Powering the future: AI, Data Centres, and the challenge of electricity supply

Key Takeaways:

• Electricity demand from AI data centres in the West is projected to increase 31-fold by 2035, potentially absorbing up to 20–25% of total annual electricity generation in the United States.

• Structural increases in electricity prices have already driven U.S. power costs up by around 50% since 2021, undermining competitiveness and the viability of new data-centre investments.

• China has doubled the United States’ installed power capacity over the past decade, benefiting from an abundant and affordable electricity supply that positions it to lead in AI development.

• The construction timelines for new power plants—up to ten years for combined-cycle facilities—are misaligned with the rapid expansion of data centres, which typically come online within one to two years, creating significant bottlenecks.

• In Spain and other European countries, governments are prioritising electricity rationing over capacity expansion, favouring “green” consumption and restricting access for data centres.

Artificial Intelligence, Data Centres, and the Challenge of Electricity Supply

Over the next decade, electricity demand is expected to rise sharply across developed economies. The primary driver of this increase is the rapid expansion of data centres dedicated to the training and inference of artificial intelligence models. Data centres are highly electricity-intensive for two principal reasons. First, the graphics processing units (GPUs) employ requires substantial power to perform complex, high-speed computations. Second, the operation of this hardware generates significant heat, necessitating extensive cooling systems that further increase electricity consumption.

The scale of electricity demand associated with these data centres is therefore considerable. If current investment projections prove accurate, the resulting increase in electricity consumption will be substantial and remains insufficiently appreciated. In terms of installed capacity, existing data centres supporting AI training and inference currently require approximately 4 GW of power. Recent consultancy estimates indicate that by 2035—within the next decade—AI data centres will require around 123 GW of installed capacity, representing a 31-fold increase relative to current levels.

By way of comparison, total installed capacity in the United States electricity system currently stands at approximately 1.2 TW. On this basis, AI data centres would account for close to 10% of today’s total installed capacity by 2035. When measured in terms of electricity generation rather than capacity, the proportion absorbed by data centres may be higher still, given that these facilities typically operate continuously throughout the year. Their annual electricity consumption could therefore exceed 1,000 TWh. By contrast, total annual electricity generation in the United States presently amounts to just over 4,000 TWh.

Absent a substantial expansion of generation capacity, data centres could thus absorb approximately 20–25% of total annual electricity generation in the United States. Similar constraints would apply in Europe. Under such conditions, the ability to scale computing capacity for AI training and inference would be materially constrained, placing these economies at a disadvantage in the global development of artificial intelligence.

One likely consequence of insufficient capacity expansion would be a sustained increase in electricity prices. Rising prices would reduce electricity consumption by households and by firms outside the AI sector, reallocating available supply towards data centres. However, higher electricity costs would also undermine the economic viability of many prospective data-centre investments, preventing projects that would otherwise be viable under lower price conditions from proceeding. This dynamic would further weaken the competitive position of Western economies in the global AI landscape.

Evidence of upward pressure on electricity prices is already apparent in the United States. Between 2016 and 2021, average electricity prices remained in the range of 13–14 cents per kilowatt hour. Since 2021, prices have increased to around 20 cents per kilowatt-hour, representing an increase of nearly 50% over a short period. While part of this rise reflects the inflationary shocks of 2021–2023, another component appears to be structural, linked to sustained increases in electricity demand from data centres.

Constraints on electricity supply in Western economies are, to a significant extent, the result of policy and regulatory choices rather than physical limitations. Over the past decade,  China has added more than 2 TW of new electricity-generation capacity, bringing its total installed capacity to approximately 3.6 TW. In comparative terms, this expansion is equivalent to nearly doubling the total capacity the United States has accumulated over its history. The resulting abundance and affordability of electricity constitutes a significant competitive advantage for China in the development of artificial intelligence, as it faces fewer immediate supply bottlenecks for the construction and operation of data centres.

That said, electricity-supply constraints also reflect timing mismatches between demand growth and capacity expansion. Data centres can typically be constructed and commissioned within one to two years, whereas most power-generation facilities require considerably longer lead times. Combined-cycle gas plants, for example—despite gas being an abundant resource in the United States—can take up to ten years to plan, permit, and build. As a result, plants not already approved are unlikely to become operational before 2035, coinciding with the projected 31-fold increase in data-centre power requirements.

Renewable energy sources such as solar photovoltaic and wind power, as well as electricity-storage systems, can be deployed more rapidly. Indeed, in the United States, approximately 92% of new electricity demand from data centres in 2025 has been met through the addition of renewable generation. Nevertheless, renewables alone are unlikely to be sufficient to meet the projected exponential growth in electricity demand over the coming decade. A diversified generation mix will be required to ensure system reliability and adequacy.

Failure to expand electricity-generation capacity at scale would have broader economic consequences. Insufficient supply would constrain economic growth and technological innovation, limiting future productivity gains and improvements in living standards.

Developments in Spain illustrate these dynamics. While there is strong private-sector demand to invest in data centres, electricity-generation capacity has not expanded at a comparable pace. In this context, policy responses have increasingly focused on managing demand rather than expanding supply. Where electricity demand substantially exceeds available supply and price-based allocation is politically constrained, rationing mechanisms become the principal means of allocation, prioritising certain uses while restricting others.

This approach is reflected in a ministerial resolution dated 11 July 2025 concerning access-capacity tenders at specific nodes of the transmission network. The resolution acknowledges the strategic importance of data centres for digital transformation and artificial-intelligence deployment, while emphasising that their base-load demand would increase electricity-generation requirements and, at times, necessitate support from thermal generation. It further concludes that, given the associated emissions implications, priority should be given to other forms of electricity consumption that more directly contribute to emissions reduction.

In summary, the capacity of Western economies to generate sufficient electricity is likely to constitute one of the principal bottlenecks in the development of artificial intelligence over the next decade. Unless electricity-generation capacity is expanded in line with demand, constraints on power supply will limit investment, innovation, and competitiveness. The choice therefore lies between enabling the necessary investment in electricity systems or accepting a gradual erosion of economic and technological leadership.