California’s AI Data Center Boom Collides With Water Scarcity

Google News has surfaced a sharp conflict in California: proposed AI data centers need dependable cooling where communities already face severe water constraints.

The immediate flashpoint is the RB Inyokern Data Center, a proposed 99-megawatt facility in eastern Kern County. Its developer forecasts annual water demand of roughly 50 acre-feet. Residents question whether that estimate captures peak cooling needs in the Mojave Desert.

That disagreement is larger than one project. California has more than 200 active data centers, while developers are proposing new facilities in communities shaped by drought and groundwater overdraft. Google News readers are encountering a fight between AI infrastructure growth and local water security.

The central question is not whether every data center consumes an extraordinary amount of water. Cooling design, climate, workloads, electricity sources, and operating schedules produce very different results.

The harder question is whether communities receive enough verified information before approving projects that create permanent demands. In Inyokern, public documents provide a water estimate, but residents dispute its assumptions and want a fuller environmental review.

This is where the AI boom meets physical limits. A cloud service feels detached from geography, yet every model runs in a building connected to a particular electrical grid and water system.

The Inyokern Proposal Puts a Number on the Conflict

The dispute begins with a 99-megawatt project and two sharply different estimates of its potential water demand.

R&L Capital proposes building the RB Inyokern Data Center on approximately 50 acres near the intersection of US Highway 395 and State Route 178. The planned building would cover about 238,000 square feet and contain six modular data halls.

According to the developer’s filings, those halls would support cloud computing and artificial intelligence workloads. Construction is scheduled to begin in 2027, with full operations anticipated in 2031.

The developer says the facility would use a hybrid cooling system. Hybrid systems can combine dry heat rejection with water-assisted cooling when temperatures or server loads rise.

Its application estimates annual water demand at roughly 50 acre-feet. One acre-foot equals about 326,000 gallons, making the estimate approximately 16.3 million gallons per year.

The Inyokern Community Services District issued a commitment to provide about that amount. Project documents characterize the demand as a small share of projected basin extraction.

Residents do not accept that framing. Nearly 90 early comments submitted to state officials raised objections, according to the Inyokern water dispute reported by SJV Water and KVPR.

Some commenters estimated demand closer to 500,000 gallons per day. That would equal approximately 560 acre-feet annually, assuming steady daily consumption.

Another estimate applied an academic average of 18.6 acre-feet per megawatt to the proposed 99-megawatt load. That calculation produced more than 1,800 acre-feet per year.

Those higher figures do not establish the project’s actual consumption. They rely on generalized assumptions that may not match the proposed cooling equipment or operating schedule.

The developer’s lower number also deserves scrutiny. The public needs the engineering assumptions behind it, including temperature thresholds, peak-day demand, water recycling, and expected hours of evaporative operation.

This distinction matters because annual averages can hide infrastructure stress. A system that normally uses little water might draw far more during the hottest days, when community demand is also elevated.

The project’s descriptions have created additional confusion. A public FAQ refers to dry cooling with evaporative assistance during peak temperatures. The original application describes hybrid cooling towers and water-cooled chillers but provides limited operational detail.

Dry cooling transfers heat into the air without continuously evaporating water. It conserves water but usually requires more electricity, especially in hot weather.

Evaporative cooling uses water to remove heat efficiently. It can lower electricity demand, but some water leaves the system as vapor and cannot immediately return to the local supply.

A hybrid system switches between these approaches. Its real water footprint depends on when those switches happen and how frequently extreme temperatures trigger water-assisted cooling.

The project is seeking a small power plant exemption because its backup system falls within the California Energy Commission’s jurisdiction. State records show an active Inyokern project docket with continued public comments through July 2026.

That process does not make the commission the general permit authority for the data center. It does, however, create a state-level environmental review tied to the project’s power system.

The proposed backup installation has also attracted attention. Project materials describe dozens of diesel-fired generators capable of supplying up to 99 megawatts during emergencies.

If the state grants an exemption, the developer would still need local approvals. Kern County and the Indian Wells Valley Groundwater Authority would retain important roles.

The project therefore remains a proposal, not a finished facility with measured consumption. Every water figure currently under discussion is a forecast.

That is the first important conclusion for readers. The disagreement is not between an established measurement and a rumor. It is between competing estimates made before final design, permitting, construction, and operation.

