Many data center operators have committed to achieving net-zero greenhouse gas (GHG) emissions by improving efficiency and using renewable energy, while also pursuing sustainability goals in areas such as water use.

In 2023, Uptime Intelligence predicted that the industry would face a crisis as these commitments become more expensive to meet, while reporting and regulation requirements tighten further. This prediction still holds, but the issue has intensified due to the increased resources demanded by AI. Today, the sector’s objectives are increasingly viewed in terms of growth versus sustainability, despite efforts to reconcile these two goals.

To accelerate construction projects, data center operators have had to build or co-opt new energy sources. Not all of these can be from renewable sources because reliable (i.e., fossil fuel) power is needed to cover the intermittency of wind and solar energy.

Despite conflicts with national net-zero plans, governments have been persuaded of the economic benefits promised by AI — some have classified data centers as critical national infrastructure and pledged support (see Critical national infrastructure status: what does it mean?). Fears of being outpaced in technology have also raised some national security concerns — governments worry about becoming overdependent on other countries.

Meanwhile, new data center projects face growing opposition from residents, environmental groups and other pressure organizations, as well as local government officials. Complaints include broken environmental commitments, high demands for power, land and water, reliance on tax breaks and limited support for local jobs.

While data center projects continue to attract investment and resources, hostility from citizens, pressure groups and local government is expected to increase. Without greater efforts from developers and operators to explain their value and minimize environmental impacts, concerns will spread across the media and among policymakers.

Context: growth trumps sustainability

Data center operators, along with other sectors and led by government, have committed to reducing energy use and transitioning to low-carbon energy sources. Many governments are aiming for net-zero emissions by 2030, aware that GHG emissions could raise global temperatures by 2.7°C (4.86°F) above pre-industrial levels before the end of 2100, according to a 2023 report by the Intergovernmental Panel on Climate Change.

Reducing emissions is a challenge, particularly for the rapidly growing data center industry. The AI boom is further driving power demands, which will almost certainly put these promises out of reach. This may push operators to use carbon-intensive on-site power, further drawing data centers into political controversy.

Additional AI power demand

Data center power demand is around 1% to 1.3% of total global electricity capacity, equivalent to between 200 TWh and 450 TWh, according to the International Energy Agency. This is predicted to grow to between 700 TWh and 1,100 TWh by 2030, or 1.5% of the world’s total capacity.

AI data centers are not included in this figure. They use GPUs, with a much higher power demand than CPUs. Virtually unknown in 2022, they are expanding rapidly: Uptime Intelligence estimates their annualized power demand at 5.8 TWh in early 2024, potentially rising to 21.9 TWh by early 2025 (see Generative AI and global power consumption: high, but not that high).

Extrapolation from these data points is dangerous, but Goldman Sachs predicts that AI will grow data center power demands by 160% before 2030.

Grid delays drive on-site power

Total electricity demand is growing rapidly, and low-carbon electricity is neither plentiful nor stable enough to meet the combined needs of all users. In prime data center locations, such as Northern Virginia in the US, London in the UK, and Ireland, grid operators are unable to provide timely grid connections. Current waiting times can be around three to six years. As a result, new data centers face long delays and opposition from other organizations that are also competing for power.

To advance projects, data center operators are moving to on-site power. While this sidesteps the connection problem, the only widely practical option is natural gas, often provided through mobile gas generators on trailers (see On-site natural gas: why some sites need it). These are intended to be a temporary measure, but often their use stretches into years and can effectively become permanent. This increases emissions compared with grid electricity in almost every location.

Faced with these conflicts, data center operators are canceling or sidelining emissions targets. In July 2024, Google withdrew a carbon-neutral claim based on cheap carbon offsets. Microsoft and Amazon have withdrawn or been ejected from the Science Based Targets initiative (SBTi), which validates corporate carbon-reduction targets, amid a complex debate about carbon credits and Scope 3 emissions.

Policy conflicts

Governments are allowing — and supporting — new data center developments, creating a conflict between national growth plans and regulations introduced under decarbonization initiatives. For example, the EU supports data center growth but requires emissions reporting and reductions in the sector, as well as 100% renewable energy consumption by 2030 under its Energy Efficiency Directive (EED).

AI is regarded as nationally strategic, with regulations such as the EU’s Artificial Intelligence Act and China’s Cybersecurity Law aiming to establish “sovereign AI”. Meanwhile, laws such as Germany’s Energy Efficiency Act (an EED implementation) make demands for heat reuse and PUE. While these pose a challenge for AI, they are likely achievable using new facilities and liquid cooling technologies.

Local objections: tax, jobs and water

Opposition to data center developments is often dismissed as NIMBYism (not in my back yard), however, the list of grievances continues to grow.

Data center operators and hyperscalers often receive tax breaks in exchange for commitments to create jobs, improve the environment and boost the local economy, but these are frequently not met. For instance, in 2015, the US state of Michigan granted tax breaks to data center developer Switch for the construction of a facility. While Switch still saves $1 million per year in taxes, by 2022, it had created only 26 of the 1,000 promised new jobs.

Although data center water use is small compared with other industries, it has become contentious in some locations. For instance, Microsoft’s AI facility in Des Moines, Iowa, uses up to 11 million gallons per month, while its Amsterdam facilities in the Netherlands have been accused of diverting water away from local farmers. In 2023, Microsoft’s data center water use rose by 35% — partially due to hotter weather — despite a promise to be water-positive by 2030.

Government support, public opposition

Governments will most likely support a faster growth in data center development, as operators claim that an AI-led boom will benefit public finances. In some countries and states, this could give data center builders privileged access to power and land, along with a relaxation of planning and environmental rules.

Meanwhile, opposition to new data center projects is likely to intensify due to the following factors:

• They are growing in size.
• They increasingly depend on locally polluting power sources.
• They conflict with climate goals.
• They are perceived as large water consumers.
• They receive tax breaks but often fail to deliver promised jobs.

Implications and further questions

Supported by government and opposed by public opinion, data center operators will likely reiterate their facilities’ social and environmental benefits, as well as their role in promoting low-carbon technologies such as wind, solar and new nuclear power.

However, governments may increasingly question the level of support they give to data center operators. In some cases, such as in Ireland, opposition parties are calling for restrictions on the construction of new data centers.

Government support may weaken if the economic benefits from AI do not show signs of living up to expectations.