A newcomer’s guide · Denver metro

Denver’s data center buildout: power, prices, and risk

Data centers are arriving in several sizes and places around Denver. The useful question is not whether they are inherently good or bad. It is how each project connects to the electric system, which costs it creates, who carries the risks, and what the community receives in return.

August 10, 2026About 12 minutesStart with the map
First orientation

“A Denver data center” can mean very different things

A data center is a building-sized computer system: servers do the work; electrical equipment converts and distributes power; cooling removes heat; batteries and generators bridge outages; networks connect the facility to users and other facilities.

The Denver-area projects below are not a complete inventory. They are enough to show why category-level claims fail: scale, utility, cooling design, construction stage, and neighborhood context differ.

ExamplePublished scaleUtility/contextWhy it matters
Parker22.5 MW critical IT; 249,000 sq. ft.; 17 acres; PUE 1.3CORE Electric Cooperative; Town says adjacent existing substation and no equipment-cooling waterA midsize regional facility whose public debate centers on electricity and cost allocation.
North Denver DE3About 18 MW; 170,000 sq. ft.Expansion of an established interconnected urban campus; operations scheduled for June 2026Shows that not every expansion is a hyperscale campus.
Aurora QTSPlanned 160+ MW critical campus capacity; 65 acresXcel territory; phased hyperscale developmentLarge enough to affect utility resource planning; backup-generation permits create a separate local air and noise question.
Do not add these MW figures casually. “Critical IT capacity,” campus capacity, utility service capacity, and actual hourly demand are different measures. Projects also ramp over years rather than appearing at full load on opening day.
Reading a specification

Four concepts prevent most bad analysis

MW is power

How much at one moment?

A 22.5 MW IT rating describes designed computing capacity, not annual energy and not necessarily the utility meter peak.
MWh is energy

How much over time?

Annual energy depends on actual load through every hour. Capacity multiplied by 8,760 hours is a full-load scenario.
PUE is overhead

What runs beyond servers?

PUE divides total facility power by IT power. At PUE 1.3, every 1 MW of IT load implies 0.3 MW for cooling and electrical overhead at the stated condition.
Load factor is behavior

How steady is demand?

A steady customer can use infrastructure efficiently; a rapidly changing or uncertain forecast can force expensive capacity to sit underused.

For Parker, one transparent scenario is:

22.5 MW IT × 1.3 PUE = 29.25 MW facility load
29.25 MW × 8,760 hours = 256.23 GWh/year at continuous full load

This is useful for scale, but it is not a forecast. Public information does not show Parker’s eventual hourly utilization, ramp schedule, or meter data.

How service gets built

A data center does not simply plug into the grid

1 · RequestDeveloper submits size, timing, reliability, and ramp assumptions.
2 · StudyUtility tests generation, transmission, substation, and feeder needs.
3 · ContractTariff and service agreements assign costs, minimum payments, and exit risk.
4 · BuildCustomer and utility construct the connection and any broader upgrades.
5 · OperateActual load, flexibility, outages, and demand charges determine economics.

The important distinction is between energy cost and system cost. The monthly electricity consumed is only one layer. A large connection can require generation capacity, transmission, a substation, distribution equipment, reserves, and reliability investments. Some assets serve only that customer; others strengthen a shared system.

The rate question is not “does the data center pay an electric bill?” Of course it does. The question is whether its rates and contracts cover the incremental infrastructure, capacity, and forecast risk it causes.
How customers can win—or lose

A large load is not automatically a subsidy or a bargain

Rates can improve

More revenue, better utilization

A stable, high-load-factor customer can spread shared fixed costs across more sales. A residential-heavy cooperative may value a commercial customer that pays reliably and uses assets outside household peaks.
Rates can stay neutral

Incremental cost is ring-fenced

Project-specific charges, long commitments, minimum bills, and exit protection can make the customer fund what it requires.
Rates can worsen

Forecast or allocation fails

Existing customers can be exposed if infrastructure enters the shared rate base, demand arrives late, a customer leaves, or system-wide generation is built on an overoptimistic forecast.

These outcomes depend on contract design and regulation, not on the word “data center.” CORE is a member-owned cooperative; Xcel is an investor-owned utility regulated by the Colorado Public Utilities Commission. Their governance, rate processes, resource portfolios, and disclosure obligations differ.

