Artificial intelligence should be integrated into a nuclear power plant project during pre-project activities and project development, well before the first concreting date (FCD).
In the approach proposed here, AI is embedded in an integrated project model and works simultaneously across project scope and requirements, engineering and design, licensing, schedule and cost, procurement and the supply chain, risk management, contracting, stakeholder and organizational interfaces, configuration and change management, project information and records, construction preparation, and commissioning readiness.
Starting this early follows from the structure of an NPP project itself. Pre-project activities and project development already include much of the work required to define the project and prepare it for contracting and construction. By the time FCD is reached, many decisions affecting the future project trajectory have already been made.
The project model is not a single AI application or model. It is a structured, machine-readable representation of project objects, activities, requirements, documents, quantitative attributes, and their relationships, together with a set of computational and AI-enabled tools that can operate on this common project structure. These tools may include verified mathematical models invoked by AI, model ensembles, AI agents, and specialized machine-learning models developed for particular project tasks. Large language models may also be used where appropriate, but they are only one component of the architecture, not the model on which the entire system depends. Together, these components can support forecasting, risk analysis, readiness assessment, consistency checking, optimization, and decision support.
I discussed this model in more detail here: Formalization of a Nuclear Power Plant Life-Cycle Model for AI and Machine Learning Applications.
AI technologies still have unresolved limitations. For nuclear projects, the main concerns are reliability, validation, explainability, data quality, cybersecurity, governance, and regulatory acceptance. I do not examine these issues in detail here.
At the same time, some aspects of the proposed approach may seem ambitious given what existing technology can realistically support. NPP projects, however, typically involve long development and implementation periods, while AI capabilities are advancing rapidly. Capabilities that are immature or difficult to implement today may become practical within the next several years.
An AI-enabled project model introduced early should therefore be designed to evolve rather than remain fixed at the technological level available at the time of implementation. Its underlying models, computational tools, agents, and AI components can be progressively improved as the project develops and new technologies become available. By the time FCD is reached, some parts of the model may already operate on a substantially more advanced technological basis than was available when implementation began.
This creates an additional challenge for the traditionally conservative nuclear industry. The project model must remain sufficiently stable, controlled, and traceable to support long-term project processes. At the same time, it must be capable of incorporating significant advances in AI and computing over the course of the project.
Artificial intelligence implementation should begin during front-end planning
The early development phase of a nuclear power plant project establishes much of the project’s fundamental definition.
Front-end planning (FEP), also called front-end loading (FEL) or pre-project planning, includes the activities undertaken at the start of a project to ensure that it is sufficiently defined, planned, and estimated to support informed decisions on whether the project is viable and ready to proceed. It also includes identifying residual risks and establishing appropriate risk responses.
Front-end planning therefore provides the natural starting point for an AI-enabled project model.
At this stage, the model can integrate project objectives and requirements with the proposed technical solution, site conditions, licensing requirements and strategy, project scope, contracting strategy, funding and financing arrangements, major procurement decisions, schedule assumptions, and identified risks.
The objective is to develop a structured project representation as the project definition evolves.
As the project progresses, this representation can be refined, extended, and populated with project-specific information rather than reconstructed later from disconnected documents, datasets, and information systems.
I discussed this topic in more detail here: Recommendations for Implementing Artificial Intelligence Technologies in Nuclear Power Plant Construction Processes.
The model should support project integration from the beginning
In an NPP project, integration is a core project management function. It requires the processes and systems needed to keep the project organization aligned and to coordinate activities, interfaces, information flows, responsibilities, and decisions across the project.
An integrated AI-enabled project model can support this function directly.
Engineering decisions can be linked to requirements and licensing commitments. Procurement activities can be connected to design maturity, manufacturing lead times, supplier interfaces, and required delivery dates at site. Schedule activities can be linked to resources, contracts, cost, and other project dependencies. Risks can be associated with the project elements, activities, organizations, and interfaces through which their consequences may propagate. Changes can be traced across engineering, procurement, construction, and commissioning.
