Who pays for AI electricity? Data centers spark debate over rising energy costs

Companies rushing to build the massive infrastructure needed for the rise of artificial intelligence face a growing backlash over electricity costs, as households and policymakers question whether data centers are driving up energy bills.

However, a recent report from SemiAnalysis, a semiconductor research company, argued that data center expansion is only part of the story and stated that market design and policy decisions play a larger role in these energy price increases than the growth of AI infrastructure alone.

From rural Virginia to the Arizona desert, communities that once welcomed investment in technology are now rejecting data centers amid growing concerns that these facilities – built by so-called AI hyperscalers – are straining local power grids, driving up costs for everyone else.

Since 2020, US residential electricity prices have increased more than 36%, from 12.76 cents per kilowatt-hour to 17.44 cents per kilowatt-hour in February 2026, and are expected to reach 19.01 cents per kilowatt-hour in September 2027, according to the latest forecast from the US Energy Information Administration.

“Retail electricity prices have risen faster than the rate of inflation since 2022, and we expect them to continue rising through 2026,” the EIA said in a March 2025 report before the Iran war.

US President Donald Trump also recently acknowledged the problem for the industry, saying data centers “need public relations help.”

Retail electricity prices in the United States reflect the costs of generating, transmitting and delivering energy, along with other factors such as taxes and investments in utilities to improve aging infrastructure.

SemiAnalysis claimed that an obscure market pricing mechanism known as Base Residual Auction accounted for most of the “runaway” power prices in the interconnection area of ​​PJM, a regional grid operator that serves 13 eastern states and hosts data centers for hyperscalers such as Google, Anthropic and Amazon.

Under this mechanism, consumers make payments for expected electricity costs two years in advance, ensuring sufficient availability of energy during periods of peak demand, such as heat waves or winter storms.

Future energy prices under the mechanism are forecasted prices based on anticipated future demand, calculated through simulations performed with proprietary models and data. But in all prediction models, the parameters may not always reflect real-world circumstances.

SemiAnalysis argued that PJM’s forecasts often overestimated future demand, particularly as many planned data centers in the area faced construction or assembly delays due to chronic memory shortages.

The report compared PJM to another energy grid overseen by the Electric Reliability Council of Texas, where it said prices have remained relatively stable since 2022, despite the development of data center complexes by hyperscalers such as OpenAI, Anthropic and Google.

In the United States, where regulations governing power grids are decentralized between states and utility providers, market design often determines how additional costs are passed on to households.

The EIA also noted regional price disparities in a March 2025 report, saying regions with high residential electricity prices could see increases above the national average.

“In a capacity-constrained market like PJM, prices have increased dramatically as demand for data centers has increased. However, other markets allow for more complete direct cost allocation,” Maeghan Rouch, partner at Bain & Company, told CNBC.

Exactly what drives increases in consumer energy prices may not always be clear, as unrelated investments in local networks, such as strengthening and modernizing networks, or general inflation, can also weigh on households, Rouch added.

“Even in the absence of data center investment, we would still expect some degree of upward pressure on price growth,” Rouch said.

Big tech companies have also worked to calm concerns about their energy use, with promises to cover electricity costs for their projects or develop alternative energy sources.

In January, Microsoft outlined a five-point plan, including a commitment to cover any additional electricity costs resulting from its data centers, among other community investments. This was followed by a similar commitment from Anthropic in February.

More recently, President Trump summoned executives from major AI corporations to the White House to affirm the Taxpayer Protection Pledge, ensuring that expenses incurred by new AI data centers are not passed on to American consumers.

Such commitments could prove particularly important to “gain support from communities that might otherwise oppose (data center) projects,” according to Chris Howard, head of data center account management at JLL, especially if data center development were accompanied by alternative investments in local communities, such as jobs or training.

But experts have questioned the legitimacy of such commitments, as hyperscalers have struggled to make profits.

“The problem is that the industry is not making money, which puts even more pressure on them,” said Marc Einstein, research director at Counterpoint Research.

Hyperscalers should also clarify their plans to address rising electricity costs, he added. “If they stay silent about it, that will allow the rumors to get out of control.”

Technology companies have also committed to meeting the needs of data centers through renewable sources.

These alternative energy sources will become increasingly important as concerns about energy availability grow due to increasing demand for data centers around the world, according to JLL’s Howard.

“The average wait time for a network connection in major data center markets is already four to six years, and up to 10 years in cities like Tokyo,” Howard said.

Globally, these energy deficits could “create enormous opportunities for energy producers, particularly when it comes to renewable energy,” he added.

However, skepticism toward renewable energy commitments within the current US administration has raised questions about the extent to which such sustainability promises will advance the country, Howard said.

Still, analysts say it may be in the corporate interest of AI hyperscalers to deliver on these promises.

“It would definitely be better public relations,” Einstein said.

But public backlash could also prompt regulators to impose new rules on hyperscalers, Einstein added, “which is actually not what they want.”