The Physics Is Fine. The Bottleneck Is… Everything Else.

The world needs roughly 300 gigawatts of new nuclear capacity. The United States is building 0.5 gigawatts per year. The target is 13.

That gap is not an engineering problem. The reactor designs exist. The safety record is established. France built 56 reactors in 15 years. The United States built more than 100 over roughly 25 years. The knowledge of how to do this at volume is not lost.

We cannot build hundreds of large reactors at sovereign scale without a stepwise transformation in all nuclear work paper-to-steel, in an industry that is sorely understaffed by an aging or missing nuclear workforce.

Enter Everstar.

Our AI platform Gordian is built to scale up the industry en masse: reactor safety, available sites, ready suppliers, and a new generation of nuclear workforce.

The Bottleneck Is the Workload

To build nuclear at volume, four things have to happen in parallel: reactor designs have to be standardized, construction has to run in series, a domestic supply chain and workforce have to be developed, and the regulatory environment has to support volume without compromising safety.

Every one of those tracks runs on the same underlying problem. Too much specialized knowledge. Too few people who hold it. Too little time to apply it at the scale the industry needs.

Darryl Willis, Microsoft's CVP for Energy and Resources, put it plainly: "Nuclear energy is the essential backbone for this future, but the industry remains trapped in a delivery bottleneck… That is where AI comes in."

Everstar is where AI comes in.

Consider what that looks like in practice. A license application chapter that a team of expert engineers estimated at over 200 person-days of work. Gordian, our AI platform for the full paper-to-steel nuclear workload, completed it in one day. National laboratory experts reviewed the output and confirmed its quality. That result came out of the DOE Genesis Mission, a collaboration with Idaho National Laboratory and Argonne National Laboratory.

One day versus over 200 person-days. That is not an incremental improvement. It is a different category of what is possible.

But permitting is one layer. The bottlenecks run through the entire deployment cycle.

The Full Stack of Nuclear AI

Reasoning

Nuclear work runs on dense technical documents. The challenge is not reading them. It is reasoning across them: understanding which regulation governs which situation, how a change in one document affects obligations in another, what the regulatory history says about how a given question has been resolved.

That same capability reaches into the supply chain. Building nuclear at volume requires qualifying thousands of suppliers to nuclear's exacting standards. Each qualification involves reviewing quality manuals, auditing procedures, and identifying gaps against a complex set of requirements. Today that process is slow enough that it constrains how fast new projects can move, or suppliers give up and leave the industry entirely.

A reasoning model can screen a supplier's documentation, identify the gaps, and produce a qualification roadmap in hours rather than weeks. One of our manufacturing customers needed a gap analysis to revise its quality standards. Using Gordian, they compressed weeks of work into hours of human review.

The same model can onboard new engineers, answer technical questions on shift, and help operators work through corrective action programs. The nuclear workforce is aging. AI can capture institutional knowledge before it walks out the door and train the next generation faster than any classroom program.

Vision

Nuclear facilities are full of schematics: piping and instrumentation diagrams, electrical circuit diagrams, isometric drawings. Engineers need to understand how components connect, how failures propagate, and what the downstream safety implications are. Vision models can identify components, map connections, and trace failure paths across a two-dimensional diagram. Gordian applies this to drawing reviews and specification comparisons, where a missed connection can have consequences that do not surface until construction or commissioning.

Then there is the physical world. Non-destructive examination generates enormous volumes of image and signal data. Today, trained inspectors review that data manually, one record at a time. Vision models can analyze NDE data faster and more consistently, flagging anomalies that human review would miss and building a continuous record of component condition over time. The same capability applies on the construction site: tracking progress, verifying as-built against as-designed, catching deviations before they become rework.

Physics

Running a full reactor safety analysis today can take days of compute time, and most programs only run a handful of scenarios. Physics models can run orders of magnitude more, which means engineers can stress-test assumptions they currently have to take on faith.

Site selection is one application. Identifying a viable nuclear site means evaluating geology, hydrology, population density, grid interconnection, and regulatory jurisdiction across federal, state, local, and tribal requirements. That process today takes years. Running those analyses in parallel compresses the timeline and expands the number of sites that can be seriously evaluated.

The tools in this category are still developing. The gap between what is possible and what is deployed is widest here, which means the upside is also largest.

Where Human Judgment Belongs

AI is not a replacement for the decisions that matter most. Nuclear safety depends on the ability to anticipate events that fall outside normal operating experience: scenarios that are low-probability, high-consequence, and sometimes novel. AI can expand the number of scenarios analyzed. It finds patterns that manual review would miss. It cannot determine which risks are acceptable.

That judgment belongs to engineers, operators, and regulators. The goal is to stop burying them in work a machine can do faster and more consistently, so they can spend their time on the problems that actually require a human.

The Infrastructure Layer for Nuclear Deployment

The nuclear industry has scaled before. The knowledge of how to do it at volume exists. What has been missing is the capacity to execute at the speed the moment requires.

Everstar is building the intelligence layer for nuclear that will enhance, capture, systematize, and automate complex nuclear work from paper-to-steel.

Gordian will unlock the most exciting industry of this century.