Who will win: the inhabitants who demand light, or Big Tech, which builds the AI empire?
Data centers in the United States already consume about 6% of the country’s electricity, and new sites are growing so fast that energy companies, local authorities and residents are increasingly asking whether the network will withstand another wave of construction.
Data centers always required a lot of electricity, but generative AI has dramatically increased the load. Large computing platforms for training and modeling can already consume as much energy as a small city. At the same time, companies continue to build new complexes, because the demand for computing is growing together with chatbots, image generators, corporate AI services and cloud platforms.
The new report of the International Data Center Authority shows the scale of the jump. The total capacity of all data centers in the world reached 67.7 GW, which is 36% more than 2 years ago. The U.S. accounts for 29.2 GW, approximately 43% of global consumption of data centers. The head of IDCA, Mehdi Parjavi, said that many very large AI factories are being launched in the country, which is why the total electricity consumption is growing sharply. According to him, data centers are already taking 6% of all American electricity.
This number is important not only in itself. The IDCA report says that serious discontent of residents and politicians usually begins after data centers reach about 5% of national electricity consumption. The U.S. has already passed this threshold. The UK is almost next to it: data centers there consume 5.8% of electricity. In Germany, the figure is even higher, 9.5%.
Resistance is growing at the state and municipal levels. According to the report, hundreds of bills have already been submitted to the United States that should regulate the construction and operation of data centers. In the state of Maine, lawmakers adopted a document that would prohibit until 2027 to build data centers with a capacity of more than 20 MW. Governor Janet Mills vetoed, and the parliament could not overcome it. Later, Mills signed a decree on the establishment of a council that will study the influence of such facilities on the staff and prepare recommendations for the beginning of 2027.
In some regions, the problem is no longer in politics, but in free power. Local planners refuse to issue new permits due to lack of electricity. In North Virginia, where the Data Center Alley area is densely built up by data centers, the developers of new projects will have to wait until 2032.
Electricity is not the only resource around which conflict is growing. In many areas, data centers also require a lot of water. Most of the facilities use water-cooled chillers or evaporative cooling towers that can consume millions of gallons of water per day. One major data center is potentially capable of consuming as much water as 6500 households. New AI sites are increasingly switching to closed liquid cooling systems, which need much less water with constant operation, but such solutions still occupy a small proportion of the total park.
The authors of the report believe that part of the discontent industry creates itself. Developers often register local companies with neutral names, by which it is difficult to understand who is really behind the project. For residents, this looks like an attempt to hide the owner and the scale of the future object, so trust falls even before the start of public discussions.
Environmentalists require more transparency. Chief researcher at Greenpeace UK Doug Parr, after the publication of the data told The Guardian that before the construction breakthrough it is worth assessing the price of the AI infrastructure for the society. According to his position, data centers should disclose the consumption of water and energy, undergo a full-fledged environmental assessment, and new polluting power plants should not be built only for the sake of AI.
At the same time, the load is created not only by new giant campuses. IDCA estimates that about 13% of cloud consumption in the US, more than 3 GW, falls on zombie loads. This is the name of the forgotten test environments, unused applications and other systems that continue to work and consume electricity, although they no longer benefit.
There is also a less noticeable layer of infrastructure: thousands of small data centers within corporate buildings and regional offices. Energy estimates often focus on large hyperscale campuses, so such sites fall out of the overall picture. According to IDCA, they provide at least 15% of total consumption of data centers, including due to lower efficiency compared to large modern facilities.
There are almost no signs of slowing down. Global spending on data centers is approaching $1 trillion a year, and the U.S. could reach $700 billion in 2026. For AI companies, this is the infrastructure of future growth. For power grids and cities, these are new lines, substations, water intake, emissions and disputes with the population.
Further, the fate of the AI boom will depend not only on models and processors. If networks fail to accept new capacities, and local communities will continue to block construction sites, data centers will become one of the main limiters of the industry. AI can require more and more calculations, but each new cluster must take electricity, water and building permits somewhere.