Why Google News Is Carrying a Local Water Story Nationwide

Google News is amplifying a local dispute because California lacks a clear system for measuring data center water use before conflicts emerge.

The California Energy Commission says the state has more than 200 active data centers. California trails only Texas and Virginia among leading data center markets, according to the commission’s infrastructure overview.

AI workloads are raising the stakes. Graphics processors produce dense heat loads, while model training and inference can keep computing equipment busy for long periods.

More computation produces more heat. Every facility must move that heat away from servers, whether it uses air cooling, liquid cooling, evaporation, or some combination.

California requires certain large buildings to disclose energy use through its benchmarking program. However, the state does not maintain a complete public inventory of data centers.

Water oversight is even more fragmented. Local utilities, groundwater agencies, cities, counties, and state bodies each control part of the approval landscape.

A 2026 Berkeley Law analysis found no California regulatory system designed specifically to manage data center water impacts. Instead, projects encounter a patchwork of rules created for water service, groundwater, land use, and environmental review.

The California water analysis also identified a significant information gap. Not every operator must disclose facility-level water consumption.

That gap leaves decision-makers comparing developer forecasts, academic estimates, utility commitments, and corporate sustainability reports. Those sources often use different boundaries and measurement methods.

One estimate may include only water consumed on site. Another may include water used by power plants that generate the facility’s electricity.

Water withdrawal and water consumption are also different. Withdrawal measures water taken from a source, while consumption measures the portion not returned for near-term reuse.

A cooling tower might withdraw water and later discharge part of it. Evaporated water counts as consumed because it no longer remains available to that local system.

These definitions can make apparently conflicting statements simultaneously accurate. A company might report low direct consumption while an analyst calculates a much larger total footprint.

California researchers estimate that off-site consumption from electricity generation dominates the state’s data center water footprint. The location and technology of power generation therefore matter alongside the cooling system.

Next 10 estimated that total California data center water consumption rose from 25.42 billion liters in 2019 to 49.91 billion liters in 2023. That represents a 96.4 percent increase.

Its analysis estimated that on-site consumption increased from 1.33 billion liters to 2.84 billion liters during the same period. The remaining footprint was associated primarily with electricity production.

Those statewide estimates do not predict Inyokern’s specific demand. They show why a narrow focus on the facility’s water connection can miss part of the resource burden.

Location determines how consequential that burden becomes. A gallon consumed in a water-rich region does not create the same risk as a gallon consumed from an overdrafted desert basin.

Indian Wells Valley has spent years confronting groundwater imbalance. Local households, businesses, the Naval Air Weapons Station China Lake, and other users depend heavily on the basin.

That history changes how residents interpret a commitment for 50 acre-feet. They are not evaluating an isolated number against California’s total supply.

They are asking whether a new industrial demand fits within a local plan already shaped by scarcity, pumping limits, fees, and legal disputes.

This explains why the story moved beyond a local permitting docket. It combines the most visible technology investment cycle with one of the American West’s oldest resource conflicts.

AI Cooling Trades Water Savings for Electricity Demand

The central tradeoff is not AI versus water alone; it is water-efficient cooling versus the electricity and equipment needed to make it work.

Data center cooling begins with a simple problem. Computing equipment converts most of its electricity into heat, and that heat must leave the building continuously.

Traditional air cooling circulates chilled air around server racks. It remains common, but dense AI systems can make air movement less efficient.

Direct liquid cooling moves fluid near processors or other hot components. Liquid carries heat more effectively than air and can reduce dependence on room-scale cooling.

Closed-loop systems recirculate coolant rather than continually consuming fresh water. However, the captured heat still needs a final destination outside the server hall.

A facility can reject that heat through dry coolers, cooling towers, chillers, or multiple systems working together. Each option carries costs and operational constraints.

Dry cooling protects local water supplies because it transfers heat into the surrounding air. Its performance falls as outside temperatures rise.

That limitation becomes important in Inyokern. Summer heat can reduce dry cooling efficiency exactly when both server cooling and community electricity demand are high.

Operators can install more equipment to compensate, accept higher electricity use, or add evaporative assistance. The project’s final design will determine which option dominates.

Evaporative systems work efficiently in dry climates because water absorbs substantial heat when it changes into vapor. Yet the process consumes water rather than returning it locally.

Hybrid systems promise a compromise. They operate in dry mode when conditions allow and use water during hotter periods.

