AI + Advanced Nuclear: 2025 Funding Signals Lift-Off

By December 2025, advanced nuclear stopped being “interesting” and started being funded at scale. That shift matters more than any single reactor spec sheet.

A market intelligence snapshot put numbers on the momentum: $1.3B in equity funding for nuclear fission companies by early Q3 2025, the highest annual total on record, representing nearly 40% of all nuclear fission equity investment since 2020. Deal volume climbed too—28 equity transactions by October 2025 versus an average of ~15 in prior years. Most of the money followed a clear theme: SMRs and microreactors captured ~75% of funding.

For energy and utility leaders tracking AI in Energy & Utilities, this isn’t just a nuclear story. It’s a deployment story. And deployment changes the AI conversation from “research pilots” to production-grade operations: forecasting, grid integration, predictive maintenance, and reliability analytics that utilities already buy—now applied to a new class of assets.

Record investment is a signal: demos are getting close

The simplest interpretation of 2025’s record funding is this: investors believe first-of-a-kind demonstrations are close enough to underwrite manufacturing, licensing, and early commercial delivery.

Three December rounds illustrate the pattern:

• Radiant raised $300M+ (Series D) to scale microreactor manufacturing and build a factory in Oak Ridge, Tennessee.
• Last Energy closed an oversubscribed $100M+ (Series C) to complete a Texas A&M pilot reactor targeting criticality in 2026, while pushing a UK regulatory pathway.
• ARC Clean Technology closed a Series B to accelerate commercialization of its ARC-100 sodium-cooled SMR, building on Canadian regulatory milestones.

This matters because nuclear projects don’t fail only on physics—they fail on schedule risk, supply chain risk, operational readiness, and regulatory execution. Capital flowing into these developers is effectively a bet that they can manage those risks.

Why investors are showing up now

Two forces are colliding:

1. Firm, always-on power demand is spiking (data centers, electrification, industrial reshoring). Many grids are capacity-constrained at the worst possible time.
2. The industry is finally making a credible push toward factory-built, repeatable nuclear—less “mega-project,” more “product.”

I’m opinionated here: utilities should treat this as a new asset class that will compete on operational excellence, not just on cost of capital. That’s exactly where AI becomes a differentiator.

What these funding rounds really mean for utilities and large energy users

Funding “Series A/B/C/D” can sound like startup trivia. In advanced nuclear, it’s more practical than that: it’s a proxy for maturity and execution readiness.

• Series B money typically goes to building the machine: manufacturing processes, supply chain, and productization.
• Series C and D increasingly fund repeatability: certification, standardized deployment playbooks, and the first few customer deliveries.

For utilities, IPPs, and large industrials, that translates into a more concrete set of questions:

• When will this technology be operated outside a lab environment?
• What’s the licensing pathway, and how predictable is it?
• Can it be manufactured and serviced like equipment—not constructed like a one-off megaproject?

Now layer AI on top: Can the operator maintain high capacity factors with smaller crews, tighter supply chains, and more distributed sites? That’s where advanced analytics, predictive maintenance, and AI-driven operations start paying for themselves.

Three developers to watch—and where AI fits on day one

Radiant: microreactors as transportable “power products”

Radiant is developing Kaleidos, a 1 MWe / 1.9 MWth helium-cooled microreactor using TRISO fuel and passive cooling. The company’s timeline is specific: first reactor test in 2026 at Idaho National Laboratory’s DOME facility, and initial customer deployments beginning in 2028.

It’s not just reactor engineering—Radiant is also building an “everything else” stack: fuel agreements (HALEU), a manufacturing facility, and customer commitments (including a purchase agreement for 20 microreactors for data center applications).

Where AI fits immediately (even before full commercialization):

• Digital twin + anomaly detection during the 2026 test campaign: AI models trained on high-frequency sensor data can spot drift, instrumentation faults, and off-normal thermal-hydraulic signatures earlier than traditional alarm thresholds.
• Condition-based maintenance for factory throughput: once you’re building units in a factory, downtime is expensive in a different way. Predictive maintenance on CNC equipment, welding stations, QA imaging systems, and test rigs becomes a direct driver of cost and schedule.
• Site energy management for remote/critical infrastructure: microreactors are often paired with microgrids. AI-based demand forecasting and dispatch optimization can reduce required spinning reserves and improve resilience.

If you’re a utility or C&I buyer, the practical takeaway is this: microreactors will win early deals based on deployment repeatability and operational simplicity, and AI can make both real.

Last Energy: factory fabrication meets a familiar reactor archetype

Last Energy’s strategy is almost contrarian: build something that looks “new” in delivery model but “familiar” in core technology. Its commercial design is the PWR-20 (20 MWe) using standard PWR fuel (<4.95% enriched, 17×17 assemblies) and a four-loop pressurized-water design—leaning on the operating history of 300+ PWRs worldwide.

The near-term milestone is the PWR-5 pilot reactor at the Texas A&M–RELLIS Campus under a DOE pilot program pathway, with a target of criticality in 2026.

