Will O’Leary’s data center actually consume billions of gallons of water per year?
Kevin O’Leary’s “closed-loop, minimal water” framing for the 9 GW Stratos / Wonder Valley data center in Hansel Valley is technically defensible only under a specific engineering design — reciprocating gas engines plus direct-to-chip liquid cooling with dry coolers — that the project has not actually disclosed, contracted, or confirmed. Under that design, realistic consumptive water use lands around 3,000–10,000 acre-feet per year (1–3 billion gallons/year, 3–9 million gallons/day), which is roughly consistent with the developer’s publicly assembled portfolio of ~13,000 AF of water rights. Under any conventional design (combined-cycle gas turbines with wet cooling towers, or evaporative data-center cooling), realistic water use balloons to 38,000–115,000 acre-feet/year (12–38 billion gallons/year) — physically unobtainable in this aquifer and 5–10× the developer’s portfolio. The single filed water application (Bar H Ranch, 1,900 AF from Salt Wells Spring) was withdrawn on May 7, 2026, after roughly 3,800 protests, and the developer himself, CEO Paul Palandjian, told KUER he could not provide a water-use number because the project is in “pre-development.” The gap between marketing (”zero water turbine,” “net positive to the Great Salt Lake,” “less water than ranching”) and unverifiable engineering reality is wide enough that the claims should be treated as directionally plausible but not yet credible — and several specific framings are demonstrably misleading.
What O’Leary and the project actually say about water
O’Leary’s most-cited public statements form a layered defense. On X on May 10, 2026, he called concerns about draining the Great Salt Lake “ridiculous” and went further: “if anything, we’ll be adding to the Great Salt Lake because the water rights on that land will be used and one of the hyperscalers even said to us, ‘Let’s use air cooled so there’s no water involved.’” On May 5 he claimed the project could blend in “air-cooled turbines,” solar, wind, and batteries — a statement that directly contradicted MIDA Executive Director Paul Morris’s April 22 testimony to Box Elder County that “100% of the power will be generated” from natural gas via the Ruby Pipeline. To CNN Business on May 9, O’Leary repeated the “ridiculous” framing and said developers would “invest in new technologies to reduce the facility’s water usage.”
The most consequential admissions come from CEO Paul Palandjian (O’Leary Digital). On KUER, he said the project would purchase 2,800 AF of water rights and use “just a fraction of it,” with another 10,000 AF available if needed, and that “there’s some rough order of magnitudes that if you go in your AI and ask, on a zero water solution power gen and on a zero water solution closed-loop system, on average, how much acre feet water we use, you’ll get as good a number as I could give you right now.” He could not provide a specific number. Hilary Venable (MIDA Project Director) told the Herald Journal long-term water demand would be “closer to what you would see from a large office campus, not traditional industrial use.” Morris told county commissioners the project would “use less water than ranching” and produce a “net positive to the Great Salt Lake.” Governor Cox’s official FAQ on Stratos states the design is a “closed-loop chilling system combined with dry (air-based) cooling. There is no continuous water draw.”
The numbers the developer has actually committed to paper
Only one water filing exists in the public record. Bar H Ranch, Inc. filed Change Application a54385 on March 25, 2026, seeking to convert 1,900 acre-feet/year from Salt Wells Spring Stream (water right 13-4148, 1904 priority) from agricultural irrigation to year-round industrial use for “Wonder Valley Utah“ — described in the filing itself as a 7.5 GW natural-gas power plant plus a “closed-loop” data center. Public protests reached roughly 3,800 formal filings plus ~2,000 letters of concern, a State Engineer’s office record. Consultant Logan Riley emailed the Division of Water Rights on May 5, 2026 withdrawing the application “in light of the County’s action on Monday” and signaling resubmission “with additional supporting information.” State Engineer Teresa Wilhelmsen confirmed cancellation on May 7. The withdrawal came one day after HB 60 took effect on May 6, narrowing the State Engineer’s authority to consider “public welfare” or “interference with more beneficial use” — a timing critics including the Great Basin Water Network and Center for Biological Diversity called a tactical reset to refile under a more permissive standard.