The model's value comes from both the project information it contains and the relationships among project elements and their associated information.
This is particularly important for an NPP project, where responsibilities and interfaces are distributed across multiple organizations, including the owner/operator, regulatory body, NPP vendor, architect-engineer or design organizations, equipment suppliers, construction contractors, subcontractors, and commissioning organizations.
The AI-enabled project model should be developed together with the project management system
The project management system must be in place from an early stage. It provides an integrated framework for managing project processes, responsibilities, organizational interfaces, information flows, and decision-making, and can be refined as the project develops.
From the outset, the AI-enabled project model should form part of the same project environment. It should represent project processes, requirements, responsibilities, interfaces, information flows, and the history of key decisions.
A process-based structure is particularly suitable because it organizes project activities through defined processes, inputs, outputs, responsibilities, interfaces, and performance criteria.
For AI, this provides more than access to documents and datasets. It gives AI the structure needed to interpret information and understand how project elements are interconnected.
Reliable AI-supported analysis also depends on clear provenance, data quality, version control, and identification of the authoritative sources for project information.
Requirements, configuration, and changes should enter the model from the beginning
From the earliest stages, the model should link project requirements with design solutions, configuration information, and supporting documentation.
This provides traceability from each requirement to its implementation and supporting evidence. If a change is proposed, the model can show which project elements are affected and how the change could influence design, licensing, contracts, schedule, cost, construction, and commissioning.
Starting early also means that configuration and change history are captured as the project develops rather than reconstructed later.
Design maturity is a precondition for effective construction
Effective project execution requires sufficient design maturity before FCD.
The design should be sufficiently mature, licensing issues relevant to proceeding to FCD should be resolved, and adequate project pre-planning should be completed.
An integrated project model can make design maturity more measurable.
It can monitor the completeness of requirements, design packages, supplier inputs, interface information, approvals, configuration status, and supporting documentation. These data can then be linked to schedule activities, procurement packages, and the construction work that depends on them.
This allows design maturity to be treated as a dynamic project characteristic rather than as a single evaluation carried out immediately before FCD.
It also helps show how unresolved engineering issues may affect procurement, licensing, construction readiness, schedule performance, and subsequent commissioning activities.
Licensing should be integrated into the model before FCD
Licensing depends on the completeness and consistency of requirements, design solutions, supporting analyses, and evidence.
An AI-enabled project model can link regulatory requirements to design decisions, justifications, documents, and approvals, helping identify gaps and dependencies before they delay licensing.
Early integration is important because unresolved licensing issues can directly affect design maturity, construction readiness, and the project schedule.
The project schedule begins long before FCD
The first concreting date is an important project milestone, but the schedule logic that determines subsequent construction performance begins much earlier.
NPP project schedules include preparatory work before FCD as well as construction activities after FCD. Feasibility studies, design, licensing, and procurement begin years before FCD, and major equipment procurement may also begin well before this milestone.
The AI-enabled project model should therefore contain the project schedule while that schedule is still being developed.
Schedule activities can be connected to engineering deliverables, licensing actions, procurement packages, manufacturing activities, site preparation, construction work packages, resources, contracts, and commissioning requirements.
The baseline schedule can then be assessed not only as a sequence of activities but as part of the wider project structure. Dependencies, constraints, assumptions, resource requirements, and uncertainty can be evaluated before the schedule becomes the principal reference for project performance.
This is particularly important because decisions made during early planning can create constraints that later determine the critical and subcritical paths of the project.
Risk management should start when the project is first considered
Integrated life cycle risk management must begin at an early stage.
Risk management should begin when a new project is first considered and continue throughout its life cycle. Some risks originate in decisions made before the formal project begins, yet their consequences may emerge much later as impacts on cost, schedule, quality, and ultimately the NPP's life cycle cost.
An AI-enabled project model can incorporate risk from this early stage.
