Data centers for AI could nearly triple San Jose’s energy use. Who foots the bill?

Those plans are forcing major grid upgrades, PG&E and city officials say, while raising questions about who pays for them and whether the state can keep the power clean.

Panelists at a CalMatters event in downtown San Jose clashed over key issues. They included a local official working with PG&E on the city’s data-center buildout, a tech advocate urging California to seize the economic moment, a Stanford energy expert pressing for a more modernized grid and a utility watchdog skeptical of AI’s promised benefits.

Their discussion centered on how quickly California should move to accommodate new demand, what information the public should be entitled to and how to keep customers from shouldering the cost of infrastructure that may never be fully used.

Proposals to more strictly regulate data center development died in the Legislature this year. Going forward, several state agencies and commissions are expected to take up further discussions, including the California Energy Commission, the Little Hoover Commission and the California Public Utilities Commission.

How much energy will California’s new data centers actually need?

The surge in AI is complicating efforts by regulators and utilities to forecast how quickly data centers will grow and how much power they’ll need. Companies can propose large facilities without committing to build them, the computing demands behind AI are changing quickly and cooling needs vary across the state. These factors make long-term energy needs hard to pin down.

According to the state’s electricity-demand forecast, utilities report that data centers, in planning documents, have requested 18.7 gigawatts of service capacity. That’s enough to power roughly 18 million homes, compared with California’s estimated 14 to 15 million. Regulators don’t expect all of those projects to be built, and assume the ones that do will come online gradually and operate at less than their requested capacity, producing a forecast of between 4 and 6 gigawatts by 2040.

Liang Min, who directs Stanford’s Bits & Watts Initiative, and a speaker on CalMatters’ panel, said that forecasting is particularly tough because companies are rolling out new AI apps — or “application layers,” as he put it — at breakneck speed. They include products like ChatGPT that use large language models. No one knows which apps will take off, and those uncertain bets are driving huge demands on the power grid.

“Right now we’re really struggling,” Min said. “The risk is extremely high in the application layers.”

The Public Advocates Office, an independent consumer watchdog within the California Public Utilities Commission, recently warned that rapid data-center growth could leave Californians paying for billions of dollars in grid upgrades if projects never materialize or use far less power than promised.

“Ratepayers could end up paying for costly infrastructure upgrades that may not be needed for many years – or at all,” the office said in its commentary.

Min said forecasting data-center load is a national challenge, but California will need better tools to keep rates in check, meet its clean-energy targets and stay competitive with states racing to attract data centers and high-paying tech jobs.

Local officials have also begun to grapple with the uncertainty. In San Jose, city energy officials say they are reluctant to procure additional power until they know which projects will actually be built. “We do not want to buy more power than we need,” said panelist Lori Mitchell, director of San Jose Clean Energy, the city’s publicly-owned electricity provider. “That’s job No. 1.”

What are the environmental concerns around the data center boom?

California’s data-center boom is bringing a wave of environmental concerns that state officials are only beginning to understand. Those concerns center on water use, the carbon emissions tied to rising energy demand and the air pollution from diesel backup generators.

Air quality is a particular concern. While back-up generators run only intermittently, their presence is concentrated in a handful of regions. In Santa Clara County, where many facilities sit close together in dense industrial areas, the local impacts could be greater simply because so much equipment is packed into a small space.

Yet the state still has limited visibility into what data centers are doing. Attempts to require more transparency stalled this year amid tech industry opposition. The only measure that became law gives regulators the authority to determine whether data centers are driving up costs — but stops short of requiring environmental reporting.

Ahmad Thomas, CEO of the Silicon Valley Leadership Group, and another panelist, said his group opposed the electricity disclosure and water reporting measures because they would make California less competitive.

“It’s very hard to see a world where California is at the top of the AI pile if we do not have an approach to data centers that is — at minimum — mildly competitive with other states,” he added.

Consumer advocates say the lack of information leaves communities unprotected. “We certainly think there needs to be more transparency — that’s a good thing,” said panelist Mark Toney, the executive director of the The Utility Reform Network, a ratepayer advocacy group..

Will data centers slow down California’s switch to clean energy?

The rapid growth of data centers could slow California’s clean-energy transition if it keeps the state tied to natural gas. And some of the carbon-free alternative energy sources that could meet their power needs are deeply controversial among environmentalists.

The state has pledged to reach 100% carbon-free electricity by 2045, yet it still depends heavily on natural-gas plants during hot summer days. A recent report by the environmental think tank Next 10 and UC Riverside estimated that data-center carbon emissions nearly doubled from 2019 to 2023 — largely from gas-fired generation — underscoring how even a relatively clean grid may struggle to absorb AI-driven load without higher emissions.

State leaders are making policy shifts as AI demand grows. California this year approved joining a broader Western power market, a move driven in part by new demands on the grid, including data centers. Critics warn the change could expose the state to dirtier electricity from other states and weaken its control over clean-energy rules.

Min of Stanford argues that California will need to rely on options some environmentalists would rather avoid. That includes holding onto existing resources like the Diablo Canyon nuclear plant. In a recent report, Min argued the state will also need more “clean, firm” power — resources that can operate around the clock — such as geothermal energy or natural-gas plants with carbon capture.

PG&E agrees. Spokesperson Stephanie Magallon told CalMatters in an email that nuclear power, carbon-capture systems and large solar-plus-battery projects are all options under consideration for powering data centers in its region. But environmental justice critics in California have opposed carbon capture technology calling it unproven tech that risks extending fossil-fuel use.

Mitchell said community choice aggregators can manage new data-center load while keeping power clean and affordable. San Jose’s mix is already 60% renewable, and she said the biggest opportunity is flexibility — getting data centers to shift use off the hottest afternoons so the city can avoid buying additional power.

Will data centers raise your electric bill?

California’s data-center boom is reshaping the fight over electricity bills, exposing a divide over whether these new customers will lower costs — or drive them higher for everyone else.

PG&E argues that adding large users like data centers can lower rates because fixed grid costs would be spread across more customers. It also claims the grid is underutilized on average — operating at about 45% of capacity — although the grid faces real strain during the hottest hours and in parts of the system that routinely run close to their limits. If data centers can be connected in places with available capacity, PG&E argues, they could help spread costs without worsening congestion.

Toney, another panelist, urged the state to slow down, warning that California is planning major infrastructure without knowing which data centers are real or how their costs will land on customer bills.

“I’m worried that we’re engaged in what I call faith-based policymaking,” he said. “The benefits are very speculative, but the costs are very real.”

Some states, said Toney, have begun tightening rules around the growth of data centers. One law in Oregon will require data-center grid costs to remain off household bills. A Minnesota law will give very large data centers their own billing category so regulators can keep their costs separate from other customers’ electric bills.

“This issue of data centers and the connection between affordability and clean energy is of national concern, and California is actually behind on this,” Toney said. “There’s this mythology about California being the leader all the time.”

This story was originally published by CalMatters and distributed through a partnership with The Associated Press.