AI data centers need large amounts of continuous, low-carbon power, and that has revived commercial interest in nuclear energy. Hyperscalers are signing long-term agreements with existing nuclear plants and backing advanced reactor developers. For investors looking to invest in nuclear energy startups, the opportunity reaches beyond small modular reactors to fuel, components and project development. The real test is which companies can commercialize at competitive costs, and at valuations that leave room for execution risk.
Why AI is driving nuclear energy's comeback
Five years ago, the US nuclear story was about closures. Plants with decades of life left were shutting down because they could not compete with cheap gas and subsidized renewables. Today, one of those shuttered reactors is being brought back to supply a single technology company, and others are being kept open under 20-year contracts with data center operators.
The reason is the type of power AI needs. Training and serving AI models runs around the clock at high utilization, so data centers need firm electricity, not just annual energy volumes. Many hyperscalers have also committed to carbon-free power targets. Few sources deliver all three at scale: continuous output, low carbon and long asset lives. Existing nuclear plants do, which is why they were the first to be signed up. Gas, geothermal, hydro, and renewables paired with storage all compete for the same demand, and the International Energy Agency expects renewables and natural gas to supply most of the growth in data center electricity this decade.
AI is an accelerator rather than the only driver. Energy security, industrial electrification and decarbonization policy all point the same way. But demand for nuclear electricity is not the same as profit for nuclear startups. Buying power from a reactor that already exists is a very different proposition from building a new one, and investors need to keep the two apart. For the wider picture of AI's power constraint, see AI data center power constraints.
The demand signal: hyperscaler nuclear agreements
The four largest cloud and AI spenders have all made nuclear commitments, but they fall into two very different categories: buying power from plants that already exist, and backing reactors that have not yet been built.
Microsoft: restarting Three Mile Island Unit 1
In September 2024, Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart the 835 MW Three Mile Island Unit 1, renamed the Crane Clean Energy Center. Constellation still targets a 2027 restart, backed by a $1 billion DOE loan, although grid operator PJM has indicated some transmission upgrades could take until 2031 and Constellation has asked FERC to transfer interconnection rights to keep the schedule.
Meta: keeping Clinton running
In June 2025, Meta signed a 20-year agreement for the 1,121 MW output of Constellation's Clinton plant in Illinois, starting in June 2027 when the state's zero-emission credit program expires. The deal includes a 30 MW uprate and keeps an existing plant open rather than building a new one.
Amazon: existing supply plus an SMR bet
Amazon is doing both. Under a June 2025 agreement, Talen Energy will supply 1,920 MW from its Susquehanna plant through 2042, ramping to full volume by 2032. Separately, Amazon led a $500 million investment round in X-energy, a developer of high-temperature gas reactors and their fuel, targeting more than 5 GW of new capacity by 2039.
Google: the first SMR order book
Google's October 2024 agreement with Kairos Power aims to bring 500 MW of advanced reactors online by 2035. The first project, Kairos's 50 MW Hermes 2 molten salt reactor in Oak Ridge, Tennessee, is scheduled to begin operating in 2030, with TVA buying its power under the first US utility contract for a Generation IV reactor.
| Agreement | Category | Capacity | Timing | Development status | Main remaining uncertainty |
|---|---|---|---|---|---|
| Meta / Constellation (Clinton) | Existing plant, life extension | 1,121 MW + 30 MW uprate | 20 years from June 2027 | Commercially operating | Contract, operating and relicensing risks |
| Amazon / Talen (Susquehanna) | Existing plant | 1,920 MW | Ramping to 2032, through 2042 | Commercially operating | Contract, operating and regulatory risks |
| Microsoft / Constellation (Crane) | Existing plant, restart | 835 MW | Restart targeted 2027 | Previously operating, being refurbished | Restart approvals, refurbishment, grid connection |
| Google / Kairos (Hermes 2, 500 MW program) | New advanced reactors | 50 MW first, 500 MW total | 2030 first, 2035 full | Pre-commercial development | Licensing, financing, construction, performance |
| Amazon / X-energy | Equity in reactor developer | Over 5 GW targeted | By 2039 | Pre-commercial development | Technology, licensing, financing, construction |
The pattern matters for investors. The most certain contracts so far are for existing plants owned by listed utilities. The commitments to new reactors are real, but they are smaller, later and conditional on the technology being built on time and on budget.
What are small modular reactors, and why do they matter?
