California signs seven laws on AI data centers

Governor Gavin Newsom on Monday, September 21, 2026, signed a package of seven data-center bills, which his office billed as "the most comprehensive data center laws in the nation." The package gives communities more information and control over water, electricity, and land use for data centers (official press release).

Key provisions:

• Electricity (AB 2383, SB 886, SB 1168): Data centers must pay their fair share of grid-upgrade costs and interconnection; the CPUC must establish separate transmission/distribution and generation tariffs for large data-center customers (threshold set by the CPUC, capped at 25 MW); tariffs must include ≥10-year payment mechanisms, upfront collateral or prepayment, early-termination fees, and wildfire-mitigation/societal-cost contributions so costs cannot shift to residential and low-income ratepayers. SB 886 requires incoming data centers to disclose whether they have applied for interconnection in other service territories. SB 1168 requires the CPUC to assess rate structures that ensure data centers pay a reasonable share of transmission/distribution and load-procurement costs and reduce residential rate pressure.
• Water (AB 2469, AB 2619): No city/county may approve a permit for a new or expanded data center unless the applicant supplies a water supply assessment and (from Jan 1, 2028) a water scarcity plan, plus projected water use and workforce disclosures, and assumes the full cost of required water infrastructure. Data-center owners must report expected/actual water use (under penalty of perjury) in business-license applications and renewals, including cooling-system type; data centers are counted in urban water suppliers' demand forecasts.
• Siting / environmental review (SB 887): Data centers are no longer eligible for blanket CEQA categorical exemptions. In exchange, a data center certified by the lead agency as meeting strict conditions (prepaid interconnection, no net increase in fossil-fuel consumption, ≥4-hour zero-carbon storage at 100% of peak, behind-the-meter zero-carbon generation, 100% zero-carbon hourly energy within 5 years of which 75% newly developed, recycled water or waterless cooling, community benefits agreements) can be certified by the Governor as an "environmental leadership development project" for CEQA streamlining. Geothermal power plants get a parallel track.
• Reporting (AB 1577): Data centers ≥10 MW IT capacity must report location/size, PUE, energy consumption, and fuel consumed by onsite generators to the California Energy Commission, with a data-center load assessment in the 2029 Integrated Energy Policy Report.

Context: The signing came two days after Newsom's Executive Order N-9-26 (Sep 18) ordering AI safety recommendations including a possible "kill switch" for frontier models — California's one-two regulatory punch on AI compute and AI safety. It follows a year of escalating local backlash: moratoriums/bans in Monterey Park (first-in-nation permanent voter ban), Coachella, Desert Hot Springs, Palm Springs, Tulare County, Richmond, LA County (unincorporated areas), and more; while Congress considers national data-center power-cost bills and other states pass similar ratepayer-protection laws.

• Reporting regime (changed): Before, California had essentially no public data-center energy/water reporting (developers used NDAs; UC Berkeley researchers found almost no public water data). After: mandatory, penalty-of-perjury water disclosures at license time, annual CEC energy reporting (≥10 MW), and water-supply assessments/scarcity plans before permits.
• Cost allocation (changed): Before, grid-upgrade and interconnection costs could be socialized across ratepayers. After: data centers must prefund/collateralize ≥10-year generation commitments, pay for transmission upgrades they trigger, and pay wildfire/societal-cost shares; local government may not approve projects without water-infrastructure cost responsibility on the developer.
• Environmental review (changed): Before, data centers could ride CEQA categorical exemptions (blanket exemptions). After: categorical exemptions are barred; fast-track certification requires demonstrable zero-carbon/water performance and community-benefit agreements.
• Governor's stance (changed): Newsom vetoed AB 93 (2025), a Papan water-disclosure bill, saying he was "reluctant to impose rigid reporting requirements." On Sep 21, 2026 he signed the resurrected, expanded versions — an explicit reversal, framed against Trump's deregulatory posture.
• National template (changed): California, the largest US economy and biggest private-AI market (32 of top 50 private AI companies), now has the most comprehensive state data-center statute, a template CalMatters and analysts expect other states to copy.