Where AI fits immediately:

• AI-driven work packaging and schedule risk prediction for modular builds: the fastest projects aren’t “lucky”—they’re managed tightly. ML models that learn from fabrication nonconformances, rework rates, and supplier lead-time variance can surface schedule slip weeks earlier.
• Smart commissioning analytics: during commissioning, you get a flood of transient data. AI can classify start-up signatures, flag sensor calibration issues, and detect subtle mismatches between expected and observed system response.
• Load-following optimization for behind-the-meter customers: many PWR concepts aim for steady output, but early buyers (data centers, industrial sites) care about power quality, ramp constraints, and integration with storage. AI-based control policies can coordinate reactor output limits with batteries and demand response.

If you’re evaluating SMRs for industrial campuses, Last Energy’s approach highlights a point many teams miss: standard components don’t guarantee standard outcomes. You still need operational analytics to achieve repeatable performance.

ARC Clean Technology: fast reactor heritage, industrial heat potential

ARC’s ARC-100 is a 100 MWe / 286 MWth sodium-cooled fast reactor derived from EBR-II, which operated for 30 years. The company completed Phase 2 of a Canadian Vendor Design Review in 2025—an important de-risking milestone because it signals no fundamental barriers have been identified at that stage.

ARC also targets non-electric applications—industrial heat, hydrogen production, and data centers—which is where AI-enabled optimization becomes even more valuable.

Where AI fits immediately:

• Heat integration optimization: industrial heat is all about pinch points, exchanger performance, and operational constraints. AI can optimize setpoints and heat flows to minimize curtailment and improve thermal utilization.
• Predictive maintenance for sodium systems: sodium fast reactors have different instrumentation and chemistry considerations than water reactors. AI models can correlate pump vibrations, temperature gradients, and chemistry signals to predict maintenance needs without excessive conservatism.
• Grid interconnection + congestion forecasting: a 100 MWe unit placed near constrained substations needs smarter dispatch planning. AI-based congestion and price forecasting helps operators decide when to run full power, when to produce hydrogen, and when to provide ancillary services.

My stance: the strongest early business case for larger SMRs won’t be “cheap electricity.” It will be high-value reliability + high-value heat, optimized with software.

The real bottleneck: operating a fleet, not proving a design

Demonstrations get headlines. Fleets pay the bills.

Advanced nuclear developers are moving toward multiple sites, smaller crews, and more standardized equipment. That creates a new operational challenge: you’re no longer optimizing one plant—you’re optimizing a distributed portfolio.

Here’s the operating model shift utilities should plan for:

• From periodic manual checks to continuous condition monitoring
• From fixed maintenance intervals to predictive maintenance
• From static operating procedures to data-driven performance management
• From local-only thinking to fleet-wide learning loops (what fails at Site A becomes a preventive fix at Sites B–F)

A practical AI blueprint for SMR/microreactor operators

If you’re building an AI roadmap for nuclear-adjacent assets (or considering it), focus on four layers:

1. Data foundation: historian strategy, sensor standards, time sync, cybersecurity segmentation, and a clear data ownership model.
2. Asset health: anomaly detection, remaining useful life estimation, and automated work order recommendations.
3. Operations optimization: demand forecasting, dispatch optimization, and microgrid coordination.
4. Regulatory-grade governance: model validation, audit trails, and change control.

A mistake I see often is starting with “cool models” and skipping governance. Nuclear operations—especially in early deployments—will be judged on traceability and discipline, not novelty.

What energy leaders should do in Q1 2026

These financings are December 2025 news, but the operational implication is a 2026 planning problem. If you’re in utilities, independent power, large C&I, or data centers, here are next steps that actually reduce risk:

1. Treat AI requirements as part of procurement. Ask vendors what data you get, at what resolution, and with what interfaces. If the answer is vague, assume you’ll pay later.
2. Build a grid integration playbook now. Interconnection queues and substation constraints are brutal. Use AI-based hosting capacity and congestion forecasting to choose better sites and reduce upgrade surprises.
3. Plan for fleet operations, even if you start with one unit. Standardize tags, alarm philosophy, and maintenance taxonomies from day one so lessons transfer.
4. Model the hybrid plant economics. Many projects will pencil out because they can switch between electrons, heat, and molecules. AI-driven forecasting is how you decide which product to make each hour.

One-liner worth keeping: The winners in advanced nuclear won’t just build reactors—they’ll run them like a software-informed fleet.

Where this fits in the “AI in Energy & Utilities” story

Most AI in Energy & Utilities work has focused on variable renewables, grid congestion, outage prediction, and utility customer operations. Advanced nuclear adds a new dimension: firm generation that still benefits from AI, especially when deployed as smaller, repeated units.

The best time to build the analytics backbone is before the first criticality milestone becomes a commercial operating problem. The funding in December 2025 says the market is getting serious about deployment. Utilities and large energy buyers should get equally serious about the operational layer.

If you’re evaluating SMRs or microreactors, start by mapping your AI priorities: predictive maintenance, demand forecasting, and grid integration optimization. Then pressure-test the vendor’s data and governance posture. The question that will matter most by 2028 isn’t “Can it produce power?”—it’s “Can we operate it predictably, repeatedly, and profitably at fleet scale?”