Nuclear and SMR Stocks in an AI-Powered Grid

In September 2024, Microsoft agreed to pay Constellation Energy to bring a dead nuclear reactor back to life. Three Mile Island Unit 1, the reactor that shares a site with the most infamous nuclear accident in American history, retired in 2019 because it couldn't compete with cheap natural gas. Now it's being rebuilt to feed data centers, under a 20-year power purchase agreement, at a cost of about $1.6 billion for roughly 835 MWe of output.

I want to be clear about why this matters. This is not a press release. This is a hyperscaler with one of the strongest balance sheets on Earth deciding that the cheapest, fastest way to get firm, carbon-free, always-on power was to resurrect a shuttered reactor and sign a two-decade contract for all of it. When the smartest capital allocators in tech start underwriting nuclear like that, the thesis has teeth. The question for an investor is whether the stocks riding that thesis are actually the ones that will capture it.

The restart is real, and it's ahead of schedule

Constellation renamed the plant the Crane Clean Energy Center, and the progress since the deal was signed has been genuinely fast for nuclear. By mid-2026 the restart was more than 65% staffed and had hit real technical milestones, including successful operation of the main generator and turbines. Constellation is now targeting power delivery to Microsoft as early as 2027, pulled forward from an earlier in-service estimate of 2028.

The regulatory and financing pieces have fallen into place too. In November 2025 the Department of Energy's Office of Energy Dominance Financing issued a loan guarantee of up to $1.0 billion to support the restart. Then on June 1, 2026, FERC approval cleared the last major grid obstacle. When you strip out the narrative, what you have is a de-risked project with a signed 20-year buyer, government backing, and a first-mover slot. Out of the 13 announced hyperscaler nuclear projects totaling more than 9.8 GW by May 2026, Microsoft's restart is scheduled to be the first one online.

Everyone piled in, and the deals got bigger

Once Microsoft moved, the rest of the hyperscalers followed, and they didn't follow quietly. The commitments escalated fast through 2025 and into 2026.

• Meta made the biggest bet. In January 2026 it signed deals for up to 6.6 GW of nuclear power by 2035, the largest single-hyperscaler nuclear commitment to date. That package includes a 20-year PPA with Constellation for the Clinton Clean Energy Center in Illinois, a 1.2 GW campus tied to Oklo in Ohio, a TerraPower Natrium fleet, and options for a 300 MW SMR at a Vistra site.
• Amazon went big on an existing plant. In June 2025 it expanded its Talen Energy PPA to up to 1,920 MWe from the Susquehanna plant through at least 2042, a deal valued around $18 billion. Separately it agreed with X-energy to deploy 960 MW of Xe-100 reactors in the 2030s.
• Microsoft got there first. The Three Mile Island restart remains the one scheduled to actually deliver electrons in 2027, while most of the rest are 2030s stories.

Notice the pattern. The deals that are near-term and bankable are restarts and PPAs off existing, licensed, operating plants. The deals that involve brand-new reactor technology, the SMRs and advanced designs, are dated 2030 and beyond. That timing gap is the entire investing story, and it's where the three headline stocks split into completely different risk profiles.

Three tickers, three totally different bets

People talk about “nuclear stocks” like it's one trade. It isn't. Constellation, Oklo, and NuScale are about as different as three companies in the same theme can be.

Constellation (CEG): the one that actually makes money

As of July 14, 2026, Constellation traded near $256 with a market cap around $92 billion. This is not a story stock. It's the largest operator of nuclear plants in the United States, it generates real cash flow today, and the AI-nuclear narrative is upside on top of an already-working business. The Microsoft and Meta PPAs give it long-dated, high-credit-quality contracted revenue on capacity it largely already owns. If you believe the thesis but you don't want to bet on a reactor that has never been built, Constellation is the low-drama way to own it.

Oklo (OKLO): the pure narrative bet

Oklo is the opposite. As of July 14, 2026, it traded near $46 with a market cap of about $8 billion, down hard from a 52-week high of $193.84. And here is the thing every buyer needs to sit with: Oklo is pre-revenue. No operating reactor. It posted an adjusted loss of $0.18 per share in Q1 2026. What it does have is roughly $2.5 billion in cash after a $1.182 billion raise in January 2026, and about 14 GW of non-binding letters of intent, including a 12 GW master power agreement with Switch running through 2044.