• Before: Data centers in California operated largely opaque — no statewide water reporting (post-AB 93 veto), no energy-efficiency reporting, no special tariffs; grid-upgrade costs partially socialized; CEQA categorical exemptions available; localities approving projects with NDAs and incomplete water/energy information; 2% of state electricity demand, projected to double in 10 years (CEC).
• Change: Seven-bill signing, Sep 21 2026 — reporting, tariff, water-assessment, scarcity-plan, CEQA, and cost-responsibility rules enacted together.
• After (from 2027–2029 as provisions phase in): Public visibility into data-center water/energy; data centers pay for their own grid and water infrastructure; ratepayers protected from cost shifts; environmental review the default with a demanding clean-infrastructure fast-track; developers weigh California's compliance costs vs. markets elsewhere; other states expected to follow the California playbook; California data-center demand still expected to grow (CAISO: +2.3 GW by 2030; PG&E had ~2,300 MW of data center capacity applications in 2024).

• AB 2383/SB 886 (CPUC tariffs): By Jan 1, 2028, the CPUC must establish separate tariffs for data-center interconnection and transmission/distribution/generation service (threshold set by CPUC, not exceeding 25 MW peak demand). Tariffs require: participating customers to disclose duplicate applications in other territories (SB 886), upfront collateral/prepayment for incremental generation costs, ≥10-year payment mechanisms, early-termination fees, cost-recovery minimums based on projected load, wildfire-mitigation/electrification/societal-cost contributions (AB 2383), and no compensation for onsite generation that shifts costs to other consumers.
• SB 1168 (rate structures): CPUC assesses (in new or existing proceedings) rate structures ensuring data centers pay reasonable transmission/distribution shares and proportionate load-procurement shares, explicitly to alleviate pressure on residential ratepayers (including CARE and FERA program customers).
• AB 2469 (permit gate): Cities/counties may not approve construction or peak-water-increasing expansion unless the applicant provides a water supply assessment, a water scarcity plan (from Jan 1, 2028) covering drought stages from "abnormally dry" to "exceptional drought," projected water use with efficiency measures, workforce disclosures (composition, classifications, duration, wages, local-hire share), and accepts full cost of any required water conveyance/treatment/storage/distribution infrastructure. Applies to charter cities (statewide-concern finding).
• AB 2619 (license reporting): Owners must disclose, under penalty of perjury, expected water use, source, and max-day/max-month/average-year volumes before and at business-license application; renewals require prior-year total and direct water use with cooling-system type (potable/non-potable/recycled by source). Establishes Type I (hyperscale: >10,000 servers or >25 MW), Type II (2–25 MW), Type III (<2 MW) tiers; data-center demand must be included in urban water suppliers' unconstrained-demand forecasts.
• AB 1577 (CEC reporting): Data centers ≥10 MW IT capacity report location/size, installed IT capacity, total energy consumption, PUE, and generator fuel quantities to the CEC; CEC includes a data-center electrical-load trends assessment in the 2029 Integrated Energy Policy Report; trade-secret protection process with no preemption of other disclosure laws.
• SB 887 (CEQA + fast track): Defines "data center" for CEQA; bars categorical exemptions for data-center entitlements; creates an environmental-leadership certification path for data centers meeting 11 conditions (prepaid interconnection, no fossil-fuel consumption increase, ≥4-hour 100%-of-peak zero-carbon storage, demand-response participation, behind-the-meter zero-carbon generation, enforceable full-payment grid commitments with early-termination fees, recycled/waterless cooling, 100% zero-carbon hourly energy within 5 years with 75% newly developed, GHG mitigation per existing leadership-project rules, community benefits agreements, and CARB leakage-review requirements). The Office of Land Use and Climate Innovation, with the CEC, must develop uniform statewide standards for the conditions.

