology rather than existing reactor restarts, reflecting a longer time horizon and higher risk tolerance.

Amazon, not to be left out, has been acquiring data center sites adjacent to existing nuclear plants and signed deals with Talen Energy for nuclear-powered data center capacity in Pennsylvania. The Susquehanna nuclear facility deal provides up to 960 MW of nuclear power directly to AWS data centers.

Stock-by-Stock Analysis: Five Nuclear Plays for the AI Era

The nuclear energy investment universe spans operating utilities, uranium miners, and pre-revenue technology developers. Each offers a fundamentally different risk-reward profile.

Constellation Energy (CEG) — The Blue Chip

Constellation is the largest nuclear operator in the United States, with 21 GW of nuclear capacity across 14 plants. The Three Mile Island restart deal with Microsoft was the catalyst that re-rated the stock from $120 in mid-2024 to over $330 by early 2026 — a 175% move. The thesis is straightforward: Constellation owns irreplaceable assets in a market where new nuclear capacity takes 10–15 years to build, and hyperscalers are willing to pay premium prices for clean baseload power.

The bull case rests on additional PPA announcements. Every new hyperscaler nuclear deal increases the scarcity value of Constellation's existing fleet. Management has guided for $8–9 per share in adjusted earnings for 2026, but if premium nuclear PPAs expand across the fleet (currently only Three Mile Island), upside to $12+ in earnings is plausible. At current prices around $330, the stock trades at roughly 37x forward earnings — expensive for a utility, but potentially cheap if the PPA pipeline materializes.

The bear case: valuation has gotten ahead of fundamentals. Most of Constellation's nuclear output is still sold at standard wholesale or regulated rates, not at the premium Microsoft pricing. If additional hyperscaler deals take longer than expected, or if SMRs eventually increase nuclear supply, the scarcity premium could compress.

Vistra (VST) — The Value Play

Vistra operates 6.4 GW of nuclear capacity through its Comanche Peak facility in Texas, plus a diversified portfolio of natural gas, solar, and battery storage assets totaling 41 GW. The stock was the best performer in the S&P 500 in 2024, rising over 260%, driven by the AI power demand narrative and the Energy Harbor acquisition that added nuclear capacity.

Vistra is cheaper than Constellation on most metrics — roughly 18x forward earnings versus 37x — because its nuclear exposure is diluted by conventional thermal assets. But this actually creates optionality: if nuclear PPAs expand, Vistra could announce similar deals for Comanche Peak. Meanwhile, the Texas power market (ERCOT) is one of the tightest in the country, with reserve margins below 10%, which benefits all of Vistra's generation assets regardless of the nuclear story.

Cameco (CCJ) — The Uranium Commodity Play

Cameco is the world's second-largest uranium producer (behind Kazakhstan's Kazatomprom) and the largest publicly traded pure-play on uranium. The thesis is simple: every nuclear reactor, whether existing or new, requires uranium fuel. Global uranium demand is projected to exceed supply by 20–30 million pounds annually through 2030 as restarts, life extensions, and new builds accelerate. Spot uranium prices have already risen from $50/lb in early 2024 to approximately $85/lb in early 2026.

Cameco's McArthur River mine in Saskatchewan is one of the highest-grade uranium deposits in the world, giving it a significant cost advantage. The company also owns 49% of Westinghouse Electric (in partnership with Brookfield), providing exposure to nuclear fuel fabrication, reactor services, and new build engineering — the full nuclear value chain.

NuScale Power (SMR) — The SMR Leader

NuScale holds the distinction of having the only SMR design certified by the US Nuclear Regulatory Commission. Its VOYGR reactor module is a 77 MW pressurized water reactor that can be factory-built and combined in multi-module configurations up to 924 MW. The stock has been a roller coaster — surging on nuclear enthusiasm, crashing when its initial Idaho project was cancelled in late 2023 due to cost overruns, then recovering as hyperscaler interest reignited the SMR thesis.

NuScale is pre-revenue and will not generate meaningful income until at least 2029–2030. This is a venture-stage bet wrapped in a public equity. The addressable market is enormous — the US Department of Energy estimates SMR demand at 300+ GW globally by 2050 — but execution risk is severe. Construction costs for first-of-a-kind nuclear projects historically exceed estimates by 2–3x. Investors should size positions accordingly: this is optionality, not a core holding.

Oklo (OKLO) — Sam Altman's Nuclear Bet

Oklo is developing a sodium-cooled fast reactor called the Aurora, designed specifically for distributed power generation at data center scale (15–50 MW per unit). The company went public via SPAC in 2024 with Sam Altman, CEO of OpenAI, as chairman of its board — a connection that immediately linked Oklo to the AI power narrative.

