Inflated Electricity Consumption for AI: A Real Threat or an Exaggeration?

Technology companies are racing to develop artificial intelligence, and their data centers are scrambling for every kilowatt of electricity. But what if all the hype surrounding massive energy consumption is not as frightening as it seems? An article in The Verge points out that forecasts of electricity demand growth due to AI are often inflated. This could push the United States to build unnecessary gas-fired power plants and pipelines, resulting in more pollution and higher costs for residents. Let us take a closer look at what is happening behind the scenes of this energy adventure.

Growing Demand

Before the rise of generative artificial intelligence, electricity consumption in the US had remained stable for many years thanks to energy savings. Now that is changing. Data centers that train and run new AI models require enormous amounts of power. According to S&P Global analyst Dan Thompson, a traditional data center rack consumes 6 to 8 kilowatts—equivalent to the consumption of three American households. But for AI systems with powerful chips, that figure jumps to 80 to 100 kilowatts, equivalent to the energy used by 80 to 100 homes. "That is the amount of energy needed for a small town," Thompson says.

This growth is not just theoretical. Between January 2023 and January 2025, companies and independent developers proposed 70% more gas-fired capacity than before, largely because of data centers. If all these projects went ahead, the number of gas-fired power plants would increase by nearly a third. In the southeastern US, where data centers are concentrated, utilities estimate demand growth of up to four times the level projected by independent analyses from the Institute for Energy Economics and Financial Analysis. According to a December 2024 report by Grid Strategies, utilities nationwide are preparing for 50% more growth than the technology industry expects.

Speculation and Inflated Estimates

The problem is that many of these forecasts are based on speculation. Speculators are flooding the market and building data centers in order to sell them quickly. To avoid long waits for grid connections, they request power before they have secured capital or customers. There is also double counting—developers approach multiple companies at once, inflating the overall estimate. According to a Southern Environmental Law Center report reviewed by London Economics International, there is an "overestimation of future demand."

Companies acknowledge this. Jim Burke, CEO of Vistra Energy in Texas, said during the company's first-quarter earnings call this year: "Proposed grid interconnection projects may be overstated by three to five times compared with what is actually built." Other analyses by BloombergNEF estimate that US data center demand will rise from 35 gigawatts in 2024 to 78 gigawatts by 2035, accounting for 8.6% of total electricity consumption. But skeptics point to constraints: The projections would require the US to consume 90% of global chip production, which is unrealistic because of supply-chain limitations.

This uncertainty is frightening. "The uncertainty is concerning," warns the Compute and Consequence report released this month by As You Sow and the Sierra Club. If the forecasts prove exaggerated, utilities will build unnecessary assets that customers will pay for through higher bills. Conversely, if demand explodes, power outages could occur.

Environmental Impacts

The worst part is that these inflated estimates are driving the construction of fossil-fuel infrastructure. In Louisiana, local utility Entergy proposed three new gas-fired power plants for a massive Meta data center that will consume as much energy as 1.5 million homes and produce 100 million tons of carbon emissions over 15 years. This conflicts with the Biden administration's goal of achieving 100 percent carbon-free electricity by 2035. The new infrastructure is moving the country in the opposite direction, increasing air pollution and slowing the transition to solar and wind energy.

The current Trump administration, supported by contributions from the oil and gas industry, is promoting fossil fuels. This threatens the slow progress in clean energy that the US has achieved through solar and wind farms. Higher electricity bills and more pollution—that is the price Americans could pay for a speculative AI bubble in which investors are pouring money into technologies that may fail.

Kelly Poole, lead author of the report by As You Sow and the Sierra Club, says: "While the AI boom presents exciting opportunities, there are many risks if its energy needs are not addressed deliberately and thoughtfully with long-term impacts in mind."

Solutions Within Reach

Fortunately, there are ways forward. Utilities should require developers to disclose how many other companies they have approached and how far along they are in the planning process. Contracts could require long-term agreements, nonrefundable deposits, and project cancellation fees. Technology companies such as Amazon, Meta, and Google, which have long purchased renewable energy, should invest in solar and wind farms. Long-term contracts for new projects could offset reductions in support for renewable energy.

AI efficiency is improving—new architectures such as "Mixture of Experts" reduce consumption—but scaling up its use may offset those gains. The key is caution: accurate estimates and prioritizing clean energy so that AI does not destroy the planet it is supposed to help. If utilities and technology companies do not get carried away by the hype, they can manage this growth without disaster.