Category: 3. Business

You don’t have permission to access “http://www.spglobal.com/energy/en/news-research/latest-news/electric-power/032726-ceraweek-in-the-case-for-clean-resources-energy-security-tops-energy-transition” on this server.

Reference #18.77257e68.1774659083.4d2a297a

https://errors.edgesuite.net/18.77257e68.1774659083.4d2a297a

The rapid expansion of data centres is emerging as a major challenge for energy policy. In January 2026, the Trump administration urged PJM Interconnection, the largest electrical grid operator in the US, to run an auction requiring tech firms to bid on 15-year contracts for new generation capacity. In March, the US administration convened executives from leading technology companies to sign the Ratepayer Protection Pledge, committing them to securing new generation capacity for their data centres and bearing the full cost of the infrastructure upgrades. Both measures appear to reflect political response to mounting pressure over energy affordability ahead of 2026 mid-term elections: US household electricity prices are 32% above their 2019 levels (IEA 2026), with even larger increases in data centre-intensive regions. 

The extraordinary rise of artificial intelligence (AI) has fuelled an extensive policy and academic debate on how the digital transition is reshaping economic growth (Carpinelli et al. 2026), the labour market (Giuntella et al. 2025, De Souza 2025), and the green transition (Bonfiglioli et al. 2025). At the heart of this transformation sit data centres – the physical backbone of the digital economy. Yet data centres are not only an economic asset but also an increasing source of pressure on electricity costs, both through the substantial investment required in new infrastructure (power plants, transmission lines, and grid upgrades) and through the upward pressure they exert on wholesale electricity prices. In this column, we document the scale and structure of the global data centre market and assess its energy footprint, combining facility-level information with subnational electricity consumption statistics for the US and Europe.

The anatomy of a booming market

Over the last 15 years, the global data centre market has evolved in both scale and structure. The number of active facilities worldwide has grown by more than 140% since 2010, exceeding 6,500 in Q2 2025 (Figure 1a). The US leads with around 1,600 sites, while Europe as a region hosts more than 1,900 facilities, concentrated mainly in Germany, France, and the UK.

More informative than the headcount, however, is the expansion in live IT capacity, i.e. the electrical power available to run computing equipment. Global IT capacity has surged by approximately 900% since 2010, reaching more than 55 GW by mid-2025 (Figure 1b). US dominance is even more pronounced here: the US accounts for roughly 50% of global IT capacity, compared with around 18% for Europe and 10% for China.

Figure 1a Number of data centres by region

Figure 1b IT capacity (MW) by region

Note: Authors’ elaborations on BNEF data; latest data available is 2025 Q2. Europe includes countries in the European Union, but also UK, Norway, and Switzerland.

Since the most computationally intensive workloads – above all AI model training – require very large IT capacity, the market has become increasingly concentrated in large facilities. Those above 10 MW account for less than 20% of global facilities but over 80% of global IT capacity, with the US hosting the vast majority (Figure 2a). Concentration is especially stark at the top of the distribution: a small group of US-based hyperscalers (AWS, Google, Meta, and Microsoft) together account for around 70% of global self-built IT capacity. The scale of individual facilities continues to grow. Sixteen sites worldwide now exceed 1 GW of theoretical IT capacity, and Meta’s data centre under construction in Louisiana is expected to reach around 2 GW, potentially expandable to 5 GW through a multi-phase build-out.

Historically, co-location facilities, where multiple clients rent space and infrastructure from a third-party operator, have represented the largest share of global IT capacity, reflecting the preference of most organisations for outsourcing their data centre needs (Figure 2b). Within this segment, wholesale data centres – where operators sell large volumes of capacity to multiple customers – account today for the largest share of global IT capacity (around 42%). More recently, self-built facilities, where companies own and operate their own infrastructure, have expanded rapidly, driven primarily by the massive investment programmes of large technology companies. Within the self-built segment, public cloud operators that own their own data centre infrastructure and rent out computing resources to external customers have provided the main impulse for IT capacity growth and now account for almost 23% of global IT capacity.

Figure 2a IT capacity (MW) of large data centres (>10 MW) by region

Figure 2b IT capacity (MW) of data centres by facility subcategory

Note: Authors’ elaborations on BNEF data; latest data available is 2025 Q2. In the top plot, C- stands for colocation (data centres operated by one party but where the services are leased by another), SB- for self-built (data centres constructed and operated by a company for its own use). Colocation subcategories refer to the nature of the customer, while self-built subcategories refer to the business of the operator.

Power hungry: The electricity footprint of data centres

Globally, data centre electricity consumption has grown by around 12% per year since 2017, more than four times faster than total electricity consumption (IEA 2025). To illustrate the scale, a 10 MW data centre consumes approximately 80 GWh annually, equivalent to the electricity use of 20,000 European households. Global estimates of electricity consumption range from 415 TWh (IEA 2025) to 450 TWh (Bloomberg 2026) and are expected to more than double by 2030, reaching as much as 1,260 TWh (4.4% of global electricity demand) in a lift-off scenario.