Letters of intent are not orders. A non-binding LOI is a company saying it would like to buy power if and when Oklo can actually deliver it, which right now it can't, because the first Aurora powerhouse at Idaho National Laboratory is targeted for late 2027 to 2028. There was real good news in May 2026: the NRC approved Oklo's Principal Design Criteria topical report for the Aurora powerhouse in under half the traditional review timeline, and the stock jumped roughly 12% to 22% intraday on it. But a design criteria approval is a milestone, not a license. Oklo still has no approved combined license, and the NRC had denied its earlier custom application without prejudice back in January 2022.

NuScale (SMR): the approved design nobody's ordering

NuScale is the strangest case, and honestly the most instructive. It's the only company with an NRC-approved SMR standard design, and not just one: it holds approvals for both its original 50 MWe module and the upsized 77 MWe version. On paper, that's the crown jewel. Regulatory approval is supposed to be the hardest, slowest part of the whole business. NuScale cleared it first.

And yet, as of July 14, 2026, the stock traded near $8.35 with a market cap around $2.6 billion, roughly 84% below its 52-week high of $57.42. The problem isn't the technology or the regulator. It's that NuScale's contracts have been painfully slow to convert from memoranda of understanding into firm orders. It's pursuing a potential 6 GW deployment program with the Tennessee Valley Authority via ENTRA1, plus the six-module RoPower project in Romania. Those are big numbers. They're also, so far, mostly agreements to agree.

Put Oklo and NuScale side by side and you get the clearest picture of the sector. Oklo has 14 GW of interest and no approved license. NuScale has the approvals and can't convert its pipeline into binding orders. Both stocks are betting the missing half shows up before the cash runs out. That's a real bet. It's just not the same bet as owning Constellation.

The demand is not the question. The supply mix is.

Let me steelman the bull case first, because it's strong. The demand side of this is not hype. The EIA's January 2026 outlook forecast the strongest four-year growth in U.S. electricity demand since 2000, driven directly by data centers. Their share of peak summer power demand is projected to rise from 4.1% in 2025 to 5.3% in 2026 and 8.5% in 2027. That is a genuine, structural, hard-to-fake surge in electricity consumption, and it's happening now.

Here's the part the nuclear bulls tend to skip. That demand is real, but it doesn't follow that nuclear captures most of it. The same EIA outlook projects nuclear generation stays essentially flat, holding around 18% of the U.S. mix in the near term and actually falling to 12% to 15% by 2050. Meanwhile solar adds nearly 70 GW and supplies the largest generation increase in both 2026 and 2027.

This is the tension you have to hold. The nuclear story is simultaneously true and overhyped. True, because hyperscalers really are signing multi-gigawatt, multi-decade contracts and paying to restart retired reactors. Overhyped, because the aggregate numbers say nuclear's share of the grid barely moves while a pre-revenue reactor developer briefly carried a $17 billion valuation on the strength of letters of intent.

How I'd actually frame the trade

I don't give buy or sell calls, but I can tell you how the risk actually stacks, because the three names are not interchangeable.

• Constellation is the thesis without the moonshot. You're buying a profitable operator whose existing plants are suddenly worth more because hyperscalers will pay a premium for firm, clean power. The Three Mile Island restart is a call option that's already mostly funded and de-risked. The downside is that a lot of good news may already be in a $92 billion market cap.
• Oklo is a venture bet in public-market clothing. The $2.5 billion cash pile buys it years of runway, and the fast-tracked NRC design approval is a genuine tailwind. But you are underwriting a company that has to build and license a first-of-a-kind reactor, on time, before the story fatigues. Size the position like the binary it is.
• NuScale is the cautionary tale about approvals. Having the only NRC-approved SMR design was supposed to be the moat. An 84% drawdown from the high tells you the market has decided that clearing the regulator means little until firm orders show up. If the TVA program or RoPower converts to binding contracts, the story changes overnight. Until then, it's a design in search of a customer.

The mental model I keep coming back to is the gap between announcement and electron. A signed 20-year PPA on a reactor that already exists, like Three Mile Island, is close to the electron. A 14 GW letter of intent on a reactor that hasn't been built is very far from it. The market keeps pricing those two things as if they carry similar certainty, and they do not.

Microsoft restarting Three Mile Island is the real signal in all of this. It's a hyperscaler voting with $1.6 billion and a 20-year commitment that firm nuclear power is worth paying up for in an AI-hungry grid. That vote is genuine, and it's bullish for the operators who already own the plants. Whether it's bullish for the pre-revenue reactor developers depends entirely on whether they can turn milestones into machines before the narrative moves on. The demand is not in doubt. The delivery is the whole game.