California pioneers artificial intelligence technologies, but can it power the many data centers AI requires? If not, the state could lose its dominant position in developing AI, according to a new Stanford University report.
California can maintain its AI lead by focusing on five key strategies, according to the readout of a recent meeting in Sacramento of experts from California utilities, technology companies, data center developers, investment firms, academia, and government. However, this will require great determination and quick actions throughout the state, said the experts.
U.S. electricity demand is rising on a scale last seen in the post-World War II industrial boom, primarily due to AI computing, electrification of transportation and buildings, and reshoring of manufacturing. By 2040, California peak power demand is forecast to increase by the amount of electricity needed to run 20 million homes, which is more than the number of homes in the state today. California’s task is to supply the new energy while continuing to decarbonize its electricity supply and making power in the state more affordable and reliable.
“The challenge is acute and urgent in the San Francisco Bay Area,” said Liang Min, managing director of Bits & Watts, the smart grid initiative of the Stanford Doerr School of Sustainability’s Precourt Institute for Energy. “PG&E’s data center pipeline has doubled just since February.”
“On the bright side, electricity growth means economic growth, so it can be a good problem to have – if we solve it,” added Min, co-organizer of the meeting and lead author of the readout report. The Precourt Institute hosted the meeting in June, which follows one the institute held in February on expanding the grid for U.S. electricity growth broadly.
Attendees pointed out that very lengthy permitting processes for new power plants and transmission lines, wildfire outages, and sluggish grid connections for data centers and energy storage facilities threaten the reliability of California’s grid – especially with load growth. Executives in the meeting, which was held under assurance that the 50 participants would not be quoted, agreed that the state could lose out on many good jobs building and operating the needed infrastructure, as well as significant tax income.
But that’s not all.
“Technology companies and AI talent will necessarily migrate to the states that effectively support the growth of data centers,” Ira Ehrenpreis, founder and managing partner of venture capital firm DBL Partners, said after the meeting.
“The states that align infrastructure, electricity, and policy with AI’s explosive growth will win,” said Ehrenpreis, who is co-chair of the Precourt Institute’s advisory council, and co-organizer of the Sacramento meeting with Min. “California should create the optimal regulatory and legislative environment to properly compete.”
Five key levers
Ten meeting participants summarized the meeting’s discussion into five recommendations based on extensive notes.
First, the state should create a one-stop permitting venue integrating local, state, and federal reviews. Most big tech companies like California’s relatively low-carbon power supply, but they expect to build and get data centers online in two years. Delays in permits and grid connections in the state make this nearly impossible, though the California Public Utilities Commission and Gov. Gavin Newsom have moved this year to streamline the processes.
California could also buy and warehouse strategic grid equipment, like power transformers, to reduce construction delays. In addition, the state’s utilities should analyze the grid needs of new data centers in clusters – given their burgeoning pipelines – rather than individually.
Second, the state needs to invest in its electricity infrastructure while maximizing its use. On average, California’s transmission system uses less than 40% of its capacity, but data center developers tend to cluster their facilities in certain areas. The state’s grid operator can guide data center developers to underutilized areas and generally improve coordination with the developers for times when power supplies get disrupted.
Third, if data centers can harness their flexibility in power use, that would avoid unnecessary utility investments to meet avoidable peak demand. Most of the facilities have backup generators and energy storage, and they can delay some AI processing. This requires more coordination with California’s grid operator and electric utilities.
Fourth, the state government should do what it can to accelerate deployment of clean, 24/7 generation, such as geothermal energy, low-carbon fuel cells, and natural gas with carbon capture and sequestration. Though such technologies are still in early deployment and relatively expensive, bringing them to scale in the mid- to long-term could substantially lower overall system costs, land requirements, and transmission infrastructure needs compared to meeting new power demand with solar panels and wind turbines alone. Building new power generation near new data centers could also greatly reduce transmission infrastructure needs.
Fifth, state regulators and utilities could advance innovation in business models and financing mechanisms, which could accelerate both load growth and climate progress. Such innovations include enhancing flexibility in electricity use throughout the power system, long-term contracts to help fund grid expansions, and low-interest loans through public-private partnerships.
Affordability imperative
California electricity is among the most expensive supplies in the United States, due to infrastructure hardening for wildfire, years of flat demand, and other factors. New AI demand needs to drive bills down, not up, to maintain political viability, meeting participants agreed.
“Greater demand is crucial to affordably modernizing California’s energy infrastructure,” said William Chueh, director of the Precourt Institute and a professor of materials science in Stanford’s School of Engineering. “Additional demand can increase transmission utilization and spread the costs of infrastructure investments we must make anyway across much more consumption, if we get this right.”
Immediate action
California has about 24 months to correct its course or risk losing its AI and data center competitive advantage to more agile states, meeting participants said. In that time, state agencies need to improve permitting, interconnections, forecasting, operations, and investments. Commercial customers, meanwhile, should commit to capacity agreements, site projects strategically, and enhance grid reliability through existing backup systems and electricity use flexibility. Likewise, California’s universities and Silicon Valley’s entrepreneurial and venture-capital ecosystem are well positioned to move promising ideas and technologies from lab to market quickly.
“Every month saved in siting, interconnection, and construction puts capital to work sooner, creates jobs, raises tax revenues, spreads fixed costs among more customers, and brings clean capacity online when it’s needed most,” said Min. “We must adapt quickly to harness AI’s once-in-a-lifetime economic and technological opportunity.”
The Precourt Institute met privately in Washington, D.C., on Sept. 9 with industry executives and executive branch employees on managing AI-fueled electricity growth nationally. On Nov. 5, it will host Powering AI Innovation Forum, which will take on a global perspective and is open to the public.