Oklo's advantage is its focus on used nuclear fuel recycling, which could reduce both fuel costs and waste. The Aurora reactor is designed to be “walk-away safe” — passively cooled with no risk of meltdown — addressing public safety concerns that have historically blocked nuclear development. However, Oklo's NRC application was rejected in 2022 for insufficient technical detail, and the company is in the process of resubmitting. This is the earliest-stage of the nuclear plays covered here, with first reactor deployment unlikely before 2031.

The Bull Case: Why Nuclear Re-Rating Has Further to Run

Three structural drivers suggest the nuclear re-rating is not yet complete, despite the massive moves already seen in Constellation and Vistra.

First, the PPA pipeline is still early innings. Microsoft's Three Mile Island deal covers one reactor. Constellation owns 14 plants with 21 GW. If even 20–30% of that fleet transitions from wholesale pricing to hyperscaler PPAs at 50–100% premiums, the earnings impact would be transformational — potentially doubling Constellation's per-share earnings by 2028–2029.

Second, nuclear plant life extensions are accelerating. The NRC has begun granting 80-year operating licenses (up from the original 40-year design life plus 20-year extensions). This dramatically extends the revenue runway for existing plants and increases the present value of nuclear assets that were previously expected to decommission by the 2040s.

Third, the political landscape has shifted. Nuclear energy enjoys rare bipartisan support in the US. The ADVANCE Act, passed in 2024, streamlined NRC licensing and provided production tax credits for nuclear power. Both parties view nuclear as essential for energy independence, AI competitiveness, and decarbonization. This political tailwind reduces regulatory risk, which has historically been the sector's largest overhang.

For investors analyzing how power demand flows through to utility earnings, our deep dive on AI infrastructure investment in data centers provides the capex and power frameworks.

The Bear Case: What Could Go Wrong

We hold nuclear as a core conviction but honest analysis requires addressing the material risks.

Valuation is stretched. Constellation has gone from a boring 12x earnings utility to a 37x growth stock in 18 months. If AI capex decelerates or additional PPA deals take longer than expected, the multiple compression could be severe. A reversion to even 20x earnings would imply 45% downside from current levels.

Construction risk for new nuclear is real. The Vogtle expansion in Georgia — the only new nuclear plant built in the US in decades — came in at $35 billion versus a $14 billion original estimate, with a seven-year schedule delay. If SMR developers face similar overruns, the economics could favor natural gas with carbon capture or next-generation geothermal.

AI power demand could moderate. DeepSeek V4 demonstrated that equivalent AI performance can be achieved with dramatically less compute. If efficiency improvements outpace demand growth, the 35 GW data center power projection could prove too aggressive. We addressed this tension in our analysis of the best AI tools for investment research in 2026, which explores how efficiency and demand interact across the AI ecosystem.

Finally, there is tail risk from a nuclear incident anywhere in the world. A Fukushima-scale event, even at a facility unrelated to any US company, would likely trigger a selloff in all nuclear equities and could restart political opposition that has been dormant for years.

Portfolio Construction: Sizing Nuclear in an AI-Focused Portfolio

Our recommended approach is a barbell. Allocate the majority of nuclear exposure to profitable operating companies (Constellation, Vistra, Cameco) and a smaller allocation to pre-revenue SMR developers (NuScale, Oklo) as optionality.

For a diversified equity portfolio, we believe 3–7% total allocation to nuclear energy is appropriate in the current environment. Within that, a 60/20/20 split between operating utilities (CEG/VST), uranium (CCJ), and SMR optionality (SMR/OKLO) balances near-term cash flow visibility with long-term upside.

Entry points matter. Constellation at 37x is a fundamentally different risk/reward than Constellation at 20x. We recommend averaging into positions over 3–6 months rather than taking full positions at current valuations, and using AI-driven monitoring tools to track the PPA announcements, NRC filings, and utility earnings that drive these stocks.

The 2030 Outlook: Where Nuclear Energy Stocks Go from Here

By 2030, we expect the nuclear energy landscape for investors will look fundamentally different than today. The operating utility thesis (Constellation, Vistra) will either be validated by a wave of additional hyperscaler PPAs — in which case these stocks will have delivered another 50–100% upside — or the PPA pipeline will have stalled, and multiples will compress back toward utility norms.

Cameco's trajectory depends on uranium supply/demand dynamics. With minimal new mine development globally and 60+ reactors under construction (primarily in China, India, and the Middle East), the supply deficit is structural. We believe uranium prices will reach $120–150/lb by 2028, which would make Cameco a strong compounder from current levels.

The SMR story will be the most dramatic. By 2030, either NuScale, Kairos, or another developer will have demonstrated that factory-built reactors can be deployed on time and on budget — which would unlock a $200B+ addressable market — or the technology will have failed to deliver, and investors in pre-revenue names will have lost most of their capital.

That asymmetry is exactly why we maintain the barbell: operating nuclear for the high-probability, moderate-return thesis, and SMR exposure for the low-probability, transformational-return thesis.

Frequently Asked Questions