Small modular reactors (SMRs) are reactors of roughly 300 MW or less, a fraction of a conventional 1,000 MW-plus plant. The promise is in the word modular: build standardized units in a factory, ship them to site, and add capacity in increments instead of betting billions on one large project. Smaller units can also fit sites that a large plant cannot, including next to industrial customers and data centers.
That promise has not yet been proven at scale. Factory economics only appear once many identical units are built, and no SMR design has reached that point in the United States.
| Technology | Defining characteristics | Key investment consideration |
|---|---|---|
| Light-water SMRs | Smaller versions of established water-cooled reactor technology | Construction economics and repeatability |
| High-temperature gas reactors | Helium-cooled, higher temperatures, potential industrial heat applications | Fuel supply and demonstration |
| Sodium-cooled fast reactors | Liquid-metal cooling, often paired with thermal storage | First-of-a-kind cost and schedule |
| Molten salt reactors | Molten salt coolant and, in some designs, fuel | Materials, licensing, technical maturity |
| Microreactors | Very small units for remote, military or distributed use | Unit economics and market size |
Microreactors are defined by size and use, not by coolant; they can use several of the technologies above. Maturity varies widely by design.
The first-of-a-kind problem. First-of-a-kind nuclear projects routinely run over budget and behind schedule. TerraPower's Kemmerer project, for example, now targets operation in 2031, five to seven years later than the original DOE timeline. The investment question is not whether the first plant is expensive, but whether the developer can show a credible path to cheaper units after it.
Policy and regulation: what has actually changed
Policy toward advanced nuclear has shifted sharply since 2024, and some of the shift has already produced concrete results.
Licensing is moving. In May 2025, the NRC issued a standard design approval for NuScale's 77 MW US460 design; NuScale's earlier 50 MW module already held full design certification, making it the only SMR developer with an NRC-certified design. In March 2026, the NRC issued TerraPower a construction permit for its Natrium reactor in Wyoming, the first ever for a commercial non-light-water power reactor, after completing its safety review ahead of schedule. A construction permit allows building to begin; a separate operating license is still required before the plant can run. Other applications, including X-energy's Long Mott project with Dow in Texas, are under review.
DOE is testing reactors directly. A May 2025 executive order told the Department of Energy to bring at least three test reactors to criticality by July 4, 2026, under its own authorization rather than NRC licensing. The target was met: Antares Nuclear reached criticality first in June 2026, followed by Valar Atomics later that month and, on the morning of July 4, Aalo Atomics, with four reactors in total reaching criticality by the deadline. These were zero-power tests, not electricity-producing plants, but they show test timelines can be compressed dramatically.
Fuel and finance support. DOE programs also support domestic fuel supply, including HALEU, and the agency's loan office is financing projects such as the Crane restart.
Policy can shorten timelines and lower the cost of capital. It cannot remove technology, construction or commercial risk, and priorities can change with administrations. Investors should value milestones that have been achieved, such as permits issued and reactors operated, above announcements and targets.
Where is the private-market opportunity in nuclear energy?
Start with what is already public. The most established nuclear revenue today, from existing plants under long-term contracts, belongs to listed utilities such as Constellation and Talen. Several reactor developers, including NuScale and Oklo, are also publicly traded. Their share prices are a useful, if volatile, reference point for private valuations in the sector.
The private opportunity therefore sits elsewhere, across four segments with very different risk profiles:
| Segment | What it does | Where value comes from | Main dependency |
|---|---|---|---|
| Reactor developers | Design and commercialize new reactors | Technology, IP, order books | Licensing, first plant, funding |
| Fuel, enrichment and HALEU | Convert, enrich and fabricate fuel, including HALEU for advanced designs | Supplying many reactor developers | Scale-up, policy support |
| Components and services | Specialized manufacturing, engineering, instrumentation | Selling into multiple projects | Volume of projects actually built |
| Project developers and operators | Finance, build, own and run plants | Long-term power contracts | Construction cost, financing |
Visual for publication: nuclear investment value chain, from reactor developers and fuel suppliers through components and project developers to electricity customers (data centers, utilities, industry).
The fuel segment deserves particular attention. Many advanced designs need high-assay low-enriched uranium (HALEU), and domestic supply is limited. A fuel supplier can benefit from several reactor developers succeeding without needing to bet on which one wins, a shovels-in-a-gold-rush position worth examining. For a deeper look, see HALEU fuel supply and investment opportunities.
The most strategically important segment is not necessarily the most attractive investment. A reactor developer with promising technology may already carry a valuation that assumes successful licensing, construction and deployment. A fuel supplier or specialized manufacturer offers a different risk profile, but its price must still reflect the scale and timing of real customer demand. The question is not only how large the potential market is, but how much of that potential is already in the entry price.