Beyond the withdrawn 1,900 AF filing, the developer’s publicly described portfolio totals roughly 13,000 acre-feet per year (~4.24 billion gallons): ~3,000 AF on-site plus ~10,000 AF under contract near Snowville. The Snowville rights have not been filed with the State Engineer and exist only as developer statements. Box Elder County Resolution 26-11 (passed 3-0 on May 4, 2026) consented to MIDA designating the 40,000-acre Stratos Project Area; the county explicitly disclaimed authority over water and air, with Commissioner Lee Perry stating “Our vote today had nothing to do with water or air quality.” The accompanying DEQ letter confirmed no Notice of Intent has been filed with the Division of Air Quality — meaning the engineering disclosures that would normally pin down water use (BACT analyses, cooling technology selection) do not yet exist publicly. HB 76 (Data Center Water Transparency Amendments) does apply: Stratos meets the ≥75 AF/year threshold and the “new large data center” definition (operations after July 1, 2026), but the law allows confidentiality claims under §63G-2-305(2) for cooling, discharge, and reuse details, and caps fines at $100/day.
What independent analysts have calculated
Three serious quantitative estimates exist, and they disagree by an order of magnitude depending on assumed technology. Utah Clean Energy (Logan Mitchell et al., May 7, 2026) calculated power-generation water alone — explicitly excluding data-center cooling and upstream gas — at 16.6 billion gallons/year for a combined-cycle gas turbine (CCCT) design (≈25,000 Olympic swimming pools) or 2 billion gallons/year for reciprocating engines (RICE); the RICE scenario uses less water but produces dramatically worse NOx emissions (12,000 tons/yr, double Salt Lake County industrial NOx). Journalist Jonathan P. Thompson (Land Desk) ran the more conventional textbook calculation — 9 GW × 60% capacity factor × 200 gal/MWh ≈ 9.5 billion gallons/year — and noted this “is in line with developers’ statements that they would eventually seek up to 13,000 acre-feet of water rights.” Zach Frankel (Utah Rivers Council) uses a general benchmark of ~5 million gallons/day per large data center, framing two large facilities as equivalent to the annual water use of Sandy, Utah (~40,000 people) or “a city of 100,000 people.” GreenLatinos translated the 13,000 AF portfolio into “more than enough water for over 20,000 households.”
USU physicist Robert Davies added a critical physics constraint: the 9 GW project produces 16 GW of thermal load that must be rejected somewhere. “My understanding is that they’re citing non-disclosure agreements and proprietary technology. That’s a red flag.” Either heat goes to atmosphere via dry coolers (driving local nighttime warming of 8–28 °F) or to water via evaporation (driving large consumption) — physics doesn’t allow both effects to be small. Ben Abbott (BYU / Grow the Flow) noted Utah needs to “decrease our water use by 40%” to restore the Great Salt Lake, calling resubmission “another breach of trust… all the hallmarks of an out-of-state mega-project with little to no concern for the local community.” Friends of Great Salt Lake’s formal protest specifically argued that “federal EIA data shows natural gas power plants at this scale consume hundreds of thousands of acre-feet per year. The application does not explain how 1,900 acre-feet is sufficient.”
What the engineering actually implies
Applying NREL TP-6A20-50900 (Macknick 2011) water-consumption coefficients to 9 GW at 80% capacity factor produces the following scenario table (consumptive water, generation + data-center cooling combined):
Scenario Generation tech DC cooling tech Total water/yr Acre-feet/yr Million gal/day Low (developer claims) Recip engines (radiator) DLC + dry coolers ~1.2 B gal 3,700 3.3 Medium (realistic mixed) Recip + some NGCC DLC + adiabatic assist 8–11 B gal 25,000–33,000 22–30 High (conventional) NGCC + wet tower Evaporative towers ~38 B gal 115,000 104
The Low scenario is technically plausible because reciprocating gas engines (Caterpillar G3520K, Wärtsilä, Jenbacher) use closed-loop radiator cooling consuming only “gallons per week” of makeup, and NVIDIA’s new Vera Rubin platform supports 45 °C warm-water direct-to-chip cooling that allows heat rejection through dry coolers in most climates. Joule Capital’s Millard County campus uses exactly this architecture — Cat G3520K reciprocating engines + DLC — with ~10,000 AF rights for ~4 GW. Novva’s West Jordan facility (1.5 M sq ft) used only 3 million gallons in 2024–25. Meta’s Eagle Mountain campus consumed just 35 million gallons in 2024 across over a terawatt-hour of operation, using air-side economization. Behind-the-meter recip-engine design with DLC dry-cooled heat rejection at Wonder Valley scale would plausibly consume on the order of 3,000–10,000 AF/yr — within the developer’s 13,000 AF portfolio, and 5–10× less than a conventional NGCC+evaporative design would need.