Data centers in the United States already consume about 6% of the country’s electricity, and new sites are growing so fast that energy companies, local authorities and residents are increasingly asking whether the network will withstand another wave of construction.
Data centers always required a lot of electricity, but generative AI has dramatically increased the load. Large computing platforms for training and modeling can already consume as much energy as a small city. At the same time, companies continue to build new complexes, because the demand for computing is growing together with chatbots, image generators, corporate AI services and cloud platforms.
The new report of the International Data Center Authority shows the scale of the jump. The total capacity of all data centers in the world reached 67.7 GW, which is 36% more than 2 years ago. The U.S. accounts for 29.2 GW, approximately 43% of global consumption of data centers. The head of IDCA, Mehdi Parjavi, said that many very large AI factories are being launched in the country, which is why the total electricity consumption is growing sharply. According to him, data centers are already taking 6% of all American electricity.
This number is important not only in itself. The IDCA report says that serious discontent of residents and politicians usually begins after data centers reach about 5% of national electricity consumption. The U.S. has already passed this threshold. The UK is almost next to it: data centers there consume 5.8% of electricity. In Germany, the figure is even higher, 9.5%.
Resistance is growing at the state and municipal levels. According to the report, hundreds of bills have already been submitted to the United States that should regulate the construction and operation of data centers. In the state of Maine, lawmakers adopted a document that would prohibit until 2027 to build data centers with a capacity of more than 20 MW. Governor Janet Mills vetoed, and the parliament could not overcome it. Later, Mills signed a decree on the establishment of a council that will study the influence of such facilities on the staff and prepare recommendations for the beginning of 2027.
In some regions, the problem is no longer in politics, but in free power. Local planners refuse to issue new permits due to lack of electricity. In North Virginia, where the Data Center Alley area is densely built up by data centers, the developers of new projects will have to wait until 2032.
Electricity is not the only resource around which conflict is growing. In many areas, data centers also require a lot of water. Most of the facilities use water-cooled chillers or evaporative cooling towers that can consume millions of gallons of water per day. One major data center is potentially capable of consuming as much water as 6500 households. New AI sites are increasingly switching to closed liquid cooling systems, which need much less water with constant operation, but such solutions still occupy a small proportion of the total park.
The authors of the report believe that part of the discontent industry creates itself. Developers often register local companies with neutral names, by which it is difficult to understand who is really behind the project. For residents, this looks like an attempt to hide the owner and the scale of the future object, so trust falls even before the start of public discussions.
Environmentalists require more transparency. Chief researcher at Greenpeace UK Doug Parr, after the publication of the data told The Guardian that before the construction breakthrough it is worth assessing the price of the AI infrastructure for the society. According to his position, data centers should disclose the consumption of water and energy, undergo a full-fledged environmental assessment, and new polluting power plants should not be built only for the sake of AI.
At the same time, the load is created not only by new giant campuses. IDCA estimates that about 13% of cloud consumption in the US, more than 3 GW, falls on zombie loads. This is the name of the forgotten test environments, unused applications and other systems that continue to work and consume electricity, although they no longer benefit.
There is also a less noticeable layer of infrastructure: thousands of small data centers within corporate buildings and regional offices. Energy estimates often focus on large hyperscale campuses, so such sites fall out of the overall picture. According to IDCA, they provide at least 15% of total consumption of data centers, including due to lower efficiency compared to large modern facilities.
There are almost no signs of slowing down. Global spending on data centers is approaching $1 trillion a year, and the U.S. could reach $700 billion in 2026. For AI companies, this is the infrastructure of future growth. For power grids and cities, these are new lines, substations, water intake, emissions and disputes with the population.
Further, the fate of the AI boom will depend not only on models and processors. If networks fail to accept new capacities, and local communities will continue to block construction sites, data centers will become one of the main limiters of the industry. AI can require more and more calculations, but each new cluster must take electricity, water and building permits somewhere.
