California data center moratorium clears key Senate panel, reviving fight tech sector thought it had won
4 min read, word count: 925SACRAMENTO, Calif. — A California state Senate committee on Friday advanced legislation that would impose an 18-month moratorium on new hyperscale artificial intelligence data centers, reviving a regulatory fight the technology industry believed it had largely settled when a similar federal bill died in the U.S. House last month.
The Senate Energy, Utilities and Communications Committee voted 9-4 to send SB 1142, the California AI Infrastructure Pause Act, to the full Senate. Authored by state Sen. Marisol Quintanar, D-Oakland, with co-sponsorship from 14 colleagues, the measure would block new permits for data center campuses drawing more than 50 megawatts of grid power until July 2027, pending a state-level review of grid reliability, water use and emissions impacts.
The vote came nine days after the federal AI moratorium pushed by Sens. Bernie Sanders and Rep. Alexandria Ocasio-Cortez collapsed in the House Ways and Means Committee on a 24-21 vote, an outcome that hyperscaler executives celebrated as the end of a months-long lobbying battle. Industry groups had spent the days after the federal vote arguing that energy-related concerns would be addressed by negotiated standards rather than construction pauses.
That argument did not carry the day in Sacramento.
“California ratepayers are not a sacrifice zone for a national compute build-out,” Quintanar said in a brief statement after the vote. “If the people who run these facilities want to argue the grid can handle them, they can make that argument with data, and they can make it in 18 months. Until then, we hit pause.”
The bill is narrower than the failed federal version, which had targeted training runs above a fixed compute threshold. The California measure focuses on physical siting — new builds, expansions and so-called “behind-the-meter” generation tied to data centers — and exempts facilities already under construction or with finalized interconnection agreements as of March 1. Aides to Quintanar said roughly 40 projects across the state would fall within the scope of the pause.
California Independent System Operator data released last week showed planned data center load in the state rising to an estimated 17.4 gigawatts by 2030, up from 6.1 gigawatts at the end of 2025, with the bulk of new requests concentrated in Santa Clara, San Bernardino and the Sacramento exurbs. CAISO’s spring reliability assessment, also published last week, flagged 2027 and 2028 as the years of greatest concern for summer peak adequacy.
Industry response was sharp. The California Chamber of Commerce, the Computer and Communications Industry Association, and a coalition that includes most major cloud providers issued a joint letter Friday afternoon calling the bill “an economically self-defeating overreach” and warning that capital projects could be redirected to Arizona, Nevada and Texas.
“You cannot lead the world in AI and refuse to build the infrastructure that runs it,” said Brendan Whitfield, vice president for state policy at the trade group TechCoalition West. “Eighteen months in this industry is a generation. This bill, if it becomes law, hands the decade to other states and, frankly, to other countries.”
Whitfield said member companies had already committed more than $34 billion in California capital expenditure through 2028 and warned of “phased reconsideration” of those plans if SB 1142 reaches Gov. Gavin Newsom’s desk in its current form.
Newsom has not taken a public position on the bill. A spokeswoman, Carla Reyes, said the governor “shares the Legislature’s concern about grid stress and water” but “also recognizes the role California plays in the national technology economy” and would evaluate any final bill on its merits.
Supporters include an unusual coalition. Environmental justice groups led by the Leadership Counsel for Justice and Accountability have lobbied hard for the pause, citing diesel backup generation in the Inland Empire. They were joined by the State Building and Construction Trades Council of California, whose leadership has criticized non-union contracting at several recent hyperscale projects, and by ratepayer advocates worried about cost-shifting from industrial customers to households.
“What changed after the federal bill died is that the energy argument got stronger, not weaker,” said Priya Ramaswamy, a senior policy analyst at the nonpartisan Energy Innovation Policy and Technology group. “The hyperscalers won in Washington partly by promising voluntary standards. Sacramento is calling that bluff.”
A similar effort is underway in New York, where Assembly Member Daniel Kweller, D-Brooklyn, introduced a companion measure last week that would empower the Public Service Commission to deny interconnection to data centers above 100 megawatts pending a study. That bill has not yet had a committee hearing.
Friday’s vote in California broke largely along party lines, with all four Republicans opposed, joined by no Democrats. Two Democrats on the committee, both from districts that host or have proposed data center projects, did not attend; aides described their absences as “scheduling.”
A Senate floor vote could come as soon as the week of May 18, according to Senate President pro Tempore Diego Calderón’s office. If the bill passes the Senate, it would move to the Assembly, where the politics are widely viewed as tougher; Speaker Adriana Loomis has not endorsed the measure.
Industry lobbyists said they would now focus on the Assembly and on a possible carve-out for projects that pair construction with new firm clean generation. Quintanar’s office said the senator was open to “narrow, measurable” amendments but not to weakening the core 18-month pause.
The state Air Resources Board is expected to release updated data center emissions projections later this month, a report that aides on both sides said could shape floor debate. Additional hearings on grid impacts were scheduled for May 12 and May 19.
Note: This article was partially constructed using data from LLM.