AI power demand revives fossil fuels even as renewables hit cost parity

For years the story in European power markets was simple: demand was flat, renewable capacity was rising, and the result was a price squeeze that made new wind and solar projects harder to finance. Then artificial intelligence arrived. The electricity appetite of data centres has upended the calculus, pulling investment back toward renewables but also handing a lifeline to the fossil fuel plants the continent had hoped to retire.

The integration bottleneck

Almost 50% of European electricity came from renewable sources in 2024, according to industry data, and the project pipeline for solar, onshore and offshore wind is substantial. The problem is no longer generation but integration. Peter Osbaldstone, research director for European power and renewables at Wood Mackenzie, told CNBC that variable supply is pressing on power markets in a way that depresses prices. "The problem we have now is integrating all of that variable supply into our power markets," he said. That pressure "undermines the economics of investments, which makes the whole process of supporting a decarbonized power mix more difficult, more expensive for governments to bear."

The International Energy Agency (IEA) still expects global renewable generation to jump 60% by 2030, reaching 45% of total output. Yet the same agency revised its 2025-2030 growth forecast downward by 5% compared with its 2024 outlook, reflecting what it calls changing sentiment and policy, largely from the United States. Europe's own pipeline remains strong, but the revenue certainty needed to finance it has eroded.

AI demand rewrites the demand curve

Data centres are the new baseload. Their requirement for continuous, high-density power has turned a stagnant European demand curve into a growth story. That should be good news for renewable developers: higher prices improve project economics. But the intermittency of wind and solar means they cannot alone guarantee the 24/7 supply that hyperscalers demand. The result, in the near term, is a reprieve for gas-fired generation.

Agate Freimane, a partner at venture capital firm Norrsken, put it bluntly on CNBC's Europe Early Edition on 8 January: "Redeploying fossil fuel energy is a 'short‑term crutch' that helps the AI roll-out get going, but 'renewable energy is the only way to win in the long term.'" In a follow-up email she added that China and the United States have both acknowledged the need for vast energy resources to power an AI future, and that renewable prices have dropped more than 90% globally. In 2024, 91% of new renewable projects were cheaper than fossil alternatives.

The self-reinforcing cost cycle

Freimane argues a feedback loop is now in motion. Cheaper clean power accelerates electrification; rising electrification boosts demand for storage and grid intelligence; those upgrades push the cost of clean energy even lower. In this framing, AI is not just a new load but an accelerator of the transition. The numbers support the first half of the loop: battery costs have fallen 90% in less than 15 years, according to a 2024 IEA report, and new chemistries for long-duration storage are under development.

The second half is less certain. Long-duration batteries face a utilisation problem. "If you've got that long-duration battery storage, its utilization in a typical year is going to be very low, because you're not going to have that many opportunities to really deploy that asset," Osbaldstone said, noting that deployment depends on weather patterns. There is also a price risk: as more batteries enter the grid, the arbitrage margin, buying low, selling high, may compress because more participants are chasing the same spreads. Faraco of Morningstar DBRS made the same point: "As you add more batteries to the grid, this arbitrage margin may compress because there will be more batteries buying electricity at the low price and more batteries selling electricity at the higher prices."

Gas stays, nuclear waits

For all the talk of batteries, the analysts converge on one conclusion: gas is not leaving soon. Faraco said gas is "impossible to phase out" for the time being, describing it as the most efficient and cleanest fossil fuel across the board. Wood Mackenzie expects gas, classified as a transition fuel by the European Union, to remain part of the energy mix as far ahead as 2060. "There is a call to be made at some point by governments: What do I do with gas generation?" Osbaldstone said. "But ultimately, the lights have got to stay on."

Nuclear is often proposed as the zero-carbon baseload partner for renewables. Faraco is sceptical of its flexibility: nuclear plants cannot be switched on and off to match variable renewable output. That leaves gas as the swing provider, able to ramp quickly when the wind drops and the sun sets. The implication is uncomfortable for climate policy: meeting AI-driven demand while cutting emissions requires either a breakthrough in long-duration storage or an acceptance that gas plants will run for decades longer than net-zero pathways assume.

AI as grid operator, not just consumer

The European Commission is betting that AI can help solve the very problems AI's electricity demand creates. A Commission spokesperson described what it calls the "twin potential of energy for AI and AI for energy." The bloc's forthcoming roadmap for digitalisation and AI in the energy sector aims to "accelerate the uptake of digitalisation and AI in the energy sector while improving energy efficiency and system reliability," the spokesperson said. "As AI rapidly advances, its potential to strengthen Europe's energy resilience and accelerate the clean transition is becoming increasingly clear. At the same time, the growing electricity needs of AI technologies call for smart, forward-looking planning."

An IEA report noted that AI-driven data analytics could improve planning, project design and real-time operational decisions, resulting in reduced fuel consumption, lower CO2 emissions and extended asset lifetimes. The Commission pointed to startups such as Vind AI, which applies machine learning to wind farm design; Granular Energy, which tracks hourly renewable generation for corporate buyers; and Juna.ai, which optimises industrial manufacturing processes. All three are Norrsken portfolio companies.

Heavy industry as the hidden lever

The biggest prize may not be data centres at all. Heavy industry accounts for around a third of global energy consumption, and Freimane argues that AI-driven optimisation is now pushing industrial efficiency forward by decades. If steel, cement and chemicals plants can cut their energy intensity significantly, the overall system pressure eases. That would reduce the need for both new renewable capacity and the gas plants that currently back them up.