Something unexpected has happened in energy technology: nuclear power is cool again. Microsoft signed a deal to restart Three Mile Island. Amazon, Google, and Meta are all exploring nuclear power for data centers. Small modular reactor startups are attracting serious venture capital.

This isn’t nostalgia. It’s a response to a specific problem: where does the electricity come from for AI?

The Energy Problem AI Creates

Training large language models consumes enormous amounts of electricity. Running inference at scale - serving millions of queries - requires more. Data center electricity demand is growing faster than grid capacity in many regions.

The numbers are striking. Training GPT-4 reportedly consumed around 50 GWh of electricity. The AI industry’s total electricity consumption could reach 100 TWh annually by 2027, according to some estimates. That’s comparable to the entire electricity consumption of a country like the Netherlands.

Where does this electricity come from? The traditional answer - the grid - is running into limits. In some regions, utilities are refusing new data center connections because they can’t guarantee supply.

Renewables are part of the solution, but they have intermittency problems. A data center needs power 24/7, not just when the sun shines or wind blows. Battery storage helps but isn’t yet economical at the scale needed.

This is where nuclear becomes interesting again.

Why Nuclear Now

Nuclear power has characteristics that make it attractive for data center loads:

Carbon-free and reliable. Nuclear operates continuously (90%+ capacity factors), providing baseload power without carbon emissions. For companies with climate commitments and continuous power needs, this is appealing.

Energy density. A nuclear plant produces enormous power from a small footprint. In a world where land and grid connections are constrained, density matters.

Independence from weather. Unlike solar and wind, nuclear output doesn’t vary with weather conditions.

The historical objections to nuclear - cost overruns, construction delays, public opposition, waste management - haven’t disappeared. But the calculus is changing.

If your alternative is telling customers your AI service is unavailable because the grid can’t supply power, nuclear’s disadvantages look different.

The SMR Bet

Small Modular Reactors are the technology bet attracting the most investor attention.

Traditional nuclear plants are massive - gigawatt-scale facilities that take a decade to build and cost $10+ billion. SMRs promise smaller units (50-300 MW) that can be factory-manufactured and assembled on-site, theoretically reducing costs and construction times.

NuScale has the first SMR design certified by the US NRC. Their project in Idaho was cancelled due to cost escalation, but the certification remains. TerraPower, backed by Bill Gates, is developing a different SMR design with novel coolant technology. X-energy, Kairos Power, and others are pursuing various approaches.

The honest assessment: SMRs haven’t yet proven they can deliver on cost promises. The theory is that manufacturing scale will reduce costs, but that scale doesn’t exist yet. We’re in the early phase where costs are still high.

What makes the current moment different is the existence of customers (tech companies) with deep pockets and urgent needs who are willing to be early adopters. That wasn’t true for previous nuclear efforts.

The Tech Company Approach

Microsoft’s Three Mile Island deal is instructive. They’re not building a new reactor - they’re restarting an existing one that was shut down for economic reasons. This is faster and lower-risk than new construction.

Google and Amazon are taking smaller positions, signing purchase agreements and exploring partnerships without committing to specific plants yet.

The common thread: de-risking through contracts. If a nuclear operator has a 20-year power purchase agreement with a creditworthy tech company, financing becomes much easier. The tech companies are essentially underwriting nuclear development with their power commitments.

This is a model that could work more broadly. Corporate demand for clean, reliable power could fund nuclear capacity that wouldn’t get built based on wholesale market prices alone.

Fusion: The Longer Horizon

While fission nuclear is getting practical attention, fusion continues to attract speculative investment.

Commonwealth Fusion Systems has raised over $2 billion and is building a demonstration plant. TAE Technologies, Helion Energy, and others are pursuing different fusion approaches.

The honest take on fusion: it’s still speculative. No fusion reactor has achieved net energy gain in a way that could scale commercially. The phrase “fusion is always 30 years away” exists for a reason.

But - and this matters for technology investors - the private fusion companies are taking different approaches than the traditional government mega-projects. They’re moving faster, iterating more aggressively, and setting nearer-term milestones.

If I were building a long-term portfolio in energy tech, I’d have some exposure to fusion. Not because I’m confident it’ll work, but because the payoff if it does is transformative.

What Innovation Managers Should Know

If you’re not in energy, why does this matter?

Electricity costs and availability may become constraints on AI deployment. If your AI strategy depends on scaling compute, understand where that compute gets its power.

Nuclear renaissance affects investment landscape. Energy tech is attracting capital that might otherwise go elsewhere. Understand how this shifts the broader innovation ecosystem.

Regulatory and policy changes are coming. Nuclear permitting reform is bipartisan in the US. New frameworks for SMR licensing are being developed globally. These create opportunities and change the competitive landscape.

Corporate energy strategy is becoming more sophisticated. Large companies are increasingly going beyond “buy renewable energy credits” to direct involvement in power generation. This might become relevant for more organizations than currently expect it.

The Investment Lens

For investors and innovation managers thinking about energy tech:

Near-term opportunities: Companies enabling nuclear restart and life extension of existing plants. Grid integration and management technology. Power purchase agreement structuring.

Medium-term opportunities: SMR supply chain. Nuclear-specialized engineering and construction. Technology for advanced fuel cycles.

Speculative opportunities: Fusion development. Advanced reactor concepts. Nuclear applications beyond electricity (industrial heat, hydrogen production).

Risks to monitor: Construction cost and timeline overruns (the historical pattern). Regulatory delays. Waste management policy. Public acceptance, which varies enormously by region.

The nuclear renaissance isn’t guaranteed. But the drivers are real - climate commitments, AI energy demand, grid constraints. For the first time in decades, nuclear has motivated buyers with resources to spend.

That changes the game, even if the game plays out slowly.