• The AI compute build-out just got its first full reporting and cost-allocation regime in the largest US market: water/electricity disclosure changes site economics; community consent shapes where future clusters go.
• It is the state-level counterpoint to federal deregulation — a direct, explicit political-legal clash (Newsom vs. Trump) over who bears the costs of AI infrastructure, with California asserting community and ratepayer protections as "the driver's seat."
• It creates a compliance template other states will copy — Virginia, Texas, Utah and others now have off-the-shelf language to adopt during the AI-infrastructure boom.
• It ties into the wider week's AI-governance wave: N-9-26 (AI kill-switch EO, Sep 18) + this package (Sep 21) + OpenAI's call for a global standards push (Sep 21) + the "slow down/lab safety" controversy — compute infrastructure and model safety are being regulated in the same window.
• The economic stakes are huge: the Data Center Coalition puts California's 2024 industry contribution at ~665,500 jobs, >$159B economic activity, and $14.1B in taxes; the same group warns the laws will push development — and those benefits — to neighboring states.
• At the same time, national backlash is already biting: a May Gallup poll found ~7 in 10 Americans oppose data centers in their communities, and a Sep 22 LA Times report said AI backlash had put $68 billion in US data-center projects on hold.

• Comparable data: First time the public, local governments, and regulators can see actual PUE, water use, and energy consumption of California data centers (2029 IEPR + license cycle disclosures).
• Ratepayer protection at scale: A legal mechanism (prefunded, collateralized, ≥10-year commitments) that prevents stranded costs from prematurely retired or under-utilized data centers from landing on residential bills — replicable in every state with investor-owned utilities.
• Clean-infrastructure "good neighbor" certification: A defined, enforceable path for a data center to earn CEQA streamlining through 100% zero-carbon power, storage, recycled water, and community-benefit agreements — an incentive design other states can adopt to reward green AI infrastructure.
• Informed land-use decisions: Local communities can weigh economic benefits against water/energy costs before approving, instead of signing NDAs.
• Political differentiation: California and the federal government now have opposing, competing frameworks for data-center governance, sharpening a 2028-election issue.

1. What MW threshold does the CPUC actually set for data-center tariffs (must be ≤25 MW)?
2. Which data centers are caught by AB 1577's 10 MW floor in practice, and what does the 2029 CEC load-trend assessment reveal about true California AI energy demand?
3. How will the "no increase in fossil fuel consumption" and "100% zero-carbon within 5 years, 75% newly developed" conditions be verified, and who audits compliance?
4. Will the federal EO/Trump administration attempt to preempt or challenge these laws, and under what theory?
5. How many Governor certifications under SB 887 materialize — does the fast track actually accelerate or merely codify?
6. Does a copycat wave hit other states in 2027, and does that converge or diverge national data-center policy?
7. Will $68B in paused US projects resume, and does California gain or lose relative share?

• Short term (2026-2027): CPUC opens rulemakings for data-center tariffs (due Jan 1, 2028); CEC designs AB 1577 reporting and 2029 IEPR methodology; Governor's Office (Land Use & Climate Innovation) drafts uniform SB 887 standards; industry litigation/lobbying on tariff specifics; other states' 2027 sessions introduce copycat bills.
• Medium term (2027-2029): License-cycle water disclosures begin to surface; water scarcity plans required from Jan 1, 2028; earliest CEQA certification decisions; first CEC load-trend assessment (2029) publicly sizes California's data-center energy demand; the $68B paused-project backlog resolves.
• Long term: Likely federal-state confrontation on AI/data-center jurisdiction; a possible national "patchwork vs. floor" debate; California remains the transparency/cost-allocation benchmark, while build-out migrates to states with cheaper power and looser rules — unless tariffs become the baseline everywhere.
• Signal to watch: Trump-administration response (December-2025 EO preemption test, federal data-center zoning/eminent-domain push); the first SB 887 Governor certification; CEC 2029 numbers.

For two years the AI story has been "compute is power and water, and nobody knows how much." California just made "nobody knows" illegal: the seven-bill package converts AI infrastructure from a secret into a measured, priced, and consent-based industry — the first comprehensive answer to the question every community in America is now asking about data centers. The honest tension the story should carry: California's laws protect ratepayers and shine light on water and power, but they also make the state a harder place to build the very compute the AI boom needs, and the industry's flight-risk warning is not baseless. The deepest takeaway is that this is a template, not an endpoint: within two years, either other states copy the good parts (disclosure, cost allocation, green fast-tracking) and create a national floor through diffusion, or the patchwork deepens and the AI infrastructure race bifurcates into "regulated-rich" and "deregulated-cheap" states. Whoever reads the room correctly — regulators, operators, or the communities themselves — decides which future wins.