AI and big data in energy projects: operational leverage and the European market

The central paradox of AI applied to energy

This session, delivered on 4 November 2025 at the Silk Road Business School to a partner delegation, starts from a contradiction few presentations take on: artificial intelligence is at once the main problem and the main solution of the energy transition.

On the problem side, data centre electricity consumption is expected to move from 460 TWh to 945 TWh by 2030, and could represent around 5% of total European electricity consumption within six years. On the solution side, AI has become indispensable to managing millions of decentralised assets, with the big data market in energy projected at $33.6 billion by 2033.

The European market for AI in energy shows the same tension: around $2,592 million in 2024, a forecast of $12,062 million in 2030, a compound annual growth rate of 29.3%.

The PMO moves from reporting to decision-making

The first application is not technical, it is organisational. A PMO that consolidates progress reports produces information about the past. A PMO instrumented with AI produces alerts about the future: predicted cost drift, critical path slippage, contractual exposure by work package.

That shift changes the profile of project control teams more than it changes the tools. It assumes clean, time-stamped project data attached to a stable breakdown structure. Without that, AI simply accelerates the production of wrong information.

Digital twin: comparing as-built to as-planned

The construction digital twin is the most profitable building block in the short term. A continuously updated virtual replica does two things: it compares executed work against plan on real geometries, and it serves as a contract execution platform by making claim-triggering events factual.

On large projects, most disputes are not about law, they are about reconstructing the facts. A well-maintained digital twin moves the debate onto time-stamped data, which shortens negotiations and reduces provisions.

What AI changes, technology by technology

In solar and wind, the contribution concentrates on generation forecasting and battery optimisation, with documented gains of 15% to 20% on storage system lifetime. In green hydrogen, the issue is electrolysis process optimisation: current efficiencies sit between 60% and 80%, and the objective of passing 90% acts directly on the main cost item, the renewable electricity consumed. In smart grids, AI supports real-time balancing and predictive maintenance, with a measurable French example: Enedis's CartoLine BT tool reaches 95% accuracy between suspected anomalies and actual faults.

AI Act and NIS2: governance before tooling

Part of the session addressed the European framework, because the regulatory calendar conditions the deployment trajectory. The AI Act classifies by risk level and places critical infrastructure in the high-risk category, which imposes lifecycle risk management, data set quality, technical documentation and human oversight. The NIS2 directive extends cybersecurity obligations to essential and important entities, energy operators included.

The operational message is simple: on an energy asset, compliance is not a layer added after the pilot, it is a design constraint that determines which architectures can actually be deployed.

Architects versus specialists: the real structure of the market

The second part mapped the players. On one side the architects, established industrial groups selling scale: Schneider Electric, which capitalises on both sides of the paradox and deploys around 2,400 sustainability consulting experts; Siemens, pushing an open digital commercial platform; Enel, using AI internally at the scale of a large utility.

On the other side the specialists, selling depth: ACCURE Battery Intelligence on battery diagnostics, with $16 million raised in a Series B in February 2025; METRON in France on industrial energy optimisation, €12.5 million raised in November 2024; Gradyent in the Netherlands on digital twins for district heating networks, €28 million in a Series B.

The observable dynamic is convergence rather than elimination, as the partnership between Cleanwatts and ABB in June 2024 illustrates.

Four recommendations

Treat project data as an asset before buying a model. Prioritise use cases where measurement is possible, and therefore contractable. Choose a position in the chain, architect or specialist, rather than claiming both. And build the AI Act and NIS2 into the design phase, because late compliance costs more than the functionality itself.