The phrase “during peak temperatures” sounds limited, but it needs numbers. Reviewers require the temperature threshold, expected annual hours, peak-day withdrawal, and behavior during heat waves.

They also need to know what happens if computing density increases. A building designed for one server configuration can receive more demanding hardware after construction.

AI hardware refreshes create uncertainty because processors and cooling technologies change faster than water infrastructure. A facility may operate for decades while replacing servers every few years.

Power usage effectiveness, or PUE, compares total facility electricity with electricity delivered to computing equipment. The developer projects a PUE of approximately 1.41.

That figure remains a design claim until operating data verifies it. PUE also says little about water consumption unless it is paired with water usage effectiveness.

Water usage effectiveness measures annual site water consumption relative to computing energy. Publishing both measures would make the cooling tradeoff easier to evaluate.

A facility could achieve a favorable PUE by relying more heavily on evaporation. Another could conserve water while consuming more electricity through dry cooling.

Neither metric alone establishes environmental performance. Reviewers need the water source, seasonal pattern, energy source, and local scarcity context.

The Inyokern proposal includes one megawatt of on-site solar generation. That is small compared with a potential 99-megawatt facility load and does not resolve the larger electricity question.

The project expects primary service from Southern California Edison. Its diesel generators would provide backup capacity rather than continuous generation, according to the application.

Backup systems still matter because air permits, testing schedules, emergency operation, and local pollution affect neighboring communities. However, diesel is a supporting issue in this water-centered debate.

The main opponent remains the developer’s efficiency promise versus the community’s demand for verifiable resource limits.

That framing avoids a common mistake. Data centers are not interchangeable machines with a universal water requirement.

A 99-megawatt nameplate does not reveal annual utilization, server density, cooling hours, or peak water consumption. Generalized calculators can illustrate risk, but they cannot replace project engineering.

The reverse is also true. A developer’s annual estimate cannot substitute for enforceable limits and transparent operating data.

The strongest resolution would connect approvals to measurable performance. Conditions could specify annual consumption, peak demand, cooling technology, reporting frequency, and corrective action after overruns.

Without those conditions, the public must trust a forecast that cannot be tested until the facility starts operating. By then, the community may already have expanded infrastructure around it.

The Missing Data Matter More Than the Biggest Estimate

The most credible concern is not that the highest water forecast must be correct; it is that residents cannot independently test the lowest forecast.

Public opponents have cited estimates far above the developer’s 50 acre-feet per year. Those figures communicate the scale of possible risk, but they should not be treated as confirmed consumption.

The estimate of roughly 500,000 gallons per day assumes substantial evaporative operation. The project says its hybrid design would rely primarily on dry cooling.

The estimate above 1,800 acre-feet applies a published average to the project’s full 99-megawatt capacity. That method does not incorporate Inyokern’s proposed equipment or operating strategy.

A careful analysis should resist both extremes. It should not present a generic industry ratio as a facility measurement.

It should also reject the idea that a developer’s projection becomes reliable merely because it appears in an application.

The missing bridge is a detailed, reviewable water model. That model should explain how air temperature, humidity, server load, maintenance, and equipment degradation affect consumption.

It should include an average year, a severe heat year, maximum monthly demand, and maximum daily demand. It should also identify uncertainty around future hardware.

The local water supplier’s capacity deserves equal attention. A commitment to serve confirms willingness, but it does not automatically establish long-term basin sustainability.

Reviewers should examine the source of the water, pumping rights, physical delivery capacity, and effects on other users. They should also assess drought and infrastructure failure scenarios.

California lawmakers have recognized this transparency problem. AB 2619 proposes requiring data center operators to provide expected water demand and the anticipated source before seeking certain local approvals.

The bill’s water disclosure rules also cover maximum-day, maximum-month, and average-year projections. Existing facilities would report annual direct water use during license renewals.

The proposal does not decide whether a facility should receive water. It gives suppliers and local governments consistent information before making that decision.

A previous disclosure effort, AB 93, was vetoed in 2025. Governor Gavin Newsom expressed concern about imposing operational reporting requirements without understanding their effects on businesses and consumers.

That veto illustrates California’s competing priorities. State leaders want to retain AI investment while communities want clearer limits on resource commitments.

The economic case for Inyokern includes a projected 48-month development and construction period. The developer forecasts an average construction workforce near 300, peaking around 600.

It also forecasts 30 to 60 permanent operating jobs. Those remain developer estimates and should be evaluated alongside public infrastructure, water, and environmental costs.