While electricity consumption from data centres remains relatively modest at the global level, certain regions host highly concentrated clusters of facilities that consume a significant share of local electricity supply. We combine data on country electricity consumption, geo-localisation of data centres, and their IT capacity to estimate electricity demand for the US and Europe.
In 2024, six US states had data centres accounting for about 10% or more of electricity supply, with Virginia leading at around 26% (Figure 3). In Europe, Ireland stands out as a clear outlier with data centres accounting for almost a quarter of national electricity demand. Among the largest European economies, the UK is the only one exceeding the 3% threshold, while the four largest EU area countries remain below 2% (Figure 4).

Figure 3 Share of electricity demand from data centres in US states

Note: Authors’ elaborations on BNEF and EIA data. All data refer to 2024, the latest available year for US states electricity consumption. For readability we do not report shares smaller than 1%.

Figure 4 Share of electricity demand from data centres in European countries

Note: Authors’ elaborations on BNEF and Ember data. All data refer to 2024, latest available year for European countries electricity consumption.

Beyond the share of local electricity supply, the concentration of data centres is associated with higher electricity costs. Figure 5 plots average retail electricity prices against data centre capacity across US areas. The correlation is positive and statistically significant. While this simple cross-state association does not establish a causal link, it is consistent with the hypothesis that concentrated data centre demand exerts upward pressure on local electricity prices.

Figure 5 Scatterplot of electricity prices and DC capacity by US Census areas

Source: Authors’ elaborations on BNEF and EIA data. r is the correlation coefficient between the two variables. Sample: 2001-2024.

To meet the massive electricity consumption of data centres, several technologies can be adopted, each with different performance, cost, and construction timelines, with solar PV and gas turbines being the most rapidly deployable options. Large hyperscalers are adopting diversified strategies to ensure stable electricity supply, including signing long-term power purchase agreements (PPAs) from both conventional and renewable sources, entering nuclear partnerships, and supporting the restart of inactive gas-fired plants.

Securing sufficient electricity generation is only one side of the challenge, as the expansion of data centre capacity also faces increasing risks from grid connection delays, particularly in regions of strong demand growth. Policymakers have adopted a range of approaches to address these challenges. Some countries, notably the US, are seeking to facilitate data centre grid connections as part of a broader strategy to support AI development, in exchange for (non-binding) commitments to develop autonomous generation capacity and pay for infrastructure upgrades. Others, such as Ireland, have imposed moratoria on new data centre connections in certain areas and require new facilities to provide their own generation and storage capacity. Similar restrictions, also reflecting land and water constraints, have been adopted in the Netherlands and Singapore.

Conclusions and policy implications

Data centres are the physical backbone of the digital economy, translating AI-related capital spending into computational capacity and, ultimately, into productive output. Our analysis highlights the remarkable scale and geographic concentration of this infrastructure and the energy tensions that come with it.

Three policy challenges stand out. The first is distributional. While the benefits of digital infrastructure accrue broadly, the costs – higher electricity prices, grid congestion, and network upgrades – fall disproportionately on households and businesses in host regions. As our subnational analysis shows, data centres already absorb a substantial share of local electricity supply in some areas, and political tensions over cost allocation are mounting.

The second challenge concerns the interaction between the digital and energy transitions. Under faster-growth scenarios, competition for electricity and grid access could intensify significantly, particularly if infrastructure investment fails to keep pace with AI-driven demand. The risk is that data centre expansion crowds out or delays decarbonization objectives, forcing difficult trade-offs between digital development and climate goals.

The third, and arguably most consequential for Europe, is strategic. Global IT capacity is concentrated both geographically and among a small number of US-based hyperscalers. Europe accounts for less than a fifth of global capacity, implying heavy reliance on foreign-controlled infrastructure for cloud services, frontier AI training, and data governance. As AI capabilities become increasingly central to economic productivity and national security, reducing these dependencies will require coordinated policies on digital infrastructure investment, grid planning, and technological sovereignty.

Authors’ note: The opinions expressed in this column are those of the authors and do not necessarily reflect the views of the Bank of Italy or the Eurosystem.   

References

Bonfiglioli, A, R Crinò, M Filomena, and G Gancia (2025), “Data, power and emissions: How AI’s growth may slow down the green transition”, VoxEU.org, 31 October.

Bloomberg (2026), “AI Data Centers Fuel Quicker Growth In Power Demand”

Carpinelli, L, F Natoli and M Taboga (2026), “From AI investment to GDP growth: An ecosystem view”, VoxEU.org, 9 February.

De Souza, G (2025), “Artificial intelligence in the office and the factory: Evidence from administrative software registry data”, VoxEU.org, 9 September.

Giuntella, O, L Stella and J Konig (2025), “Artificial intelligence and workers’ wellbeing: Lessons from Germany’s early experience”, VoxEU.org, 21 July.

IEA (2025), “Energy and AI”.

IEA (2026), “Electricity 2026 – Analysis and forecast to 2030”.