Data centers for AI could nearly triple San José’s energy use. Who foots the bill?

PG&E and San José officials say a wave of proposed AI data centers could nearly triple the city’s peak electricity use, with utilities reporting 18.7 GW of requested service capacity (enough for roughly 18 million homes) though regulators currently forecast 4–6 GW by 2040 due to uncertainty from speculative project filings and rapidly evolving AI loads. The pipeline is prompting major grid upgrades and a policy fight over who pays—consumer advocates warn ratepayers could be left footing billions for infrastructure that may never be used—while environmental and reliability concerns (water use, diesel backup emissions and potential reliance on fossil or controversial “firm” resources) complicate California’s clean‑energy goals. State agencies are expected to revisit transparency, cost allocation and market rules, so investors should price in elevated grid capex, regulatory uncertainty and locational constraints that could affect data‑center deployment and offtake economics.

Analysis

PG&E and San José officials report a pipeline of data-center requests that could nearly triple the city’s peak electricity use, with utilities noting 18.7 GW of requested capacity (equivalent to roughly 18 million homes) while state regulators model a more conservative 4–6 GW by 2040 due to speculative filings and uncertain AI load growth. Forecasting is challenged by rapidly evolving AI application layers and variable cooling needs, creating a wide gap between requested service and likely realized demand.

The pipeline is already driving plans for major grid upgrades and a policy fight over who bears the costs: the Public Advocates Office warns ratepayers could be left with billions for infrastructure that may never be used, while PG&E argues large customers can lower per-customer fixed costs if connected to available capacity; the grid averages ~45% utilization but faces acute local peak constraints. Legislative attempts to increase transparency stalled, and several state bodies (CPUC, CEC, Little Hoover Commission) are expected to revisit cost allocation, disclosure and market rules.

Environmental and reliability risks include concentrated air pollution from diesel backups, rising data-center emissions (reported nearly doubling from 2019–2023) and potential reliance on ‘‘clean, firm’’ resources—nuclear, geothermal, or gas with carbon capture—or large solar-plus-battery projects to maintain 100% carbon-free goals. San José’s 60% renewable mix and emphasis on demand flexibility (shifting loads off hot afternoons) are practical mitigants but scaling them depends on regulatory incentives and contract structures.

For investors, this means elevated utility and transmission capex, regulatory uncertainty around cost allocation, locational constraints on where data centers can reliably connect, and a multi-year policy process that will determine which projects proceed versus remain speculative.

Key Decisions for Investors

• Monitor forthcoming CPUC, California Energy Commission and Little Hoover Commission actions and pause increasing long-dated utility exposure until cost-allocation and transparency rules are clarified
• Reassess exposure to PCG for potential higher grid capex and regulatory/rate-case risk and consider hedging or trimming positions if rule changes increase the likelihood that ratepayers will fund speculative buildouts
• Favor opportunities in providers of clean, firm generation, large-scale solar-plus-battery projects, demand-response technologies and transmission/grid-upgrade contractors that are likely to benefit from required infrastructure investments
• Track actual buildout versus the 18.7 GW request (compare realized additions to the regulators' 4–6 GW forecast) and stress-test portfolios for scenarios where many announced projects never materialize but upgrade costs are socialized