California Market

Market Deep Dive

A 10 GW Pipeline in the Heart of Silicon Valley

California — led by Silicon Valley and the greater Bay Area — is re-emerging as a major data center growth market after years of being considered too expensive and power-constrained. PG&E announced in July 2025 that it is working to serve 10 GW of new data center demand over the next ten years — enough to power roughly 7.5 million homes — up from 8.7 GW reported in May 2025 and 5.5 GW in February 2025.

Crucially, part of this pipeline is close to real: PG&E reports 17 data center projects totaling ~1.5 GW are in the final engineering phase (the last step before construction), projected to begin operations between 2026 and 2030. Most are in San Jose, Silicon Valley, and the Bay Area, with some in the Central Valley and Sacramento. CalMatters reports AI data centers could nearly triple San Jose's energy consumption.

Sources: PG&E Corporation press release (Jul 31, 2025); CalMatters (Dec 2025).

Project Pipeline: From Speculative HPC to the Santa Clara Cluster

California's tracked pipeline is anchored by a mix of speculative AI campuses and some of the most mature colocation markets in the world. Our database currently flags a 250 MW Silicon Valley HPC site in San Jose as Speculative (confidence score 35/100), with an estimated $2.0B investment; signals include a PG&E large-load inquiry (Jul 2026) and a CAISO cluster study (May 2026). The project, linked to an unconfirmed major AI company, illustrates the core California problem: demand is real, but power delivery timelines are the binding constraint. See the full record on DataPowerDemand Projects.

The established heart of the market remains Santa Clara and the Silicon Valley colocation cluster, where Equinix, Digital Realty, Stack Infrastructure, Vantage Data Centers and others operate dense multi-campus footprints Confirmed (operator portfolios). Vantage's Santa Clara campus and Digital Realty's Silicon Valley portfolio together account for hundreds of MW of built capacity Estimated, making the city one of the largest data center load pockets on the West Coast.

Beyond the Bay, Prime Data Centers' Sacramento campus positions the Central Valley as a lower-cost spillover market Confirmed (company announcements), while Los Angeles continues to grow as an interconnection and latency hub. PG&E has said its final-engineering projects cluster in San Jose, Silicon Valley, the Bay Area, the Central Valley and Sacramento — consistent with the pattern below.

Sources: DataPowerDemand project records; company announcements; PG&E (Jul 2025); Silicon Valley Business Journal (2025).

PG&E Deep Dive: Converting a 10 GW Pipeline into Served Load

PG&E's investor communications trace a steep ramp: 5.5 GW of prospective data center demand in February 2025, 8.7 GW by May, and 10 GW over ten years by July 2025 Estimated (company projections). Management describes the funnel as spanning early-stage inquiries through final engineering; only the 17 projects / ~1.5 GW in final engineering carry firm scopes, cost estimates and 2026–2030 in-service dates.

The financial logic rests on utilization. PG&E argues its system runs at roughly 45% average utilization, so high-load-factor data centers can absorb fixed grid costs without proportional new infrastructure. On investor calls, the utility has framed Rule 30 as the delivery mechanism: large-load customers fund transmission up front, receive accelerated interconnection, and — in PG&E's telling — the incremental revenue compresses rates for everyone else by 10%+ over time.

Execution risk is substantial. PG&E carries the sector's heaviest wildfire-liability overhang and the highest residential rates among large continental-U.S. utilities Confirmed (EIA data). Investors will watch quarterly pipeline disclosures, final-engineering conversion rates and the CPUC's permanent Rule 30 decision as the market's de-facto progress markers.

Management has framed the buildout as self-funding: data center customers prepay for dedicated transmission under Rule 30, limiting stranded-cost exposure for existing ratepayers. The open question is how much shared network upgrade cost — substations, 230 kV lines, system reinforcements — ultimately lands in general rates, which is precisely what the CPUC's permanent rulemaking must resolve Probable.

Sources: PG&E investor presentations & earnings calls (Feb–Jul 2025); EIA Electric Power Monthly (2025).

An Unusual Pitch: Data Centers Could Lower Bills

PG&E is making a contrarian, utilization-based argument: because its grid is only about 45% utilized on average, spreading fixed grid costs across a large new base of high-load-factor data center demand could lower customer electric bills by 10% or more, with every 1 GW of new data center load potentially saving customers 1–2% on monthly bills long-term. PG&E also touts 50,000 construction jobs, $1.25–1.75B in added property-tax revenue, and $2.5–3B in added sales-tax revenue.

