Europe's AI ambition clashes with climate targets as energy demands surge

Europe's attempt to become a serious player in artificial intelligence is colliding directly with its legally binding climate commitments, creating a tension that fund managers and infrastructure investors say can no longer be papered over. The continent needs massive amounts of reliable electricity to power the data centres that AI requires, yet its regulatory framework was designed to push fossil fuels off the grid, not to accommodate a new source of baseload demand that is growing faster than renewable supply.

The energy bottleneck at the heart of Europe's AI dilemma

Globally, energy has become the primary constraint on building out AI infrastructure. In the United States, utilities are restarting fossil-fuel plants and signing deals with nuclear operators to meet the surge. Europe, by contrast, requires developers to disclose energy and water efficiency measures before projects can proceed, adding procedural steps that slow deployment. Dan Ives of Wedbush Securities described the moment as a fork in the road: the bloc can either "play in the future" or risk "missing a big part of this technology wave." He characterised the region as "anti-entrepreneur," a reputation that is already pushing European technology companies and startups toward the United States, the Middle East and Asia.

The numbers illustrate the scale of the problem. A single large language model training run can consume as much electricity as a small town. Hyperscalers, Microsoft, Google, Amazon and their European counterparts, are scouring the continent for sites with grid connections of 100 megawatts or more. Such connections are scarce in Germany, the Netherlands and Ireland, where moratoriums on new data centre connections have already been imposed in some regions. The European Commission acknowledges the challenge. A spokesperson told reporters the bloc "is fully prepared to seize these opportunities while safeguarding the stability and reliability of Europe's energy system," and pointed to a forthcoming roadmap for AI in the energy sector under the Apply AI Strategy. The Commission did not answer specific questions about whether sustainability legislation would be rolled back to accommodate the AI push.

Policy rollbacks signal a shift in priorities

Over the course of 2025, the EU has quietly diluted several flagship green measures. On 16 December, member states agreed to water down the effective ban on new combustion-engine cars from 2035, allowing a continued role for e-fuels and plug-in hybrids. A week earlier, on 9 December, the Council approved a one-year delay to the implementation of the second emissions trading system (ETS2) for buildings, road transport and small industries, a mechanism designed to extend carbon pricing beyond power generation and heavy industry. At the same meeting, ministers reaffirmed a legally binding target to cut net greenhouse gas emissions by 90% by 2040 compared with 1990 levels, a figure that many analysts consider unreachable without drastic acceleration.

Earlier in the year, the Corporate Sustainability Due Diligence Directive (CSDDD) and the Corporate Sustainability Reporting Directive (CSRD) were both narrowed in scope and their deadlines pushed back. The Sustainable Finance Disclosure Regulation (SFDR) is currently under review. Paul Jackson of Invesco sees a pattern. "You can see in the U.K. that we're already rowing back on some of our commitments," he said, "and Europe will likely follow suit." He argued that climate policy is one of the easiest items for legislators to deprioritise when economic conditions tighten and competing interests, such as industrial competitiveness and energy security, crowd the agenda.

The pragmatism argument: regulation revamp or retreat?

Not everyone views the changes as a retreat. Nick de la Forge, a general partner at Planet A Ventures which backs climate-technology startups, described the reshaping of directives as "quite pragmatic, and we think that's an improvement." He acknowledged the tension: "We are always at the edge of navigating into a position where it becomes so unattractive to be present in Europe that it doesn't make sense anymore. And on the other hand, a lot of the regulation is direly needed." In his view, the revisions to CSRD, CSDDD and SFDR represent a "healthy revamp" that reduces compliance burdens without abandoning the underlying objectives. The distinction matters: if the EU is merely streamlining reporting requirements, the climate architecture holds. If it is creating loopholes that allow emissions to persist, the 2040 target becomes a paper exercise.

Carbon credits and the reality of 'energy addition'

The December agreement on ETS2 included provision for carbon removal credits to count toward the 2040 reduction target. That decision reflects a broader shift: AI hyperscalers, while maintaining headline decarbonisation goals, are increasingly relying on carbon credits and renewable energy certificates (RECs) to claim green credentials. One credit represents the removal or avoidance of one metric tonne of carbon dioxide. Jim Wright, who manages the Premier Miton Global Infrastructure Income Fund, put it bluntly: "Because, in reality, they will use some gas, and they may even use some coal," referring to the actual composition of the grids where data centres plug in. The result, several investors noted, is an era of "energy addition rather than transition", new clean capacity is built, but fossil capacity is not retired at the same pace because total demand is rising too quickly.

This dynamic is welcomed by oil and gas executives, who have long argued that the world needs more energy of all kinds. For Europe, it creates a paradox: the more successful the AI build-out, the harder it becomes to close the remaining coal plants. Jags Walia of Van Lanschot Kempen warned that "electricity wise, we might not be able to afford to close down coal power plants, which is going to be a real headache for the energy transition and energy security as well." The UK grid is already coal-free; Europe's is not. Germany, Poland, the Czech Republic and Bulgaria still rely on lignite and hard coal for a meaningful share of generation. Data centres require constant, uninterrupted power, a profile that wind and solar cannot provide without massive storage or backup generation that does not yet exist at scale.

Energy security risks mount as demand outpaces supply

Beyond emissions, the surge in data centre demand raises immediate security-of-supply questions. Paul Jackson noted that the AI race "puts a lot more strain on our energy infrastructure, and as we have seen in recent years, we're not terribly resilient when it comes to that." He was referring to the 2021-22 energy crisis, when Russian gas cuts exposed the fragility of European grids and sent industrial power prices to record highs. Adding what amounts to a new baseload, data centres typically operate at 90% capacity factor or higher, onto a system that is simultaneously decarbonising and decentralising could make pricing more volatile and, in extreme scenarios, force rationing. The European Commission's forthcoming AI-energy roadmap is expected to address grid reinforcement, demand-side flexibility and the role of small modular reactors, but none of those solutions will materialise before the early 2030s at the earliest.

Climate risk remains the elephant in the room

For all the focus on competitiveness, the physical risks of climate change have not disappeared. Kokou Agbo Bloua, global head of research at Société Générale, called it "a massive elephant in the room" and one of his biggest worries looking forward. Speaking on 22 December, he said: "We're sort of toast … pun intended, actually, because we're on the path of two-and-a-half, three degrees [of warming above pre-industrial levels]. And if you look at green technologies, [they're] being used for data centres, as opposed to replacing fossil fuels." His point is that every gigawatt of renewable capacity absorbed by a new data centre is a gigawatt not displacing gas or coal elsewhere. The net effect on global emissions depends on whether the AI workload itself generates efficiency gains, optimising grids, accelerating materials discovery, improving building energy management, that outweigh its direct consumption. That calculus remains unproven at scale.

The timeline for a formal abandonment of climate targets may be longer than the market assumes. Jags Walia noted that "sometimes on sustainability goals, what countries do is, if they are going to walk away from a goal, they try to leave it till the last minute." The 2040 target is enshrined in the European Climate Law; changing it would require a new legislative proposal from the Commission, approval by the Council and the European Parliament, a process that takes years. In the meantime, the gap between the trajectory implied by current policies and the legal target will widen, creating legal risk for the EU institutions and political risk for governments that have made the green transition a central pillar of their legitimacy.