Blog

Enormous political and media attention is being given to the increasing demand for electricity to power AI and data centers with proposals to spend hundreds of billions of dollars. President Trump has said U.S. power output will need to double to meet AI demands. One of his first announcements was support for a new company, Stargate, with plans to invest up to $500 billion in new AI infrastructure (contrary to his claim, the initiative was first announced much earlier in March 2024).

Much less attention is being given to what turns out to be an even larger and more difficult challenge — meeting the power needs of an increasing global need for air conditioning. Attaching the proper importance to these two very different sources of demand will be critical not only to the future reliability of the grid but for meeting climate and other environmental goals.

AI and ChatGPT searches require much greater electricity than Google searches. Data centers are much larger and more energy intensive. Whereas ten years ago, a 30-megawatt (MW) center was considered large, today, a 200-MW facility is considered normal. In many parts of the U.S., powering such facilities is already adversely impacting the quality of power for residents within 20 miles. Analysis by McKinsey consultants concludes demand for AI-ready data center capacity will rise at an average rate of 33 percent a year between 2023 and 2030 in a midrange scenario. Data centers require power supplies with high levels of redundancy and reliability, as well as cooling to keep systems from overheating.

Demand for electricity to power air conditioning is, surprisingly, much larger than for data centers. In the U.S., powering data centers is projected to rise to about 6 percent of demand in 2026. In contrast, according to the Energy Information Administration, air conditioning requires substantially more, about 15 percent across households, commercial buildings, and manufacturing. While powering data centers is projected to grow rapidly in the U.S. in the years to come, so will demand for AC thanks to the increasing frequency of extreme temperatures (more than 80 days above 110F in Phoenix last year including locations like Portland, Oregon, and Seattle, Washington where cooling was rarely if ever needed. The result has already been emergency planning, delayed closure of power plants and commissioning of new ones to avoid rolling blackouts.

The disparity globally is much greater. According to the IEA’s 2024 World Energy Outlook, by 2030, electricity for AC will increase more than three times the extra demand from computer data centers. (Electricity demand for electric vehicles is growing even faster — if not slowed by Trump policies — but provides substantial offsetting benefits through reduced oil consumption and the potential for battery storage.) Many countries most vulnerable to rising temperatures and extreme heat have very low rates of air conditioning — 5 percent of households in Nigeria, 15 percent in Indonesia, and 20 percent in India. And thanks to climate change, the frequency and intensity of extreme heat events is now certain in almost every part of the globe. The combination of rising incomes and increasing global temperatures will generate additional global demand for cooling by 2035 greater than the entire Middle East’s electricity use today. As Faith Birol, head of the IEA said in a recent interview, “It is a blind spot for many decision makers to see how important air conditioners are as a driver of global electricity consumption.”

For several reasons, meeting the power demand for air conditioning is also much less predictable and manageable than for data centers. Data centers are increasingly being planned with dedicated power plants on-site, the approach Trump proclaimed in Stargate. Another strategy is to locate in areas with both power availability and relatively low energy prices — near Dallas, in ERCOT; Northern Virginia and Pennsylvania, in the PJM region; and near Atlanta. In contrast, the timing and location of heat extremes is both highly dispersed and much less predictable as shown by recent record breaking temperatures in the Pacific Northwest and parts of Canada during late June through mid-July 2021. Air conditioners also tend to be operated during afternoons and early evenings adding to peak demand. (Charging EVs in the evening may also add to peak demand, a potential problem.)

A recent detailed analysis of the fastest and cheapest strategies for powering large AI data centers in the American southwest makes a compelling case for the advantages of off-grid solar microgrids. The time required for site acquisition and permitting is less than for either new grid connections or dedicated gas turbines, the cost is roughly equivalent to natural gas turbines, and (although not of interest to the Trump Administration) substantial greenhouse gas emissions would be avoided. In sum, as the climate activist Bill McKibben notes, “essentially all you have to do is put up a bunch of solar panels and some batteries and run a wire to your data center.”

And the search for better solutions continues. There is considerable potential for technologies already available and under development to power AI applications using much less power. China has already developed AI systems that are cheaper and more efficient without the use of chips made in the U.S. — a development now roiling the stock market. Multiple ways have been proposed to use AI to enhance grid reliability through better analysis of issues relevant for permitting, designing, and operating power plants, storage systems, and transmission lines. A program sponsored by the Electric Power Research Institute, the DCFlex Initiative, is bringing utilities together with data center and technology developers to demonstrate innovative strategies for integrating data centers with the grid. Nuclear power is also receiving renewed attention. Microsoft is seeking approval to reopen the Three Mile Island nuclear power plant, while Amazon and Google are among those considering investment in developing a new generation of small modular reactors. Consequently some very knowledgeable tech leaders like Bill Gates argue AI systems are no reason for anxiety.

Meanwhile much less attention has so far been given to finding ways to meet the growing cooling related demand for power. To make things worse, President Trump wants to lower appliance efficiency standards in order to “safeguard the American people’s freedom” to choose. As the worst heat tends to be in urban areas, a growing list of large cities are implementing programs providing emergency shelters, planting trees, and other measures in response to heat waves — helpful but an inadequate substitute for air conditioning. Many countries have energy efficiency standards for appliances, but these are typically way below the most efficient levels available due to higher first cost and resistance from consumers. The International Finance Corporation, the private sector lending arm of the World Bank, offered grants to early-stage companies with innovative cooling solutions for developing countries and identified dozens that appear promising. However, the resources and technical support necessary for these companies to scale and grow rapidly is not readily available.

Satisfying growing electricity demands while reducing greenhouse gas emissions consistent with avoiding catastrophic climate change is one of the greatest challenges of our time. While solutions for powering data centers appear increasingly feasible, finding equally sustainable ways to meet the energy needs for cooling requires much greater effort and support.

Alan Miller is a former climate change officer in the International Finance Corporation (2003–13) and climate change team leader, Global Environment Facility (1997–2003). Besides other engagements, Alan is an active editor for Climate Conscious submissions on Medium.