To accelerate connections, the CPUC approved interim implementation of PG&E's Electric Rule 30, which lets large-load applicants who fund necessary transmission infrastructure up front get accelerated grid connections. Separately, Silicon Valley Power (Santa Clara) remains a distinct municipal utility serving the densest part of the Valley with competitive rates.

Sources: PG&E (Jul 2025); CPUC Rule 30 interim decision; NCPA.

Silicon Valley Power: The Municipal Incumbent

Silicon Valley Power (SVP), the City of Santa Clara's municipal electric utility, is the quiet giant of California data center power. It already serves the state's densest concentration of colocation load and has publicly discussed a large-load request queue measured in the hundreds of MW to ~1 GW-plus range Probable (city/utility statements as reported by local press). Unlike PG&E, SVP owns local generation — including the ~147 MW Donald Von Raesfeld combined-cycle plant Confirmed — and buys additional supply through the Northern California Power Agency.

SVP's historic pitch is rates materially below PG&E's (often cited at roughly 15–25% lower for large users) Probable, plus a reliability record shaped by serving mission-critical load for two decades. Its constraint is physical: the Santa Clara substation and 230 kV delivery network must be expanded serially, and the utility has cautioned that very large single-site requests face multi-year timelines.

For developers, the practical implication is a two-track Bay Area strategy: Santa Clara/SVP for the most latency-sensitive colocation, and PG&E territory (San Jose, Central Valley) for larger AI campuses where Rule 30 acceleration can compress schedules. Compare utility profiles on DataPowerDemand Markets.

SVP's municipal structure is a quiet advantage: decisions run through a city utility rather than a CPUC-regulated IOU rate case, which historically allowed faster, more flexible large-load contracts. The trade-off is scale — SVP's total system peak is a fraction of PG&E's, so a handful of 100–300 MW AI campuses can absorb most of its near-term headroom Estimated.

Sources: City of Santa Clara / Silicon Valley Power materials; Silicon Valley Business Journal (2024–2025); NCPA.

Regulatory & Policy Landscape: Rule 30, CAISO, CEC, BAAQMD, Title 24

California's data center siting stack is the most layered in the U.S. At the top sits the CPUC's Rule 30 proceeding, which will define permanent cost-allocation rules for large loads in PG&E territory — how much transmission developers prepay, refund terms, and minimum-take contract structures Confirmed (interim decision issued; permanent rulemaking pending). Wholesale interconnection runs through the CAISO queue, where cluster-study reforms have slowed near-term deliverability for new large loads.

The California Energy Commission shapes demand-side expectations through forecasting and efficiency standards, while Title 24 building-efficiency rules push new facilities toward economization, higher-temperature operation and on-site renewables Confirmed. In the Bay Area, the Bay Area Air Quality Management District (BAAQMD) is a decisive permitting gate: diesel backup-genset permitting is tightly constrained, forcing designs toward fewer engines, cleaner fuels, or battery-backed ride-through Confirmed (district rules; industry reporting).

The practical consequence: California projects must clear air-permitting and building-code hurdles that Texas and Arizona largely avoid, adding 12–24 months for power-reliant campuses Probable. Developers increasingly pre-negotiate genset strategies and storage sizing before land close.

• CPUC Rule 30 — large-load cost allocation & acceleration (interim approved; permanent pending)
• CAISO queue — cluster-study reform; multi-year deliverability for new load
• CEC / Title 24 — efficiency, economization, on-site clean-energy expectations
• BAAQMD — strict backup-diesel permitting in the Bay Area air basin
• Local zoning — Santa Clara/San Jose substation-siting and noise constraints

Sources: CPUC dockets; CAISO interconnection materials; CEC; BAAQMD rules; DCD (2024–2025).

The Clean-Energy Mandate Tension

California's climate framework collides head-on with data center growth. SB 100 requires 100% zero-carbon retail electricity by 2045, with 60% renewables by 2030 Confirmed. Layered on top is the state's decision to extend Diablo Canyon — 2.2 GW of firm nuclear — from its original 2024/2025 retirement to 2029 (Unit 1) and 2030 (Unit 2), with possible further extension under federal support Confirmed (state legislation; DOE credits).