Job projections matter in eastern Kern County, but they do not settle the water question. A project can offer economic value and still require stricter resource accounting.

Comparisons with other California proposals show why scrutiny is increasing. A planned 330-megawatt Imperial Valley facility has sought hundreds of millions of gallons annually after recycled-water plans encountered obstacles.

That facility and RB Inyokern involve different developers, water sources, climates, and cooling designs. They should not be treated as one project.

They do demonstrate a shared siting challenge. Developers see available land, energy connections, tax benefits, and construction opportunities in inland communities.

Residents see long-term industrial demand entering places where water already shapes agriculture, housing, public services, and local identity.

A 2026 Next 10 study examined operating and planned facilities through water availability and social vulnerability. Its environmental impact findings warn that statewide totals can obscure concentrated local burdens.

This distributional question is more important than declaring data centers uniquely thirsty. Agriculture, cities, energy production, and ecosystems already compete for California water.

Data centers add a new demand with unusual characteristics. They need high reliability, can create sharp cooling peaks, and may provide fewer permanent jobs than traditional industrial facilities.

They also support services used throughout the economy. The benefits of cloud computing and AI can spread nationally, while water and infrastructure costs remain local.

That imbalance strengthens the case for public reporting. Communities asked to host infrastructure should be able to see how projected benefits compare with enforceable resource commitments.

Transparency would also help responsible developers. Verified performance could distinguish water-conscious projects from designs that rely on vague efficiency claims.

What California Should Watch Next

Three signals will determine whether the Inyokern debate produces accountable infrastructure or another unresolved fight over competing estimates.

The first signal is the California Energy Commission’s environmental review. Its treatment of cooling assumptions, peak water demand, and cumulative effects will establish the factual foundation for later decisions.

A detailed analysis would strengthen confidence in the process, even if reviewers ultimately accept the developer’s annual estimate. A narrow analysis would leave the central dispute unresolved.

The second signal is the project’s final cooling design. Readers should watch for equipment specifications, dry-mode performance, evaporative thresholds, and a complete water balance.

Any material change between the application and construction should trigger updated analysis. Cooling equipment cannot remain a marketing description when it controls the project’s central environmental risk.

The third signal is California’s disclosure policy. AB 2619 and related proposals will show whether the state adopts standardized facility-level reporting.

Passage would not eliminate conflicts over siting. It would make those conflicts more evidence-based by requiring consistent projections and operating results.

Failure would leave California dependent on fragmented local requests. Well-resourced jurisdictions might demand detailed data, while smaller communities could negotiate with less information.

Google News will continue carrying stories about data centers, electricity, and water because AI infrastructure is expanding faster than its accountability systems.

Readers should treat dramatic water comparisons cautiously. The most alarming estimate is not automatically the most accurate, and the smallest estimate is not automatically safe.

The decisive evidence will come from enforceable limits and measured operations. Peak-day demand deserves particular attention because desert heat creates simultaneous pressure on water and electricity systems.

California should also track indirect consumption from power generation. A facility that saves water on site may shift part of its footprint to the electrical system.

For developers, the lesson is straightforward. Water strategy must become a primary siting and design decision, not a secondary answer supplied during permitting.

For local officials, a service commitment should begin the analysis rather than end it. They need drought scenarios, infrastructure costs, groundwater effects, and transparent performance requirements.

For AI customers, the issue reaches beyond one valley. Businesses buying cloud computing increasingly make environmental claims that depend on infrastructure they do not operate.

Those buyers should request location-sensitive water data from providers. Statewide or global sustainability totals cannot reveal whether computing demand falls in a stressed basin.

The Inyokern case does not prove that AI data centers will consume the valley’s remaining water. It proves that broad efficiency claims cannot resolve a site-specific resource conflict.

The next few months should bring more review documents, engineering responses, and legislative movement. Those records will show whether the 50-acre-foot estimate survives technical scrutiny.

Until then, the responsible conclusion is conditional. Inyokern’s proposed hybrid system might keep direct consumption relatively low, but the public evidence has not settled that claim.

Follow the project docket, cooling specifications, and California’s reporting legislation rather than any single viral number. Then ask a practical question: would the project remain viable under binding water limits and public monthly reporting?

That test separates a credible water plan from an optimistic forecast. It also gives Google News readers a clearer way to judge the next AI infrastructure proposal.