In advanced nuclear, a company's financing stage does not necessarily match its commercial maturity. A business may have raised substantial private capital while still facing major technology, licensing, construction and financing milestones before it earns commercial revenue. A company with a construction permit and a utility contract is a very different investment from one that has just reached zero-power criticality, even if their latest rounds look similar.
How to evaluate nuclear energy startups: the ARCA Scarcity Test applied to nuclear
The five questions of the ARCA Scarcity Test, introduced in our analysis of AI infrastructure investing, apply directly to nuclear, with sector-specific evidence behind each one.
| # | Test | The question | Nuclear evidence to look for |
|---|---|---|---|
| 1 | Scarcity | Does the company control a genuinely scarce or hard-to-replicate resource? | Licensed or approved designs, demonstrated reactors, secured fuel supply, permitted sites |
| 2 | Revenue quality | Is demand backed by creditworthy customers and contracts that outlast the assets they pay for? | Binding power agreements with creditworthy buyers, versus memoranda of understanding and "collaborations" |
| 3 | Capital efficiency | How much more capital is needed to grow, and who provides it: customers, equity or debt? | Overnight cost per kW, total financed cost, a credible first-of-a-kind to nth-of-a-kind cost curve |
| 4 | Valuation | What future does the entry price already assume, and what happens to returns if growth arrives two years late? | Valuation relative to milestones achieved, and the effect on returns if commercial operation slips |
| 5 | Liquidity | What are the realistic exit routes and timelines, and what transfer restrictions apply in the meantime? | IPO readiness, public peer valuations, transfer restrictions, likely further dilution |
Visual for publication: Scarcity Test graphic, matching Article #7.
The second question deserves the most skepticism. In this sector, announcements are cheap and contracts are expensive. A company can list partnerships with hyperscalers, utilities and governments without holding a single binding obligation to buy power. Ask which agreements are binding, at what price, and what conditions must be met before the buyer is committed.
On capital efficiency, levelized cost of electricity (LCOE) estimates are useful but fragile: they depend on assumptions about construction cost, financing, capacity factor and plant life that are rarely tested for a new design. Treat a developer's LCOE as a target, not a forecast.
The risks of investing in SMRs and advanced nuclear
1. Regulatory and technology risk. Licensing can take longer or impose new requirements, and designs that work in a test reactor may not perform reliably at commercial scale.
2. Construction risk. First-of-a-kind plants routinely exceed budgets and schedules, and nuclear construction has a long history of overruns.
3. Fuel risk. Advanced designs that depend on HALEU face a supply chain still being built.
4. Commercial risk. Customer interest may never become a binding, financed project.
5. Financing, valuation and liquidity risk. Developers may need several more funding rounds before revenue, diluting early investors; high entry prices and slow paths to exit can erase the gains from technical success.
Nuclear development and investment timelines can run well beyond typical technology product cycles. Investors used to software or even semiconductor companies should expect longer development periods and larger capital needs before any return.
Frequently asked questions
How can I invest in nuclear energy startups?
Accredited investors typically access private nuclear companies through direct participation in funding rounds, venture or growth equity funds, special purpose vehicles (SPVs), or secondary purchases where the company permits them. Publicly listed nuclear developers and utilities offer a separate, more liquid route. For how private structures compare, see SPV vs. venture fund.
Are small modular reactors commercially operating in the United States?
Not yet. NuScale holds the only NRC-certified SMR design, advanced reactors have construction permits, and several test reactors reached criticality in 2026, but no commercial SMR is yet supplying power to the US grid. The first advanced reactor projects under hyperscaler agreements target around 2030.
What is HALEU and why does it matter?
High-assay low-enriched uranium is enriched to between 5% and 20% uranium-235, higher than fuel for conventional reactors. Many advanced designs need it, and limited domestic supply makes it a potential bottleneck and an investment opportunity in its own right.
Conclusion: demand is growing, but commercialization remains the test
AI has given nuclear energy something it lacked for decades: large, creditworthy buyers willing to sign 20-year contracts. So far, most of that demand has gone to existing plants owned by listed utilities. The opportunity in new reactors is real but earlier, and it spans more than reactor designers: fuel, components and project development each offer a different mix of technology, commercial and capital risk.
AI may be accelerating nuclear energy's next chapter, but rising electricity demand alone will not determine the winners. For private-market investors, the opportunity lies in identifying companies capable of turning nuclear innovation into commercially viable infrastructure, at valuations that leave room for the risks ahead.