Risks can be linked to the assumptions, activities, organizations, contracts, systems, interfaces, and external conditions that generate them. Quantitative attributes can then be used to estimate their effects on schedule and cost.
I discussed this approach in more detail here: Mathematical Optimization and Forecasting of Nuclear Power Plant Construction Progress using Probabilistic Models and Machine Learning Methods.
The project risk profile can be updated continuously as assumptions change and new project data become available.
Risk management thus becomes part of the same project representation used for planning, engineering, procurement, and project decisions.
This also moves risk analysis beyond a stand-alone risk register. Each risk can be examined together with the specific project objects, activities, interfaces, decisions, and dependencies that determine how it may affect project performance.
Cost and financing should be linked to the project model early
Schedule, risk, and cost are closely connected in an NPP project, and their interaction begins well before FCD.
An AI-enabled project model can link schedule uncertainty, risk exposure, expected expenditures, and financing requirements, allowing the financial consequences of delays and alternative scenarios to be assessed as the project develops.
Early integration matters because major financing and investment decisions are made before many project uncertainties are fully resolved.
Contract strategy should be defined early
The contracting model affects how responsibilities, interfaces, risks, and financial commitments are distributed across the project. It should therefore be defined while the project organization, financing structure, design maturity, and major uncertainties are still being established.
An AI-enabled project model can connect contract scope and responsibilities with schedule, cost, risk allocation, change control, and financing. This is especially important before the final investment decision, when the contract and financing strategies need to be consistent with each other.
Long-lead equipment makes early integration essential
NPP procurement provides another strong reason for early implementation.
Major equipment, including reactor vessels, steam generators, reactor coolant pumps, and turbine generators, may require lead times of 60 months or more. Procurement planning must therefore account for supplier capabilities, manufacturing cycles, and the extensive interfaces between suppliers and design organizations.
An integrated AI-enabled project model can connect each major procurement package with its design requirements, technical specifications, supplier information, and manufacturing milestones. The model can then track the package through documentation and inspections, transportation and site delivery, installation, and commissioning.
This preserves the connection between engineering, procurement, manufacturing, and subsequent project execution.
The model can also support probabilistic forecasting of procurement and manufacturing milestones and help identify equipment whose delay could affect the critical path or subsequent commissioning activities.
Because these procurement commitments are made well before much of the construction work on site, late implementation would mean that some of the most important schedule and supply-chain dependencies had already been established before the model began operating.
Construction production capacity should be planned before FCD
The construction strategy should define early how fabrication and assembly work will be distributed between the site, the site construction base, and external manufacturing facilities.
An AI-enabled project model can link expected construction volumes with available production capacity, productivity, logistics, construction methods, and the schedule. This can help determine whether the planned production system can supply structures, assemblies, and other construction outputs at the rate required by the project.
These decisions need to be made before FCD because they affect the required construction base, industrial facilities, fabrication strategy, and the organization of construction flows.
Early implementation creates project-specific knowledge
At the outset, the model can use information from reference plants and previous projects, along with relevant regulatory requirements and standards, historical performance data, and generic project structures.
As the project progresses, it accumulates project-specific information, including actual regulatory review durations, design evolution, supplier performance, contract conditions, schedule data, cost information, identified risks, key decisions, and corrective actions.
Over time, this becomes a project-specific knowledge base.
Lessons learned from previous projects, together with experience gained on the current project, can then be used to improve project performance.
An AI-enabled project model can retain this knowledge in a structured form and apply it to subsequent decisions.
The model should become more valuable as the project advances. It starts with reference information and assumptions; actual project data gradually replace or supplement them.
By the time major approval gates are reached, the model can contain both the current project state and the history that led to it.
Approval gates should use the accumulated project model
NPP projects are commonly divided into phases separated by approval gates. Before each gate, the project organization reviews progress, project status, development gaps and risks, continued feasibility, the detailed plan for the next phase, and the required budget and resources. Approval gates may be associated with project development, early preparation, and the final investment decision, while proceeding to FCD requires a dedicated assessment of overall project readiness.
