For a long time, the race for artificial intelligence was associated mainly with processors, language models and software. Today, however, it is becoming increasingly clear that the real bottleneck is the energy system. With the rapid construction of large data centres, access to cheap electricity is becoming a key factor determining the scale and pace of this boom. Companies linked to AI are trying to secure access to low-cost power. One way to do this is by building modular SMR reactors. Similar reactors are also expected to appear in Poland within the next few years.
The workloads generated by artificial intelligence require enormous amounts of electricity. It is estimated that running ChatGPT alone may currently consume as much as 40 million kWh per day. This means that ChatGPT’s annual energy consumption exceeds the total electricity consumption of 117 countries with the lowest levels of use. Training and operating large models requires not only vast amounts of energy, but also stable supply and high-quality voltage parameters. This is because data centres are filled with sensitive equipment and expensive processors that must be protected against voltage fluctuations.
This is why grid infrastructure and dedicated in-house energy sources are becoming a new strategic battleground. Electricity is beginning to serve as the backbone of artificial intelligence development. According to a report by the International Energy Agency, data centre energy demand rose by 17% in 2025, while demand from AI-specialised data centres grew even faster. This pace was significantly higher than the increase in global electricity demand, which stood at 3%.
In their effort to secure stable energy supplies for power-intensive data centres and AI models, Microsoft and Google are turning to nuclear energy. Microsoft is involved in the restart of the Three Mile Island Unit 1 reactor, while Amazon is investing in partnerships aimed at deploying small modular nuclear reactors. This trend also applies to Poland, where small modular reactors are increasingly being seen as a complement to large-scale nuclear power. The most advanced projects are being developed by Orlen and Synthos, based on BWRX-300 technology, but this is not the only path. Other proposals also remain in play, including British Rolls-Royce SMR reactors, which Industria Nuclear Solutions is working on. Small-scale nuclear power is ceasing to be a vision of the future and is becoming a real foundation for large industrial energy consumers.
The way utilities and energy infrastructure companies are perceived is also changing. They are less and less often seen solely as part of traditional energy. Increasingly, they are becoming a key foundation of the digital economy. Siemens Energy is a good example, supplying transformers and switchgear needed to connect data centres to high-voltage grids. At the same time, the company’s gas turbines can provide local power generation, working together with batteries and replacing traditional diesel generators. This allows data centres to expand partly independently of the existing grid. In a situation where Microsoft, Google and Meta are expanding data centres almost at any cost, companies supplying critical energy infrastructure are gaining a very strong negotiating position.
Alongside Siemens Energy, a new group of power producers is emerging as major beneficiaries of the AI race. In the United States, NextEra Energy is using its portfolio of wind, solar and energy storage projects to sign long-term contracts with cloud and AI companies. At the same time, the company is not limiting itself to renewables. It is also developing gas projects and supporting the restart of the Duane Arnold nuclear power plant, which is intended to help supply Google’s data centres. Brookfield Renewable has taken a similar direction, securing energy supplies for Microsoft. These agreements show that demand generated by AI is beginning to translate directly into cash flows for energy and infrastructure companies.
The strain on power grids extends far beyond data centres themselves. The accelerated adoption of electric vehicles is creating a second wave of rising demand, as millions of charging points and increasingly fast charging infrastructure consume enormous amounts of energy. This creates additional pressure on grids that were never designed for such a level of demand. At the same time, geopolitical tensions in recent years, from the war in Ukraine to the blockade of the Strait of Hormuz, have made the strategic importance of energy security impossible to ignore.
The result is a structural, multi-year shift in electricity demand that goes far beyond a single technology cycle. AI data centres, electric vehicle fleets and the broad electrification of industry are all placing pressure on the same power grid at the same time, while the companies building, modernising and supplying that grid sit at the intersection of all these trends. In this context, the biggest bottleneck in the current AI boom may not be chips, but the ability to build and control the power systems that supply the entire digital world.
