How to evaluate small modular reactor stocks in the 2026 hyperscaler buying spree

9.8 gigawatts. That is the volume of nuclear capacity U.S. hyperscalers have already committed to buy, and signed rather than announced. Microsoft, Google, Amazon and Meta have each locked in long-term power purchase agreements for small modular reactors (SMRs). The first electrons from a restarted reactor are scheduled to reach an AI data center in 2027.

As we wrote in August, AI data centers began buying fusion power a decade before it will exist. The SMR trade is different: the reactors are real, the permits exist, and the construction has started. The question for an investor is no longer whether this sector gets built. It is which of the sixteen public vehicles gets paid first.

The buying spree that changed the sector

Every major U.S. hyperscaler now has at least one nuclear deal. Microsoft locked in 835 MW via a 20-year power purchase agreement for the Three Mile Island Unit 1 restart, targeting 2027–2028. Google committed to 500 MW from Kairos Power. Amazon invested $700 million in X-energy for up to 12 Xe-100 reactors, and Meta signed agreements across TerraPower's Natrium, Oklo's Aurora, Vistra and Constellation, for up to 6.6 GW in total.

These are not speculative options. A power purchase agreement of this size commits the offtaker to buy output for 15 to 20 years, which is exactly what a first-of-a-kind reactor needs to raise $1 billion to $3 billion per plant in construction finance.

That is the structural change. The buyers are creditworthy, desperate for firm power, and willing to underwrite projects a decade before electrons flow.

The pattern repeats outside the United States. The European Commission unveiled a full SMR strategy in March 2026 backed by a €200 million guarantee, explicitly modeled on the U.S. hyperscaler-PPA approach. Rolls-Royce SMR won the UK selection process and is targeting first concrete at Wylfa as early as 2027. Poland is procuring a BWRX-300 fleet, and Romania's NuScale-based RoPower project targets 2029. The American template, a deep-pocketed offtaker underwriting a first-of-a-kind build, is now export policy.

The reactor designs on the table

NuScale remains the only SMR design with an approved standard design from the U.S. Nuclear Regulatory Commission. That certification matters more than any roadmap, because utilities and hyperscalers can reference it in their own license applications. Its fuel runs on conventional low-enriched uranium, a practical advantage over the sodium-cooled and high-assay designs that depend on a HALEU supply chain still being built.

Beyond the three leaders sit a long tail with different risk profiles. Oklo's Aurora is a fast reactor aimed at remote and off-grid use, and the company received DOE startup authorization for its Groves test reactor in July 2026, one step short of first criticality. Kairos Power is building two demonstration reactors in Tennessee under its Google agreement. X-energy's Xe-100, backed by Amazon, is a high-temperature gas reactor. Last Energy sells 20 MW microreactors on a buy-the-plan, not buy-the-shares model. Each design has its own licensing path, fuel requirement and failure mode. The sector is not one trade.

The money

Sixteen publicly traded names span three very different businesses: pure-play reactor developers, diversified industrials with reactor programs, and the fuel and component suppliers who get paid regardless of which design wins.

In 2025, the pure plays went vertical. NuScale and Oklo gained 200% to 300% on a wave of data center partnerships. In 2026, both gave back 20% to 30% as the market began asking a harder question: when do signed announcements turn into billed megawatts?

The money behind these deals is now institutional rather than venture. X-energy's April 2026 IPO raised $1 billion. Antares closed a $470 million Series C in July led by Paradigm. Radiant Energy, Standard Nuclear and Last Energy each raised more than $100 million since December. Holtec filed for its own IPO in February targeting a $10 billion valuation. The funding base has shifted from angel-led speculative rounds to infrastructure-scale capital, a tell that the market has begun pricing these as real assets rather than research projects.

Cameco's move matters. Uranium supply is the one segment where revenue exists today and demand is structurally guaranteed by every design on this list.

The fuel bottleneck everyone avoids

The cleanest way to play this cycle without touching reactor schedule risk is the fuel chain. The binding constraint sits at one point: high-assay, low-enriched uranium. Every non-light-water design, from Natrium and Xe-100 to Kairos's fluoride-salt reactor and Oklo's fast reactor, needs HALEU enriched to roughly 20% U-235. In 2024 the material was a documented bottleneck. The demand signal from big tech is now pulling capital into commercial fuel production, and Centrus Energy is the only U.S. producer with HALEU enrichment capability at scale.

The other durable play is conventional uranium. Cameco is the largest listed producer. Its decision to fold Westinghouse into its stack, pairing mining with reactor supply, signals an industry expecting a decades-long buildout. Whatever the reactor-design race resolves to, the fuel demand is already committed.

The fuel thesis inverts the usual risk. Reactor developers fail on schedule; fuel suppliers fail only if the whole sector collapses. In a market where the buyers are underwriting 15-year offtakes, that asymmetry is worth more than any roadmap.

Costs that still embarrass the sector

The honest number comes from Lazard's analysis: first-generation SMR electricity at about $214 per megawatt-hour. That is higher than every newly built power plant except the most expensive gas turbines. The entire investment thesis rests on the factory-built learning curve kicking in, and that curve has not arrived yet.

