‘Sovereign AI’ in energy does not fit neatly into one box. It is not just a technology question. It is not just a cyber question. It is not just a data question. And it is certainly not something that can be left entirely to either the IT function or the security function. It is becoming a board-level question about control, resilience, capability and trust.

A year or two ago, some of this might have sounded theoretical. Sovereign AI was a phrase people used in strategy documents and conference panels. Important, yes, but still slightly abstract. That has changed.

We are now moving very quickly into a world where the computing power, the models, the data, the cloud infrastructure and the operating systems behind AI are becoming part of the critical infrastructure of modern economies. And the question that follows is a simple one: who actually controls them?

Making this about any one country, one company or one technology provider would miss the point. The point is that whoever controls the computing, the models and the data has leverage over everyone who relies on them. The country or organisation holding that leverage today may not be the one holding it in five years’ time.

That is a strategic issue for the economy, and an operational issue for energy, because energy is not a normal sector. Energy is the system that every other system depends on. If energy fails, everything else fails with it.

And at the same time, we are asking the energy system to do something extraordinarily difficult. We are asking it to decarbonise at pace, remain reliable and affordable, integrate much higher levels of variable renewable generation, and support the electrification of transport, heat and industry. And now we are asking it to power a sharp rise in digital infrastructure and AI.

AI sovereignty matters because whoever controls the computing, the models and the data has leverage over everyone who relies on them. The country or organisation holding that leverage today may not be the one holding it in five years’ time.
 

The numbers are striking. The latest Energy Institute Statistical Review of World Energy shows that electricity is becoming more central to the energy system. Electricity generation now represents around 19% of global total energy supply. In China that share has almost doubled since the turn of the century, reaching 24%. Transport is one of the clearest signs of that shift. Internal combustion engine vehicle sales have been falling since 2017, and in 2025 more than a quarter of new cars sold globally were electric. China remains the largest electric vehicle (EV) market, accounting for 60% of EVs sold worldwide.

And then there are data centres. For the first time this year, the Energy Institute Statistical Review of World Energy included data centre power demand as a dedicated table. Global electricity consumption for data centres in 2025 was 788TWh. Some 40% of that was in the US alone.

So, when we talk about AI we are not talking about something floating in the cloud, disconnected from the physical world; AI has a physical footprint. It has an energy footprint. It has a grid footprint. And, increasingly, it has a sovereignty footprint.

The energy sector is therefore on both sides of the AI equation. On one side, energy will be essential to enabling AI: powering the data centres, the networks, the computers and the cooling that advanced AI requires. On the other side, AI will become essential to running energy: forecasting demand, balancing grids, integrating renewables, improving maintenance, optimising assets, supporting safety decisions, identifying cyber threats and helping people make sense of systems that are becoming too complex to manage by traditional means alone.

Using AI in industry
AI can help us operate more safely, more efficiently and more intelligently. It can help limit downtime through predictive maintenance. It can support load balancing and fuel optimisation. It can help specialists carry out hazard assessments. It can help engineers, operators and decision-makers see patterns they might otherwise miss.

But, and this is the important point for today, applying AI well is not just about whether the algorithm works. It is about whether the whole system works. It covers technology, organisation, people, governance, skills, interfaces, accountability and culture. The human factor is not a soft issue here. It is central.

In high-risk industries, people do not disappear when automation increases. Often, their role becomes more difficult. They move from doing the task to monitoring the system. They are expected to know when to trust the machine, when to challenge it, when to intervene and when to take back control. That is not easy. Highly automated systems can put human operators in the position of watching rather than acting. If the system makes a poor decision, the person may have only seconds to understand what has happened and intervene effectively.

That raises difficult questions. How should functions be allocated between people and AI? How much authority should an AI system have?

Who is accountable when something goes wrong? And how do we design interfaces that allow people to understand, challenge and query AI outputs rather than simply accept them? These are not abstract design questions. In energy, they go to the heart of safety, resilience and public trust.

The question for energy is not ‘Should we use AI?’ We will use AI. In many areas we already are. The question is whether we use it deliberately, safely and in a way that respects sovereignty, or whether we drift into dependency before we have agreed what good looks like.

What does sovereign AI mean for us?
Sovereign AI is still more of a phrase than a standard. It can mean domestic supply. It can mean data residency. It can mean model assurance. It can mean regulatory control. It can mean supply chain resilience. It can mean operational independence. It can mean all of those things at once.

But if we cannot define it in a way that operators can apply, boards can govern, regulators can test and government can support, then it will remain a slogan. And slogans do not run infrastructure.

National strategy increasingly assumes some degree of sovereign capability, but many operational systems still sit on foundations controlled elsewhere. That may be manageable in some contexts. But when we are talking about the infrastructure that powers the country, we need to be much clearer about the risks we are accepting, the dependencies we are creating and the safeguards we expect.

The energy sector still has a window to shape this agenda on its own terms. But that window will not stay open indefinitely.

The Sovereign AI for UK Energy Security Forum was organised with Applied Computing and supported by Wipro and AWS.

• Further reading: ‘Why the energy transition increasingly depends on AI’. Artificial intelligence is driving up electricity demand through data centres and advanced computing. Yet speakers at All-Energy argued that it may also be essential for managing the increasingly complex, decentralised energy systems needed to achieve net zero.
• ‘How much energy does AI actually consume?’ There are serious concerns about the energy consumption of AI systems – in particular large language models (LLMs) such as ChatGPT. Find out how much energy they use to answer your queries.