They are among the fastest-growing water users in the most stressed basin in America. That also makes them the fastest way to put water back. Every operator below can reach zero net Colorado River water — here is exactly how.
Lake Powell has lost more than half its surface area since 2019 and this year's snowpack is the lowest in the satellite record (both measured on our satellite layer). Into that, the basin's cities are courting data centers that draw municipal water sourced from the Colorado. Most operators do not disclose per-site water use, but the scale is real: a proposed 330 MW site in the Imperial Valley is seeking roughly 287 million gallons a year straight from the river, a figure from its own filing, now in litigation. The point is not to stop the build. It is that the same companies, with the same balance sheets, can make themselves water-neutral or water-positive and hand the basin a model.
Switch to air-cooled, closed-loop or immersion cooling. The single biggest lever: it zeroes the evaporative draw at the source.
Microsoft agreed to air-cool its Goodyear campuses amid local water stress; Novva runs near-zero-water cooling. ref
Source 100% of cooling and campus water from municipal recycled effluent or on-site blackwater reuse, so no Colorado River molecule is consumed.
Mesa exchanges treated effluent with the Gila River Indian Community; on-site advanced purification treats campus wastewater to reuse. ref
Fund potable-reuse capacity that puts an equal, measured volume back into Lake Mead or the river — the SNWA return-flow model, verified by lake elevation.
Southern Nevada recycles ~99% of indoor water back to Lake Mead for return-flow credits. ref
Pay farmers to fallow, with the saved water measured field-by-field by satellite (OpenET) and left in the system — a clean volumetric water-benefit credit, not a paper offset.
~$150–400 per acre-foot, structured with a water district; measured on OpenET. ref
Bank treated or brackish water in the aquifer for storage credits, and finance watershed/river restoration (invasive removal) that returns water through reduced evapotranspiration.
Meta's Rio Grande restoration (8 projects near Los Lunas) restores over 172M gallons/yr. ref
Where scale allows, develop brackish-groundwater desalination for a drought-proof local supply, powered by the campus's own solar so it adds no grid or river burden.
Arizona and New Mexico sit on large untapped brackish aquifers.
Drive absorption chillers with concentrated solar-thermal collectors (e.g. Solar Steam, ~0.5 MWt cooling per acre → ~2 acres per MW). Daytime solar cooling hits exactly when the heat, the cooling load, and the evaporation are highest — displacing both water and grid power at the source.
Land-rich desert campuses (Goodyear, Los Lunas, Imperial Valley, Eagle Mountain) have the acreage; see the siting screen below. ref
Cooling load ≈ the site's IT load, and solar-thermal collectors deliver ~0.5 MWt of cooling per acre, so a site needs roughly 2 acres of collectors per MW. Raw campus size is not the answer — most of that land is already built out. So we measured, from 10 m satellite land-cover (ESA WorldCover), the open, non-built land within 3 km of each campus and set it against the requirement. Built-up land is excluded; cropland is shown separately (buildable, but active farmland).
| Operator | Location | Load | Collectors needed | Open land ≤3 km (measured) | Already built | Verdict |
|---|---|---|---|---|---|---|
| Stream Data Centers | Goodyear, AZ | 280 MW | ~560 ac | 3,014 ac open +1,570 crop | 2,108 ac | ample |
| Imperial Valley Computer Manufacturing (Proposed) | Imperial, CA (Imperial County) | 330 MW | ~660 ac | 1,583 ac open +4,312 crop | 957 ac | feasible |
| Vantage Data Centers | Goodyear, AZ | 176 MW | ~352 ac | 2,831 ac open +698 crop | 3,081 ac | ample |
| QTS Data Centers | Eagle Mountain, UT | 120 MW | ~240 ac | coords pending | — | likely (est.) |
| QTS Data Centers | Aurora, CO | 120 MW | ~240 ac | 4,537 ac open +967 crop | 1,310 ac | ample |
| CyrusOne | Chandler, AZ | 169 MW | ~338 ac | 2,355 ac open +361 crop | 3,582 ac | ample |
| Digital Realty | Chandler, AZ | 54 MW | ~108 ac | 852 ac open +51 crop | 5,313 ac | ample |
| Novva Data Centers | West Jordan, UT | 200 MW | ~400 ac | 4,884 ac open +88 crop | 1,634 ac | ample |
| Novva Data Centers | Colorado Springs, CO | 200 MW | ~400 ac | 2,672 ac open +5 crop | 627 ac | ample |
| Aligned Data Centers | West Jordan, UT | 80 MW | ~160 ac | 1,562 ac open +97 crop | 4,068 ac | ample |
Open land = shrub + grassland + bare (WorldCover 20/30/60) within a 3 km radius of the grounded campus coordinates; "already built" = WorldCover built-up class. A finer parcel/ownership check is the next step before siting. Even excluding built-out land, every measured candidate has the acreage — QTS Eagle Mountain and the Imperial Valley cropland have it in abundance.
Water use is an engineering estimate — the site's IT load times a water-intensity band for its cooling type (evaporative draws water, air-cooled barely does). Actual use varies with operator efficiency and reuse. We show a specific reported figure only where a primary source (a water permit, the operator's own environmental report, or a government/utility document) confirms an annual number — that is a small minority of sites; most per-site water use is simply not disclosed. Buildable land is measured from 10 m satellite land-cover within 3 km of a confidence-checked campus coordinate, excluding already-built land; where we could not verify the exact campus location, we make no land claim. Project costs and timelines are grounded estimates to size the opportunity, not quotes. Where a site matched a primary EPA facility record (an air or Clean-Water-Act permit in EPA's ECHO database) we link it — these confirm the facility and its generators or stormwater, not cooling-water volumes, which most data centers send to the sewer rather than a permitted discharge. For each site we also identify the local sewer authority that holds the cooling-tower blowdown (industrial-user / pretreatment) records — those flow figures are almost never posted online and need a public-records request to the named agency; we found exactly one public figure (Microsoft Goodyear, 1.2 MGD). This is independent analysis; verify each figure before acting on it.
Every one of the 31 sites is costed and time-lined on its own page, filterable by region and by how it cools today, with capex, timeline and a viability rating grounded in that location's real power and water availability. Of the 31: 14 already run water-free or closed-loop, 5 still cool evaporatively and have the most to gain, and 12 do not report their cooling at all.
The data behind this page is published. Every site, its costed projects, the water model, the permit matches and the coordinates are in data/dc, so any figure here can be checked against the record it came from. Two things are held back and neither is used as evidence above: the per-operator project menus, whose five headlines are rendered but whose bodies are business-development material rather than analysis, and the intermediate costing scratch that costing.json supersedes. Agency email addresses are stripped.
Every figure here is either linked to a primary source inline or is a labeled estimate built on a stated method. In detail:
Every operator on this page can get to zero net Colorado River water using tools that already exist and that some of their peers already use. The menu is public, the projects are costed, and the savings are measurable field-by-field. Who leads?
Independent analysis by Steps Ventures, not affiliated with any operator, utility, or agency. Grounded research; verify each site's cooling and water source before acting. See our measured basin data at the satellite layer.