The other forgotten history is NuScale's canceled Carbon Free Power Project. In November 2023 the utility consortium walked away as the target price rose toward $90 per megawatt-hour. It was a cost problem, not a technology problem, and it is the exact risk profile that still hangs over every developer.

The skeptics have receipts

There's a lot of talking, but nobody is writing the checks. The investment required to deploy even just one SMR unit is more than that.— Jacopo Buongiorno, professor of nuclear science and engineering, MIT

Buongiorno made that point in April 2025, and the intervening year has been kind to it. The construction permit for TerraPower's Natrium came with a completion target of 2030, six years from groundbreaking, against an industry habit of doubling the timeline. The Nature Communications study published in August 2026 models the economics across the U.S. grid and industrial sectors; its most favorable deployments are hydrogen and ammonia plants where the 2033 production tax credit deadline forces decisions.

No Gen III+ SMR in the United States has yet delivered a plant on time, on budget and to performance targets. The sector is selling a promise backed by permits, not by operating reactors.

The race for firm power has rivals

SMRs are not the only technology bidding for the hyperscaler megawatt. Enhanced geothermal, gas paired with carbon capture, and long-duration storage each pitch the same promise: firm, carbon-constrained power delivered behind the meter. The comparison is blunt. Geothermal startups in Nevada and Utah can match the baseload story without a spent-fuel debate, and their projects come online in four to six years rather than a decade. Gas-plus-capture wins when interconnection upgrades are cheap and carbon-intensity targets are loose.

SMRs win when the opposite is true. Where a hyperscaler faces strict carbon accounting, a state grid queue measured in years, and political sensitivity around new gas, nuclear is the only option that checks every box. That is precisely why the data center crowd has adopted it as the default answer to the question nobody else can answer cleanly.

That competition disciplines the sector. Every delay, every cost overrun on a first-of-a-kind build hands an argument to the geothermal and storage lobbies, and those lobbies are well funded. The SMR thesis does not require beating them outright. It requires the hyperscaler procurement teams to keep concluding that only nuclear delivers the volume. So far, the 13 signed deals say they do.

The economics of the comparison are still being written in real time. Monte Carlo models run by engineering firms pit nuclear baseload against hydrogen peakers powered by curtailed wind, and the winner varies county by county. That is not an argument against the sector. It is an argument for reading each project's power purchase agreement instead of the press release.

The regulatory runway is the true gate

Technology risk in this sector has largely been retired by certification; schedule and regulatory risk have not. The NRC's ADVANCE Act, passed in July 2024, streamlined the licensing process, cut fees, and directed a risk-informed framework. One executive order now requires the agency to review new reactor applications within 18 months, a dramatic shortening of a process that once stretched past a decade. But design review is not construction. NuScale's certified design still has not cleared the separate construction-permit and operating-license steps that can take 30 months or longer.

The result is a two-speed market. Certified designs and permitted sites trade at one valuation; unlicensed roadmaps trade at another. TerraPower holds a construction permit for Natrium but needs a separate operating license before it can run. Oklo cleared DOE startup authorization yet remains one criticality test away from a single operating microreactor. The spread between those states, not the reactor blueprint, is where the meaningful underwriting risk sits.

For a private-market investor the cleanest position is the project-level debt or streaming instrument tied to an already-licensed design and a signed hyperscaler offtake. For a public-market investor it is the fuel chain, which now has customers under multi-decade contract and no licensing gate at all.

How to weigh the vehicles

Three ways to hold the sector exist, and they are not equivalent. Direct equity in developers is a bet on schedule execution. The ETF route, the VanEck Uranium and Nuclear ETF or the leveraged SMR ETFs launched in 2026, bundles the winners with the losers and the fuel suppliers. The least-advertised route is the debt and streaming structures forming around individual projects, where a hyperscaler offtake contract converts into a fixed-revenue instrument.

For each, the same diligence applies. Does the company hold an NRC certification or permit, or only a roadmap? Does it sell fuel or components with existing revenue, or does it burn cash until a first plant completes? What does its 2029 calendar show: an operating reactor, a construction milestone, or another capital raise? The pure plays currently price in successful completion of everything they have announced; the suppliers price in nothing going catastrophically wrong.

Watch the financing structures as closely as the technologies. The moment a signed PPA reaches financial investment decision, the sector stops being a venture theme and becomes an infrastructure asset class with institutional debt attached.

The certification gap also explains why the sector's valuation spread keeps widening. A developer holding a standard design approval can be referenced in any future license application, which compresses its time-to-revenue for every subsequent site. A competitor still in pre-application review must discount every announced customer until the NRC signs off. That asymmetry compounds: the certified design wins more customers, funds more capital, and pulls further ahead with each regulatory milestone.

For anyone allocating capital in this cycle, the two speeds matter more than the reactor blueprints. Own the fuel or the certified designs, and the schedule risk sits with someone else's license docket.

Signals to track

The sector crosses from speculation to infrastructure the day a signed PPA reaches financial close. Watch that, not the stock tickers.

The sector has spent a decade selling roadmaps. Now it is selling permits, contracts and one operating reactor at a time. That is progress. The margin between the two is where the value will be made.

Sources