Data centers are black boxes, but California wants to change that

California Gov. Gavin Newsom signed bills requiring data-center operators to begin disclosing more information next year on their electricity and water usage. The measures respond to community concerns that AI-related data-center expansion could increase power bills, strain electricity grids, and pressure local water supplies, though the rules will not provide a complete usage picture. The policy could raise transparency and compliance requirements for data-center operators in California.

Analysis

The investable consequence is not the disclosure burden itself, but the precedent it creates for cost allocation. If California regulators can isolate data-center load and water intensity, utilities will face stronger pressure to require upfront interconnection payments, demand-response commitments, and dedicated generation/storage procurement rather than socializing grid upgrades across retail ratepayers. That raises the delivered-power cost for hyperscalers and weakens the economics of marginal California capacity, favoring data-center development in power-surplus markets such as Texas, PJM, and parts of the Midwest.

For Digital Realty (DLR) and Equinix (EQIX), the relevant risk is a 6-18 month increase in permitting lead times and customer pass-through friction rather than an immediate revenue hit; their diversified footprint should allow migration of incremental workloads, but California assets could warrant lower utilization assumptions if power contracts become less certain. The larger second-order winners are merchant generation and flexible-capacity providers—Vistra (VST), Constellation Energy (CEG), and gas-turbine/storage supply chains—where large-load customers increasingly need firm, contractable power rather than generic renewable PPAs. Hyperscalers with owned generation, long-dated nuclear contracts, or superior geographic flexibility should have a relative advantage over smaller AI infrastructure operators dependent on local utility interconnection queues.

Consensus may overstate the near-term regulatory threat to AI capex: reporting requirements alone do not cap consumption, and large customers can absorb modest compliance costs. The sharper risk is that transparent load data makes politically difficult rate cases and water permits harder to approve after the next heat-wave, drought, or reliability event; that would produce episodic project delays and localized valuation discounts, not a broad AI-demand reversal. Falsification for the restrictive-power thesis would be California utility filings showing data-center load served under fully customer-funded tariffs with no material delay in new-service approvals.

Key Decisions for Investors

• No broad short in AI infrastructure on this development; treat it as a location-specific permitting watch item rather than a demand signal over the next 1-3 months.
• Maintain a 6-18 month relative long VST or CEG versus DLR, sized modestly: firm-power scarcity and direct large-load contracting should gain value if California-style cost-allocation rules spread, while colocation operators face incremental siting friction. Exit the relative thesis if utility tariffs demonstrate full, rapid pass-through without interconnection delays.
• Monitor PG&E (PCG), Edison International (EIX), and Sempra (SRE) regulatory filings for dedicated data-center tariffs, transmission-capex recovery, and queue timing. A tariff that requires customer-funded grid upgrades would be a potential catalyst for utility rate-base visibility but a negative for California data-center expansion.
• Watch DLR and EQIX quarterly disclosures for California leasing, power-cost pass-through, and development-pipeline changes. A material reduction in West Coast bookings or a rising share of delayed projects would justify reassessing a short DLR/EQIX basket against long VST/CEG; absent that evidence, do not initiate.