Three caveats matter. First, “closed loop” is marketing shorthand that conceals the dominant engineering choice — outer-loop heat rejection. A closed inner loop coupled to an outer wet cooling tower still evaporates large volumes. The developer has not published which outer-loop method it will use. Second, Hansel Valley summer ambient temperatures (95–100 °F) degrade dry-cooler efficiency and may force adiabatic/wet-assist mode for several weeks per year, adding 10–30% to nominal water use. Third, the 1,900 AF Salt Wells filing is sized only for the Low scenario — Friends of Great Salt Lake’s protest is correct that under any conventional design it would be wildly insufficient. This actually argues, perversely, that the developer intends the Low-water design, but the absence of any disclosed manufacturer, contract, vendor, hydrologic analysis, or air permit BACT filing means none of this is verified.
Where the public claims diverge from reality
Several specific O’Leary and project-team claims do not survive scrutiny. The “net positive to the Great Salt Lake” assertion (Morris, Cox FAQ, Palandjian, O’Leary) rests on the premise that converting historical irrigation use to industrial use reduces consumption — but USU’s David Tarboton notes that “if you use water for irrigation, only a very small fraction reinfiltrates,” meaning the assumption that retiring an irrigation right liberates water for the lake is contested and unsupported by published hydrologic analysis. The “zero water turbine” technology cited at the April 22 commission meeting has no named manufacturer, model, or contract; Grow the Flow’s Samantha Hawkins flagged that “there’s no publicly available hydrologic analysis or independent review to support those claims.” O’Leary’s claim that the project could be partially powered by solar/wind directly contradicts MIDA’s “100% natural gas” statement and the Bar H application’s own language (”power generation”). Palandjian’s admission that he cannot give a water number is fundamentally inconsistent with simultaneous project-team claims that consumption will be “similar to a large office complex” — you cannot know one without knowing the other.
The “adding water to the Great Salt Lake” framing has a narrow defensible reading (closed-loop blowdown reinjection or watershed offset accounting) and a misleading marketing reading (the data center produces water). Only the narrow reading is engineering-credible, and it has not been substantiated. The “closed-loop” framing is technically accurate for the inner cooling loop but is being used to imply zero atmospheric heat rejection — which is impossible at 16 GW of thermal load.
Bottom line
O’Leary’s claims sit in an awkward middle ground: not outright fabrications, but not yet credible disclosures either. The most likely engineering outcome — reciprocating gas engines plus direct-to-chip liquid cooling with dry coolers — would in fact consume far less water than critics’ high-end estimates (which assume conventional NGCC + evaporative towers) and would be roughly consistent with the developer’s 13,000 AF portfolio. In that sense, Frankel’s “city of 100,000” framing is a reasonable upper bound for water-rights control but probably overstates consumption; Utah Clean Energy’s 16.6 billion gallon CCCT scenario is almost certainly not what will be built. Conversely, “zero water,” “net positive to Great Salt Lake,” and “less than ranching” are not currently substantiated — they require engineering disclosures, vendor contracts, and hydrologic studies that do not yet exist publicly, and they rely on accounting choices (irrigation-to-industrial offset) that hydrologists actively dispute. The legitimate residual uncertainty is large: final cooling technology is not locked in, the air permit has not been filed, and HB 76’s reporting requirements will not produce verified numbers until July 2027 at the earliest. Until then, the most accurate characterization of O’Leary’s public claims is plausible if the project builds what it implies it will build, misleading in the framing it has chosen to communicate that intent, and unverifiable on the timeline at which Utah’s water-rights decisions are being made.