Silicon Valley’s AI Boom Hits a Wall: Data Centers Are Built but Can’t Turn On

(TNS) — A few blocks away from San Jose's airport, an empty concrete and glass structure is waiting to be turned on.

Work began six years ago on the four-story data center, called SJC37 and owned by Digital Realty, one of the world's biggest data center landlords. The tan exterior is finished, but there's no equipment installed — and it could be years before the project gets the power it needs to start operations.

Silicon Valley Power, a public utility owned by the city of Santa Clara, is working on a $450 million system upgrade to meet demand from data centers and other users, but isn't scheduled to finish until 2028. Another nearby project owned by data center company Stack Infrastructure is also sitting empty and isn't expecting power until 2027.

The delays, first reported by Bloomberg, are emblematic of the Bay Area's disparities in the AI industry.

The region is the runaway leader in artificial intelligence, home to both the world's deepest engineering talent pool and status as headquarters of industry giants like Nvidia, Google and OpenAI.

But data centers, the physical architecture needed for AI computing, are a different story — and energy shortfalls are a key reason why the Bay Area is losing ground to other U.S. regions when it comes to data center construction.

As of June, Silicon Valley's data center power capacity grew by only 25.2 megawatts for a total of 484.4 megawatts compared to a year earlier, making it the sixth-largest market in the country, according to real estate brokerage CBRE. Larger markets like Atlanta added a whopping 969 megawatts over the year.

"There's lots of reasons not to put data centers in California — from a cost perspective and an energy supply perspective," said Bill Dougherty, CBRE executive vice president of data center solutions, who is based in San Francisco. "From the supply side, there's lots of physical and political constraints."

Though some data centers need to be physically close to the AI companies using them, many aren't constrained by location and can seek the cheapest power, Dougherty said. "The market is naturally constrained by the amount of power we can get," he said.

The biggest data center cluster in the world is in Northern Virginia, home to the first non-governmental Internet Exchange Point, a vital connection point for various network service providers, built in 1992. Historically low power costs and a state sales tax exemption on equipment purchases for major data centers cemented its dominance as "one of the best places in the world" for the industry, said Dougherty. The region has seven times as much power capacity as Silicon Valley, and nearly three times as much as Atlanta, the next largest region, according to CBRE.

In contrast, California's energy costs are some of the highest in the country, as well as its taxes — and no tax breaks are offered for data centers. Land near urban areas, particularly in Silicon Valley, is pricey and constrained.

Some economic experts are fine with the Bay Area missing out on the data center boom.

Ted Egan, San Francisco's chief economist, is skeptical that data centers bring much economic benefit beyond their construction activity and some tax revenue.

"They need some electricity, but they don't need a lot of people," he previously said. There's far more economic upside to having highly paid AI engineers working in offices, he said — a key reason why Mission Bay is seeing more growth in economic activity compared to the rest of San Francisco.

Egan has also said that if AI turns out to be another tech bubble that ends up bursting, data center projects around the country could be cancelled first, rather than offices shuttering. That could mean the Bay Area is less vulnerable to an AI downturn than rural data center boomtowns.

There's also significant environmental concerns around power and water usage by data centers, as well as disputes over whether more data centers will raise or lower power bills. "Everyone in data centers has to be concerned about the environmental impacts," Dougherty said.

A spokesperson for Digital Realty, the owner of the powerless Santa Clara data center at 641 Walsh Ave., said in a statement that the Bay Area is a "critical piece" of the company's global strategy, and it serves over 315 companies in the region.

"We remain optimistic about bringing capacity to market as soon as utility infrastructure allows," the spokesperson said, adding that "sustainability is part of every decision" and the company is focused on supporting "grid improvements to support the growth of AI without wasting resources."

Unlike other regions, the Bay Area has largely avoided big electricity price increases attributed to data centers, while the data center-rich Mid-Atlantic region has seen soaring bills attributed to the industry's boom. But studies are split on data centers' impact on energy bills, which have risen in much of the country in the past year.

In a study published earlier this month, Lawrence Berkeley National Laboratory researchers found that data centers have likely reduced power bills by spreading out utility upgrade costs among more customers. But the study said it was difficult to forecast whether the benefits would continue as energy demand from data centers continues to grow.

Meanwhile, a June Carnegie Mellon and North Carolina State University study warned that more data centers, along with growth in cryptocurrency, could cause U.S. electricity prices to spike 8% by 2030.

This week, three Democratic senators sought information from tech giants Google, Microsoft, Amazon and Meta, as well as Digital Realty and other data center owners, on data centers' role in increasing energy costs.

At the same time, an Amazon-commissioned study by Energy and Environment Economics found that the company's data centers across four states including California were a net positive for customers, with a 100 megawatt data center generating $3.4 million more for utility companies beyond the costs to power it.

That figure is projected to reach $6.1 million by 2030, if utilities continue to adjust their rates based on market demand.

"I think it showed that we are continuing to grow the business smartly without negatively impacting residential ratepayers," Brandon Oyer, head of energy and water for the Americas at Amazon Web Services, said in an interview. "There's no evidence that California residents are paying to power our Amazon data centers ... data centers can actually help spread out infrastructure costs over a broader base."

PG&E has also embraced data centers as a way to help pay for its costly infrastructure and wildfire prevention upgrades.

The utility said in May that for every 1 gigawatt of new data center energy demand, customers could save between 1-2% on monthly bills. Demand for PG&E electricity has swelled to 10 gigawatts for new projects planned over the next decade as of July, enough to power 7.5 million homes.

In San Jose, PG&E and developer Westbank are seeking to build a net-zero project with 4,000 apartments.

Their neighbors? Three data centers.

© 2025 the San Francisco Chronicle. Visit www.sfchronicle.com. Distributed by Tribune Content Agency, LLC.