Advanced Nuclear Funding Surge: Where AI Fits Next

Nuclear fission companies raised $1.3 billion in equity funding by early Q3 2025, the highest annual total on record—and nearly 40% of all fission equity since 2020. That number matters less as a brag and more as a signal: investors are no longer paying for slide decks. They’re paying for manufacturing plans, licensing progress, fuel contracts, and demonstration dates.

Three December announcements—Radiant (more than $300M), Last Energy (more than $100M Series C, oversubscribed), and ARC Clean Technology (closed Series B)—show the same pattern: advanced nuclear is shifting from “interesting designs” to “assets that must run.” And the moment you commit to real operations—whether it’s a microreactor at a remote site or a fleet of SMRs supporting data centers—AI stops being optional.

In the AI in Energy & Utilities series, we usually talk about AI for grid optimization, demand forecasting, and predictive maintenance. Advanced nuclear is quickly becoming one of the best test beds for those capabilities because it’s being built with modern sensors, modular manufacturing, and digital-first operations. The developers that treat AI as part of the plant—not a dashboard bolted on later—will be the ones that hit schedule, availability, and cost targets.

Why 2025 capital is flowing into SMRs and microreactors

Answer first: money is moving because advanced nuclear developers are showing credible pathways to deployment, not just innovation.

Net Zero Insights reported that SMRs and microreactors captured roughly 75% of total nuclear fission funding in 2025. That split makes sense: large reactors still struggle with long timelines and complex site-specific construction. Factory-built systems flip that equation by pushing more work into repeatable manufacturing.

What’s different about this wave of financings is the “proof points” investors are looking for:

• Regulatory progress (design reviews, agreements, pilot pathways)
• Fuel strategy (especially HALEU access for certain designs)
• Factory and supply chain plans (real facilities, not placeholders)
• Early customers (even if initial deals are small, deposits matter)

I’m opinionated on this: advanced nuclear won’t win because it’s elegant. It will win if it’s operationally boring—predictable outages, stable heat rates, repeatable commissioning, and high availability. That’s exactly where AI-driven operations earns its keep.

The hidden driver: financiers now underwrite “execution risk”

Venture capital exists to fund high-risk technology, but these rounds are increasingly about execution risk rather than “will the physics work.” Once a company announces a test in 2026 or targets early customer deployments in 2028, the questions change:

• Can you manufacture modules at volume with quality control?
• Can you staff operations and maintenance efficiently?
• Can you integrate into local grids without curtailment or penalties?

These are data problems. And data problems are AI problems—if you build the plumbing early.

Three developers, three deployment models—and three AI opportunities

Answer first: Radiant, Last Energy, and ARC are pursuing different reactor sizes and markets, but they share a need for AI in reliability, grid integration, and fleet operations.

Radiant: transportable microreactors that must “just work”

Radiant raised more than $300M in mid-December to move from demonstration toward manufacturing—specifically preparing to break ground on its R-50 factory in Oak Ridge, Tennessee. Radiant’s Kaleidos is a 1 MWe helium-cooled microreactor using TRISO fuel, designed for rapid deployment to defense, disaster response, remote industry, and critical infrastructure.

Radiant also outlined a tight set of milestones: test its first reactor in 2026 at Idaho National Laboratory’s DOME facility, and target initial deployments beginning 2028. It’s also stacking commercialization signals: a deal (with deposits) for 20 microreactors for data center applications and agreements for a U.S. military base deployment.

Where AI fits: microreactors won’t have the staffing model of a traditional plant. That creates a strong push toward autonomous operations within strict safety boundaries.

Practical AI applications that map well to microreactors:

• Predictive maintenance for balance-of-plant equipment (pumps, heat exchangers, valves, generators) using vibration, thermal, and electrical signatures
• Anomaly detection tuned to “small system” physics, where small deviations can matter more because margins are tighter
• Remote operations support with AI-assisted procedures, diagnostics, and parts forecasting (especially for remote sites)

The business point is blunt: if a microreactor requires a large on-site crew or frequent specialist visits, the economics collapse.

Last Energy: modular PWRs betting on speed and repeatability

Last Energy closed an oversubscribed Series C of more than $100M and plans to use proceeds to complete its PWR-5 pilot at Texas A&M–RELLIS under the DOE’s Reactor Pilot Program, targeting criticality in 2026.

Its commercial product, the PWR-20 (20 MWe), takes a different approach than many “novel” SMRs: it leans on established pressurized water reactor experience—standard full-length fuel and familiar architecture—while pushing hard on factory fabrication, modularization, and a compact site footprint (about 0.3 acres). The company is also advancing UK licensing for a four-reactor project in South Wales.

Where AI fits: once your pitch is “we can deliver in under 24 months,” schedule becomes a technical requirement.

High-impact AI use cases for a modular PWR program:

• AI-driven construction planning and modular logistics: constraint-aware scheduling, materials availability prediction, and quality documentation tracking
• Digital twin for commissioning: using operational data from the pilot to tune procedures and reduce rework for follow-on units
• Fleet learning: every unit should make the next unit faster to build and easier to run, with AI capturing patterns across sites

Most companies get fleet learning wrong. They treat each plant as a project. The winners treat plants like software releases: consistent baselines, controlled changes, and fast feedback loops.

