Nuclear power spent a decade in the penalty box. Regulators stalled, public opposition held firm after Fukushima, and financing dried up. Then something changed -- not the politics of nuclear energy, but the economics of electricity. Data centres started consuming power at a rate that no grid planner had anticipated, and grid managers discovered that the only dispatchable, zero-carbon baseload source that can operate around the clock without interruption is nuclear. That single realisation has rebuilt the investment case for nuclear power faster than any government subsidy programme ever could.
The demand signal is clear. Hyperscale data centre operators -- Microsoft, Google, Amazon, Meta -- have collectively signed or announced nuclear offtake agreements worth hundreds of terawatt-hours. These are not expressions of intent. They are binding purchase commitments that provide project developers with the revenue certainty needed to attract institutional capital. When a major technology company signs a twenty-year power purchase agreement with a nuclear plant, it changes the financing calculus completely. Lenders who would not touch nuclear on merchant risk can price a contract-backed project. That shift is already moving capital off the sidelines.
The Data Centre Demand Surge Is Structural, Not Cyclical
The electricity demand from AI training and inference infrastructure is not a temporary spike. Model training runs require sustained, high-density power delivery for weeks at a time. Inference serving -- the continuous operation of deployed models -- consumes power permanently. Data centre operators are building not just for current demand but for projected demand three to five years out, and their current projections consistently assume electricity consumption will multiple several times over within the decade.
Grid planners across North America and Europe have updated their long-range load forecasts dramatically upward. The US Energy Information Administration revised its commercial sector electricity outlook three times in eighteen months, each revision higher than the last. In Ireland and the Netherlands, data centre load has already strained national grids to the point where connection moratoriums were imposed. The UK National Grid's Future Energy Scenarios now treat data centre demand as a primary planning variable rather than a rounding error.
For data centre operators, the consequence of grid instability is catastrophic. A facility running AI inference workloads cannot tolerate unplanned downtime. Power quality -- voltage stability, frequency consistency -- matters as much as raw supply. Variable renewable sources introduce exactly the kind of intermittency that AI workloads cannot absorb. Nuclear power, operating at capacity factors above ninety percent with essentially zero output variance, is the ideal technical match for what data centres actually need from the grid.
The Investment Case Has Three Legs -- And All Three Are Strengthening
The first leg is contracted revenue. Long-duration power purchase agreements with creditworthy counterparties -- the hyperscalers -- have resolved the merchant risk problem that made nuclear financing impossible for a decade. The second leg is policy support. The US Inflation Reduction Act's production tax credits for nuclear generation, extended in 2025 legislation, provide a floor under operating economics for existing plants and improve the return profile for new builds. European governments, from France to Poland to the Czech Republic, have moved nuclear back to the centre of their energy policy frameworks. The third leg is technology development support. Small modular reactor programmes are receiving government co-investment across multiple jurisdictions, reducing the technology risk premium that investors must price.
These three legs are not independent. Government policy makes contracted revenue more attractive by reducing political risk. Contracted revenue makes private investment viable. Private investment funds technology development. The virtuous cycle that nuclear could never establish during its difficult decade is now building momentum, driven primarily by the economic necessity of meeting data centre demand rather than by any change in public opinion about nuclear power itself.
Existing Plants Are the Immediate Opportunity
New nuclear capacity takes a decade or more to build. The near-term investment opportunity is concentrated in operating existing plants longer and more profitably. Fleet operators in the United States have seen their capacity factors and market prices rise simultaneously -- a rare combination. Units that were marginal during periods of cheap gas are now running at full output because grid managers need their baseload characteristics and because wholesale electricity prices have risen in markets with heavy data centre penetration.
Plant life extensions are proceeding across the US fleet. The Nuclear Regulatory Commission has approved multiple sixty-to-eighty year licence extensions, each requiring capital investment in safety systems and component replacement. That capital is now flowing because the revenue outlook justifies the spend. In Europe, France's decision to extend the operating lives of its existing reactor fleet represents a capital deployment programme in the tens of billions, all predicated on the same demand-driven logic.
