Site Selection
What GlobalAI's Windsor Approval Means for Colorado Data Center Sites
What GlobalAI’s approved footprint, power disclosures, and cooling estimates reveal about evaluating Colorado data center land.
GlobalAI's approval near Windsor gives Colorado landowners and developers a useful reference point. We can now study a specific project that cleared the county's land-use process and identify the conditions that made the approval possible.
On September 9, Weld County commissioners voted 5-0 to approve permit USR26-0019 for Global AI (Colorado), LLC. The project will reuse an industrial building at the former Kodak/Carestream campus. It was the first data center reviewed under the county's new process. Future phases require additional approval. Weld County's decision announcement establishes that scope.
For someone looking for the next Colorado data center site, the useful question is straightforward: which characteristics of this approved project can we measure and compare with other properties?
Start with the approved footprint, the existing industrial use, the power documentation, the cooling plan, and the conditions attached to the permit. Together, those details give us a more specific screening standard.
Start with the portion that received approval
The reported data center footprint is 41 acres within a larger 438-acre property. Those numbers describe two different areas that belong in a site database. KUNC's coverage of the approval identifies both.
Dividing 41 by 438 puts the reviewed footprint at approximately 9.4% of the larger property. That calculation gives the project a useful physical context. A relatively small share of the assemblage supports the current land-use case, while the remaining acreage creates questions about later phases and supporting uses.
For comparable sites, record three acreage figures: total ownership, the area covered by the application, and the usable development area established through engineering. Each number answers a different question about the project.
Total ownership shows how much land is under common control. The application boundary identifies the area under review. Usable development area accounts for the space the proposed facility can actually occupy after its layout and constraints are evaluated.
A data center land comparison should identify the acreage covered by the approved application. At GlobalAI's Weld County property, the reported 41-acre footprint represents about 9.4% of the 438-acre assemblage. That ratio describes this project's review boundary and ownership context. A comparable site's buildable area still depends on its own layout, infrastructure, and land-use requirements.
That is a useful benchmark because someone can reproduce it. The inputs are public, the calculation is visible, and the conclusion stays tied to the specific property.
Industrial zoning belongs at the beginning of the search
Weld County adopted its data center code changes in April 2026. The county established a Use by Special Review path in the I-1 district and Site Plan Review in I-2 and I-3. It also prohibited data centers in the agricultural zone. The update requires electricity and water documentation and establishes noise standards. The county's ordinance announcement explains those changes.
Those distinctions should affect the first round of site screening.
A Weld County search should begin with the current zoning district and the application path available in that district. An I-1 property goes into a review group built around the special-review process. An I-2 or I-3 property goes into the applicable site-plan process. Agricultural land requires a separate assessment of whether a zoning change is feasible.
This approach gives the research team a specific assignment for each parcel. It also makes the cost of pursuing a site easier to estimate because the team can identify the next approval it needs.
GlobalAI's existing industrial setting is especially relevant here. A developer studying a similar reuse property can compare the proposed data center with the building and land use already on the site. A developer studying vacant land needs to build that comparison from a different starting point.
The two groups need separate timeline assumptions. Reusing a building raises questions about its condition, configuration, and infrastructure. Developing vacant land raises questions about the entire site layout and the approvals needed to establish it.
Put dates beside every power figure
The initial operating load has been reported at 27.5 MW. Earlier reporting also described a Poudre Valley Rural Electric Association letter addressing 35 MW of service, with capacity changing in later years. Colorado Public Radio's August review of the project records is particularly useful because it discusses the timing attached to the power figures.
The arithmetic is simple: 27.5 MW equals 78.6% of 35 MW. The numerical difference is 7.5 MW.
The next step is to put those figures on a calendar. The date covered by a utility letter, the developer's requested delivery date, and the timing of an infrastructure upgrade all belong in the same comparison. A dated service commitment is much more useful than a single capacity field.
For the Windsor benchmark, preserve the letter version used in the application and record any replacement document. That lets a reviewer see whether the power assumptions changed between the initial submission, the hearing, and later construction milestones.
For another parcel, the power record should include the serving utility, requested initial load, requested final load, delivery dates, and the document supporting each figure. It should also identify the party responsible for the next study or agreement.
This creates a practical way to compare two sites. One may have a shorter route to existing infrastructure. Another may have more developed service documentation. The comparison should show both attributes and the work required to advance each one.