ARC Clean Technology: a 100 MWe fast reactor with industrial heat upside

ARC closed a Series B to accelerate commercialization of its ARC-100 (100 MWe / 286 MWth), a sodium-cooled fast reactor derived from EBR-II operating heritage. The design targets a 20-year core life without refueling and a 60-year plant operating life.

ARC also logged meaningful regulatory progress in Canada, completing Phase 2 of a Vendor Design Review in 2025 and supporting site licensing work for a proposed demonstration at Point Lepreau. It’s also participating in U.S. DOE programs and building partnerships for deployment in North America.

Where AI fits: ARC is aiming at industrial heat, hydrogen production, and data center applications—markets where value depends on high capacity factor and tight integration with adjacent processes.

AI opportunities that matter in a 100 MWe class plant:

• Advanced process control for combined heat-and-power configurations (heat delivery is often more complex than electricity export)
• Market-aware dispatch optimization: bidding strategies, ramp coordination with renewables, and minimizing wear from cycling
• Condition-based maintenance using multi-modal data (thermal, acoustic, sodium system instrumentation) to reduce forced outages

AI isn’t a “nice-to-have” for advanced nuclear—it’s how you hit the business case

Answer first: advanced nuclear will live or die on availability, staffing, and integration costs, and AI directly reduces all three.

If you’re a utility, an IPP, a campus energy operator, or a data center developer considering SMRs or microreactors, you’re not buying a reactor. You’re buying an outcome: 24/7 power, predictable pricing, and operational resilience.

AI contributes to that outcome in three concrete ways.

1) Higher availability through predictive maintenance

Forced outages are expensive for any generator, but they’re especially punishing for early fleets because they damage bankability. Predictive maintenance models (paired with a rigorous reliability program) can reduce unplanned downtime by:

• Detecting early bearing wear, insulation breakdown, or valve drift
• Prioritizing maintenance windows based on risk, not calendars
• Improving spare-parts strategy (what to stock, where, and when)

The nuance: AI doesn’t replace nuclear-grade engineering. It makes engineering decisions faster and more consistent.

2) Better grid and load integration (especially with renewables)

As more wind and solar come online, the grid needs resources that can provide:

• Firm capacity
• Frequency support and voltage control
• Predictable ramp behavior

SMRs and microreactors can fill part of that gap, but only if they’re integrated with modern grid operations. AI helps by forecasting demand and renewables output and then optimizing dispatch strategies that respect plant constraints.

For behind-the-meter deployments (data centers, campuses, industrial parks), AI can also coordinate:

• UPS and battery systems
• Demand response programs
• Heat recovery and thermal storage

3) Lower operating cost via digital-first staffing models

A common misconception is that nuclear’s cost challenge is mostly concrete and steel. For smaller reactors, operations staffing can dominate lifecycle economics.

AI-enabled operations centers—combined with strong procedures and cybersecurity—support:

• Centralized monitoring for multi-site fleets
• Automated reporting and compliance data workflows
• Faster troubleshooting with curated historical cases

If a microreactor fleet needs traditional staffing at every site, it won’t scale. Full stop.

What energy and utility leaders should do in 2026 planning cycles

Answer first: treat AI readiness as a procurement requirement for advanced nuclear, not an IT project.

With demonstrations targeted in 2026 and early commercial deployments discussed for 2028 and beyond, 2026 planning cycles are where buyers can shape outcomes. Here’s what I’d put on a practical checklist.

A buyer’s AI readiness checklist for SMRs and microreactors

1. Data architecture commitment: confirm sensor strategy, historian design, and data ownership terms (especially with vendor-operated models).
2. Digital twin roadmap: insist on a plan that starts at commissioning, not after commercial operation.
3. Reliability program maturity: ask how AI insights translate into maintenance work orders and engineering change management.
4. Grid integration plan: verify interconnection studies, power quality approach, and controls integration with EMS/DERMS.
5. Cybersecurity-by-design: require segmentation, monitoring, and incident response integration across OT/IT.

Questions that quickly reveal whether a project is real

• “What operational data will you generate on day one, and who can access it?”
• “How will lessons learned from Unit 1 change the build and commissioning of Unit 2?”
• “What’s your strategy for remote operations support at scale?”

If the answers are vague, expect surprises later.

Where this goes next: the most valuable nuclear “product” may be the operating system

Capital is flowing because the advanced nuclear sector is finally building toward deployment. That’s the encouraging part. The harder part is what comes after first criticality: turning one-off demonstrations into fleets that perform.

My take: the developers that win won’t be the ones with the flashiest reactor renderings. They’ll be the ones that pair solid engineering with an AI-driven operating model—predictive maintenance, grid optimization, and fleet learning—so performance improves every quarter instead of resetting at every site.

If you’re mapping your 2026–2028 energy strategy—whether for utility planning, industrial decarbonization, or data center growth—now is the time to evaluate advanced nuclear alongside your AI in energy roadmap. The question to carry into next year’s planning meetings is simple: when the first fleets arrive, will your grid and operations teams be ready to run them like modern infrastructure?