Meanwhile, several GW of aging gas capacity are slated for retirement this decade, tightening evening net-load ramps even as data centers add flat, 24/7 demand Probable (CAISO planning). The state's answer is storage: California leads the U.S. in grid batteries, with 10 GW-plus installed Probable, and data centers are increasingly expected to pair load with storage and clean PPAs.

New large loads may also face expectations — formal or informal — to bring their own clean supply. Regulators and advocates have floated requiring data centers to procure incremental zero-carbon generation or storage as a condition of accelerated interconnection, mirroring clean-energy pairing debates now live in several Western states Possible.

For operators, the tension is reputational as much as operational: AI campuses drawing from CAISO's gas-heavy evening mix face scrutiny under SB 100's trajectory. We explore the carbon-accounting trap in our blog on data center carbon realities.

Sources: SB 100 statute; CPUC/CEC Diablo Canyon decisions (2022–2023); CAISO; Utility Dive (2024–2025).

Water, Cooling & Siting Constraints

Water is California's quiet siting filter. Evaporative-cooled hyperscale facilities can consume millions of gallons per day at peak Estimated (industry norms), a hard sell in drought-prone counties. Bay Area and Central Valley jurisdictions increasingly steer projects toward air-cooled or closed-loop designs, trading higher fan energy for near-zero water use — a compromise AI halls with 50–100 kW racks often accept via direct-to-chip liquid cooling Probable.

Siting is equally constrained by land and neighbors: Santa Clara's remaining industrial parcels are scarce and expensive, pushing larger campuses toward San Jose's north valley, the Central Valley (Tracy, Stockton corridors) and Sacramento. Seismic design requirements and wildfire-driven PSPS (Public Safety Power Shutoff) risk add cost layers absent in Texas or Arizona. See our analysis of data center water consumption for cooling-design tradeoffs.

Noise and air rules compound siting. Backup-generator testing is limited by air-district permits, and cooling-tower plumes and fan noise have drawn local opposition in dense Santa Clara neighborhoods Probable (planning-commission reporting). The result is a premium on brownfield and industrial-reuse sites with existing substations, where entitlement timelines can be cut from years to months Estimated.

Sources: CEC water/energy research; local planning documents; DCD (2024–2025).

Competitive Positioning vs. Texas & Arizona

California's disadvantages are structural: industrial retail rates among the highest in the continental U.S. (roughly 2–3× Texas levels) Confirmed (EIA), layered permitting, seismic codes, and BAAQMD limits on backup diesel. Texas offers cheaper power and faster ERCOT interconnection; Arizona offers cheap land, low water-cost air-cooling norms, and aggressive incentives.

California's counter is irreplaceable demand-side gravity: proximity to AI customers and talent, the lowest-latency paths to West Coast users and Asia-Pacific subsea cables, and SVP's comparatively competitive municipal rates in Santa Clara. PG&E's utilization pitch — if the 10 GW pipeline converts — could narrow the rate gap by spreading fixed costs Estimated.

The wildcard is time-to-power. If Rule 30 delivers connections materially faster than Texas's congested ERCOT large-load queue, California's higher all-in power cost becomes tolerable for latency- and customer-anchored AI workloads Probable. If not, marginal growth continues leaking to Phoenix, Dallas and the Pacific Northwest.

Sources: EIA Electric Power Monthly (2025); CBRE North America Data Center Trends (2025).

Outlook

California's data center growth hinges on whether PG&E can convert its 10 GW pipeline into served load while holding its "lower bills for all" promise — a politically delicate claim in the highest-rate state in the continental U.S. Watch Rule 30's permanent form, interconnection timelines in Santa Clara and San Jose, and whether the 1.5 GW in final engineering actually reaches operation on the 2026–2030 schedule. Wildfire-driven liability and clean-energy mandates add layers other markets don't face.

Over the next 12–24 months, we expect the ~1.5 GW final-engineering cohort to begin breaking ground, SVP's Santa Clara queue to keep the city supply-constrained, and speculative AI projects like the 250 MW San Jose HPC site to either firm up with Rule 30 agreements or stall on deliverability Probable. A CPUC permanent Rule 30 decision, CAISO cluster-study outcomes, and BAAQMD permitting test cases are the three catalysts to watch.

Net view: California remains a capacity-constrained but demand-anchored market. Hyperscale AI growth will be slower than Texas but stickier — anchored by customers who cannot leave the Valley.

Sources: DataPowerDemand analysis; PG&E (Jul 2025); CPUC; CAISO (2025–2026).