An AI-enabled project model can use the accumulated project state to support these decisions throughout project development.
It can assess readiness across multiple areas, identify unresolved dependencies, quantify remaining uncertainty, and evaluate the consequences of proceeding under different assumptions.
This is particularly important before the final investment decision.
At this stage, major manufacturing and construction commitments are normally made, while project expenditure and the financial consequences of delay increase rapidly. Design maturity, the construction license, and other factors that may affect the critical path therefore need to be sufficiently confirmed and managed.
The model can support decisions at each approval gate by preserving the accumulated project state, its assumptions, unresolved issues, identified risks, and the rationale for key decisions.
The benefit of early implementation is clearest at this stage. A model introduced only shortly before an approval gate would have to reconstruct much of the project's earlier history. A model developed alongside the project already contains that history.
Construction readiness should be assessed before FCD
Before proceeding to FCD, the project should be checked for readiness across the main areas that can affect construction. These include design and licensing status, documentation, procurement and contracts, site preparation, schedule, financing, and risk.
An AI-enabled project model can bring these areas together, identify unresolved dependencies, and estimate the consequences of moving to FCD while gaps or incomplete work remain.
This gives decision-makers an integrated view of project readiness before proceeding to FCD.
The same project model should continue into commissioning
The move from construction into commissioning depends on the readiness of systems, rooms, documentation, equipment, and work fronts.
If the same AI-enabled project model is used at this stage, construction completion can be checked against system turnover requirements and commissioning readiness criteria. This makes it easier to identify incomplete work or missing information before these gaps delay commissioning.
Keeping the same model in use during commissioning also keeps construction progress and commissioning readiness connected within the same project structure.
The model's value increases across a series of projects
The benefits of early implementation become especially significant when multiple NPP units or projects are developed as a series.
An AI-enabled project model allows actual data, lessons learned, risks, durations, costs, and performance from earlier units to improve forecasting and decision support for subsequent ones. Each new project can then build on the accumulated experience of preceding projects rather than starting primarily from generic reference information.
These benefits are strongest when projects share a common structure, semantics, data definitions, and reference configuration, allowing experience to transfer consistently across projects.
Across the series, the model becomes increasingly informed by actual project performance as it accumulates project-specific data.
AI becomes part of the project architecture
AI should therefore be implemented from the earliest stages of an NPP project.
During pre-project activities, the model can support project definition, feasibility assessment, requirements development, and early risk analysis.
During project development, it can integrate engineering, siting, licensing, financing, procurement strategy, contracting, schedule development, and the management of organizational interfaces.
During contracting and construction, the same model can use actual project data to support forecasting, risk analysis, configuration and change management, procurement control, and construction planning.
During commissioning, it can support system readiness, turnover, completion status, and the transition to operation.
This continuity is consistent with a process-based management approach in which relevant processes extend across the NPP life cycle and safety, health, environmental, security, quality, and economic considerations are integrated from the beginning of the project.
The central point, therefore, is not simply to introduce AI early.
The AI-enabled project model should develop together with the NPP project, beginning with front-end planning and project definition, accumulating requirements, interfaces, decisions, risks, and actual performance data, and remaining connected to the project management system throughout construction and commissioning.
Early in the project, many of the most consequential decisions are still open. They have not yet been locked into approved designs, contracts, organizational arrangements, manufactured equipment, construction work, or financial commitments. That room for change narrows as the project advances.
This is why the earliest project phases offer the greatest opportunity for an integrated AI-enabled project model to influence the future project trajectory rather than merely analyze its consequences.
Its value is therefore not limited to predicting what may happen later. When implemented early enough, the model can help improve project definition, strengthen planning and decision-making, identify emerging constraints before they become fixed, and increase readiness for each subsequent project phase.