Crédit Mutuel AM: Will AI spark a surge in electricity costs?

January 7, 2026

By Océane Balbinot-Viale, Gabrielle Capron and Céline Zanella, Analysts - Financial and Extra-Financial Research division, Crédit Mutuel Asset Management

Generative artificial intelligence is not just about data. Its running requires a massive amount of electricity. Behind its impressive capabilities lie energy-intensive infrastructures, powered by data centers functioning non-stop, whose growth is skyrocketing.

A hyperscale data center with a 100 MW capacity, running at full power, consumes as much electricity annually as approximately 100,000 households. This high energy demand stems from the extensive use of power-hungry graphics processors and by continuous cooling systems.

On a global scale, electricity demand from data centers is expected to more than double, rising from 460 TWh in 2024 to over 1,000 TWh by 2030 and reaching 1,300 TWh by 2035, according to the AIE. In this context, US tech giants are investing at an unprecedented pace: up to $300 billion projected for AI in 2025, 20% more than the total investment in the US energy sector.

Santa Clara’s November shutdowns reveal a growing risk: insufficient power to meet demand. To address these challenges, the energy mix will need a profound transformation. According to the IEA’s latest World Energy Outlook, by 2035 renewables are expected to become the dominant energy source, driven by their competitiveness and rapid deployment.

Their share in the global mix is expected to increase by around 10 percentage points compared to today, while fossil fuels are expected to decline from nearly 80% to 71%. However, this transition brings major challenges: grid adaptation and the deployment of storage solutions, such as batteries or pumped-storage facilities, are essential to offset the intermittency of wind, solar and hydro power.

Current gas and nuclear capacities will remain strategic for flexibility, but significant constraints weigh on new installations: gas is currently facing shortages among equipment suppliers, while nuclear and small modular reactors (SMR) provide low-carbon energy but require long lead times – 10 to 15 years for conventional nuclear – with the first SMR expected in China in 2026, not before 2028 for GE-Hitachi in Canada and after 2030 in the United States. Meanwhile, the commercial viability of SMRs remains uncertain. Coal is likely to remain dominant in Asia in the medium term.

Energy realities: different geographies, different challenges

The impacts of this race vary significantly across regions, both in terms of demand, with China and the United States expected to account for nearly 80% of global growth in data center electricity consumption by 2030, and in terms of market structures and electricity supply availability, which remain highly heterogeneous.

• In the United States, growth is explosive, with an increase estimated at over 500 TWh by 2030, while grid tensions are already visible. Data center capacity could account for between 30% and 65% of new electricity demand depending on the scenario. Meeting these needs and replacing aging plants will require 85–90 GW by 2030. Of this, about 25 GW could come from gas (subject to capacity constraints), implying 60 GW from renewables or 200 GW at 30% utilization, which is in line with the current pace of 40 GW per year. Even in the U.S., the energy mix is shifting toward renewables, with 55% low-carbon energy expected by 2035 versus 44% in 2024. Energy shortages are cynically feared in states like Texas, where public contracts have namely excluded “anti-fossil” players.
• In Europe, the increase in demand from data centers, estimated at between 115 TWh and 130 TWh, is more moderate due to stricter regulation (AI Act) and energy-saving measures following the war in Ukraine. With the recent development of renewables, Europe has a production surplus, with fossil capacities largely underutilized. Europe should also be able to cover the entire incremental demand through additional renewable capacity, projected to reach 710 TWh by 2030. Concerns focus on grid sizing and peak loads, especially when renewable supply is unavailable and these peaks saturate planned capacities in gas, hydropower, nuclear and biomass. In this respect, regulations are evolving to encourage grid expansion by offering better returns on investment. The existence of surplus fossil capacities will continue to play a key role during periods of grid stress.
• In the APAC region, the dynamic is the most aggressive, though data center consumption in China and India remains marginal. The rise of AI and the cloud are driving the expansion of data centers in Asia-Pacific, fueled by low-cost open-source models (e.g., DeepSeek early 2025) and digital sovereignty policies. According to the IEA, installed capacity should more than double, from 36 GW in 2024 to 92 GW in 2030 (CAGR 20%). This growth is causing a rapid increase in electricity consumption, up 15% to 20% per year, reaching 300 TWh in 2030 compared to 150 TWh in 2024. Required investments are estimated between $90 and $110 billion, with 60% in renewables and 40% in fossil fuels, excluding grid infrastructure. Regional disparities are significant: in massive power systems like those of China and India, dominated by heavy industry and large populations, data centers will remain marginal, accounting for no more than 2% of total consumption in 2030 (vs. 0.5% to 1% in 2024). However, the energy challenge remains critical: in a region where the electricity mix is still dominated by coal (nearly 70% in 2024 and still over 55% in 2035); coal remains the main option to secure supply, despite the rise of renewables and nuclear.

AI’s hidden cost: the race for energy efficiency

The explosion in energy demand raises a social question: who will pay the bill? In the United States, retail prices have already risen by an average of 6% in 2025, with spectacular increases like +19% in New Jersey.

Wholesale prices are expected to reach $51/MWh in 2026 (vs. $47/MWh on average in 2025), largely attributed to strong data center demand. In Europe, the increase could be contained thanks to the growth of renewables, but pressure on peak loads remains strong.

If infrastructure does not keep up, households and SMEs could bear a disproportionate share of the cost of supplying large energy consumers, exacerbating social tensions around the energy transition.

This issue goes beyond price: Texas recently passed legislation to allow the grid operator to limit data center consumption (starting at 75 MW) during critical periods to guarantee household access to electricity, illustrating the growing intervention of the state to regulate what the market cannot ensure: resource availability, whether energy or water.

Amid these challenges, efficiency becomes a strategic priority. Improving PUE (Power Usage Effectiveness) from 1.35 to 1.20 would reduce consumption by 8% to 12% by 2030, saving hundreds of TWh.

Levers include advanced cooling systems, optimized data center design and hybrid microgrids combining solar, batteries and modular gas turbines. Grid digitalization via AI promotes predictive maintenance, load forecasting and renewable integration.

Operators can also secure supplies through PPAs (Power Purchase Agreements) to guarantee long-term green electricity. Ultimately, the availability of decarbonized electricity will become a key criterion for data center location, even if grid reliability remains the top priority.

AI: beyond the energy debate

The rise of AI is not just about who will pay the bill. It questions our energy models, sovereignty and ability to reconcile innovation and transition. Behind the race for computing power lie multiple considerations: environmental, with increasing water consumption for data center cooling and a worrying carbon footprint; social, with job transformation, elimination of certain functions and emergence of new skills; and governance-related, where debates on ethics, system resilience and stakeholder responsibility become central.

For Europe, the challenge is twofold: maintain digital competitiveness while reducing fossil fuel dependence. More than ever, we need a sovereign model, capable of combining digital performance and energy autonomy. While AI promises efficiency gains and major innovations, it also raises systemic challenges that go far beyond the energy bill. Are we ready to face them?