For investors, operating nuclear plant owners offer the most immediate exposure to the nuclear renaissance. These companies generate contracted, zero-carbon electricity at sites with existing NRC or national regulator approval -- the most valuable asset in electricity markets where data centre operators are competing intensely for reliable power. Plant operators including Constellation Energy, Vistra and Exelon in the US have seen their equity valuations re-rate significantly as the market has begun to price this dynamic correctly.
Small Modular Reactors Are the Medium-Term Story
Small modular reactors represent the next wave of nuclear investment, targeting first commercial operation in the early-to-mid 2030s. The SMR proposition is factory fabrication of standardised components, shop assembly rather than site construction, and deployment at data centre campuses rather than grid-scale utility sites. If the fabrication and cost assumptions hold, SMRs would allow data centre operators to own and operate their own nuclear generation on-site, eliminating the grid connection problem entirely.
The qualification is significant: the fabrication and cost assumptions have not yet been validated at commercial scale. First-of-kind SMR projects currently under development -- NuScale, X-energy, Kairos Power in the US; Rolls-Royce SMR in the UK; GE-Hitachi in Canada -- are all at early stages. Cost overruns and schedule delays in nuclear construction are historically common. Investors pricing SMR developers should apply a substantial technology and execution discount relative to the optimistic projections those companies publish.
That said, the offtake demand for SMR output is demonstrably real. Microsoft's agreement with Constellation to restart Three Mile Island Unit 1, and its subsequent announcement of interest in SMR power, illustrates the direction. When the largest corporate electricity buyer in history is willing to sign twenty-year agreements to buy nuclear power, the revenue side of SMR business cases is validated even before the technology side is. That asymmetry -- validated demand, uncertain supply cost -- is actually a positive signal for patient capital. The market will reward the developers who solve the cost problem because the buyers are already waiting.
Top 10 Companies in This Space
- Constellation Energy -- largest US nuclear fleet operator, primary beneficiary of data centre offtake demand and life extension economics
- Vistra Corp -- operates nuclear and thermal generation, actively repositioning fleet toward contracted zero-carbon output
- Exelon Corporation -- regulated utility with substantial nuclear generation exposure in PJM and ComEd service territories
- EDF (Electricite de France) -- operates the world's largest nuclear fleet, driving France's life-extension and new-build programme
- NuScale Power -- first SMR developer to receive NRC design certification, pursuing modular reactor deployment
- X-energy -- advanced pebble-bed SMR developer with DOE funding and Microsoft partnership commitments
- Kairos Power -- molten salt SMR developer, Google has signed agreements to purchase its output
- Rolls-Royce SMR -- UK government-backed small modular reactor developer targeting European deployment
- Cameco Corporation -- world's largest publicly traded uranium producer, primary input supplier for nuclear renaissance
- BWX Technologies -- nuclear components manufacturer serving both defence and commercial nuclear sectors
Our Take
Nuclear power's rehabilitation is genuine and durable because it is driven by physics rather than policy preference. AI compute needs baseload power that does not fluctuate, does not emit carbon, and does not depend on weather. Nuclear is the only source that satisfies all three conditions at scale. The investment case that has been rebuilt by data centre demand is more robust than any previous nuclear investment cycle because it is underwritten by the strongest corporate balance sheets in history rather than by government guarantees alone. The near-term opportunity is in operating fleet owners who are already capturing the revenue uplift. The medium-term opportunity is in the SMR developers who can solve the cost problem -- but with eyes open to the historical difficulty of executing nuclear construction on budget and schedule. Investors who want nuclear exposure without construction risk should focus on plant operators and uranium producers first, reserving SMR developer exposure for a small portion of a broader clean energy portfolio. -- Elena Vasquez, MarketsNXT