Turn the cooling estimate into a water budget
The approval coverage reports a closed-loop cooling requirement, an initial system fill of approximately 65,000 gallons, and annual replenishment of approximately 6,500 gallons. Those are cooling-system estimates with a defined purpose. KUNC reports the figures and the cooling condition.
Using GlobalAI's reported cooling estimates, one 65,000-gallon fill plus ten annual replenishments of 6,500 gallons would total 130,000 gallons. Half of that modeled cooling-water volume would be the initial charge. The calculation provides a way to separate commissioning demand from recurring demand when evaluating a phased data center.
The calculation assumes one initial fill and ten replenishments at the reported rate. It covers that cooling-system scenario. A full site water budget should separately account for domestic use, maintenance, irrigation where proposed, and other demands identified by the design team.
Water volume and water timing deserve separate attention. An initial fill creates a delivery requirement during commissioning. Annual replenishment creates a recurring requirement. A water provider needs to understand both.
The same structure helps with future expansion. If another building adds a separate cooling loop, the project team should identify its commissioning fill and recurring demand. Keeping each phase visible makes the water discussion easier to follow.
A comparison table can capture the main inputs without compressing them into a broad claim about efficiency:
| Measure | Publicly reported input | Calculation or use |
|---|---|---|
| Current project footprint | 41 acres | Approximately 9.4% of the 438-acre property |
| Initial power demand | 27.5 MW | 78.6% of the reported 35 MW service figure |
| Initial cooling fill | 65,000 gallons | Commissioning requirement |
| Annual cooling replenishment | 6,500 gallons | Recurring estimate |
| Fill plus ten replenishments | Calculated | 130,000 gallons |
Calculations use the reported inputs above. Acreage describes the application and ownership areas. The power ratio compares stated figures whose service dates require review. The water total covers the specified cooling scenario.
Noise conditions should become design inputs
Weld County's data center rules include a 65 dB(C) property-line limit, alongside the applicable dBA requirements. The county's approval announcement identifies the continuing noise obligations.
For another site, I would give the acoustic consultant the actual property boundary, nearby residences, proposed equipment locations, and operating assumptions early in the layout process. Those details should be coordinated with the electrical and cooling design.
The site database can help that work. It should preserve the distance from proposed equipment areas to the nearest relevant property boundary. It should also distinguish existing residences from planned residential development and identify the source date for each layer.
A parcel centroid provides a convenient map label. Equipment locations and receiving properties provide the geometry needed for the next design question. That is where I would spend the effort once a candidate reaches detailed review.
Track the version of the equipment schedule used in the acoustic analysis. Changes to fan counts, generator arrangements, operating hours, or screening should prompt a review of the associated sound assumptions.
The goal is to keep the permit condition connected to the drawings that are supposed to satisfy it. That connection becomes valuable when the project moves between planning, engineering, construction, and operations teams.
Municipal services can shape later phases
The property's relationship with Greeley and Windsor deserves its own record. The Independence Institute's September 11 coverage reports that Greeley approved a pre-annexation agreement on September 8, creating an 18-month negotiation window. That report links the annexation development to the broader project.
Windsor's own project information page also describes a water and wastewater service letter conditioned on annexation into the town. The town's explanation makes the relationship between municipal boundaries and service discussions explicit.
For site selection, record the current land-use authority, the prospective annexing municipality, the water provider, and the wastewater provider as separate fields. Attach the agreement or letter that supports each relationship.
That structure makes later changes easier to evaluate. If an annexation agreement advances, the reviewer can identify which assumptions about services, fees, development standards, and future phases need to be revisited.
A map should carry that history forward. The public-facing location name may stay the same while the approval path and service arrangements evolve. A dated jurisdiction record keeps the development model aligned with the documents.
Use Windsor to improve the next site decision
I would add GlobalAI to a comparison library with the permit number, application boundary, zoning district, power assumptions, cooling estimates, noise requirements, and expansion conditions. Every entry should point back to a source and show when that source was checked.
Then I would compare a new candidate against that record one question at a time. Does the candidate have an identified land-use path? Can the proposed first phase fit within a defined area? Is there a dated power plan? Can the cooling design produce a phase-specific water budget? Are the noise conditions reflected in the equipment layout?
The strongest result of this work is a clear next action. A parcel might need a zoning interpretation, a revised site plan, a utility study, or a water-service meeting. That action should follow directly from the gap in its record.
GlobalAI's approval gives Colorado a concrete example of how those pieces can come together at an industrial reuse site. The value for the next project comes from studying the conditions closely and carrying them into the first serious site review.