The Town of Parker says CORE represented that Parker’s project will pay for its own infrastructure and energy and will not increase rates for other members. That is relevant evidence. It is not the same as independently reviewing the service agreement or cost-of-service analysis.

Xcel’s 2026 proposed large-load tariff illustrates the protective tools available: project-specific generation, transmission, substation, and interconnection costs; minimum monthly payments; long-term commitments; financial security; and early-exit charges. It does not govern Parker, but it makes the mechanics visible.

Strongest case from every seat

What reasonable participants are optimizing

Developer
Best case: Compute is valuable infrastructure. Concentrated facilities are more efficient and secure than scattered server rooms. Customers need firm timelines and power certainty.
Hard question: Will the developer guarantee enough revenue to cover assets built around its forecast?
Utility
Best case: A large steady load can improve utilization, diversify sales, and finance grid expansion.
Hard question: Is the forecast credible, and can reliability be maintained without shifting risk?
Existing customer
Best case: Properly priced new load can broaden the customer base and reduce pressure on shared fixed costs.
Hard question: Which costs and risks remain socialized after the headline assurances?
Local government
Best case: Data centers can create construction work, property-tax revenue, and demand relatively few ongoing public services.
Hard question: What incentives are granted, how many durable jobs result, and what alternative use of land and infrastructure is displaced?
Grid and climate planner
Best case: New load can underwrite new clean generation and transmission; some computing may become flexible.
Hard question: What serves demand hour by hour before new clean resources arrive?
Nearby resident
Best case: A well-sited facility can be quieter and less service-intensive than many industrial alternatives.
Hard question: What are the measured noise, generator emissions, traffic, setbacks, and enforcement terms at this exact site?
Environmental questions

Electricity dominates—but design and location still matter

Carbon: MW alone does not reveal emissions. The answer depends on the marginal resources serving the load, construction timing, transmission constraints, storage, and whether clean-energy claims match hourly consumption or only annual purchases.

Water: Cooling designs differ. The Town says Parker will not consume water for equipment cooling and will use closed-loop systems; ordinary building water and upstream water used in electricity production remain. Other facilities may choose evaporative systems that exchange water use for energy efficiency.

Backup generation: Diesel generators generally operate infrequently, but their local impact is concentrated during testing or outages. Aurora’s QTS campus has sought permits for a large generator fleet, making equipment controls, test schedules, cumulative emissions, and noise legitimate project-specific questions.

Land and public value: A facility can generate substantial taxable value with limited demand for schools and services, but it also creates fewer permanent jobs than many similarly sized industrial developments. Neither fact alone decides whether the land use is good.

What supplies the power?

SMRs are one branch of a much larger decision tree

A large-load plan can combine existing capacity, new transmission, wind and solar, batteries, gas generation, geothermal, nuclear, demand flexibility, or delayed interconnection. The mix must be compared on cost, firmness, emissions, water, land, construction time, and who bears failure risk.

Douglas County’s exploration of small modular reactors and microreactors concerns broader resilience and economic-development questions. Parker’s 22.5 MW rating does not establish the need for an SMR. Equally, uncertainty about first projects does not make nuclear irrelevant to long-term firm clean-power planning.

A sound study starts with load scenarios and system needs, then compares portfolios. Naming the technology first risks turning the analysis into advocacy for or against a machine.
How to form an opinion

Ask for the evidence that changes the answer

  1. What exactly is being measured? IT capacity, facility demand, contracted utility capacity, projected peak, or actual annual energy?
  2. What is the ramp? When does each block of demand become firm, and what happens if it is delayed?
  3. What must be built? Identify generation, transmission, substations, feeders, and backup systems separately.
  4. Who pays each layer? Distinguish direct connection assets from shared network improvements.
  5. Who carries forecast risk? Look for minimum bills, long terms, collateral, and termination charges.
  6. What serves the marginal MWh? Annual renewable certificates do not answer the hourly grid question.
  7. Can the load flex? Curtailment or staged computing may reduce peak capacity needs—but only if contractually real.
  8. What is local? Cooling water, generator exhaust, noise, traffic, and setbacks must be evaluated facility by facility.
  9. What does the community receive? Use net tax revenue after incentives, durable employment, and infrastructure value—not announcement totals.
  10. What alternative was displaced? Compare the project with realistic alternative land uses and power investments, not an imaginary zero-impact baseline.
A responsible “yes” can mean approval with transparent cost allocation, enforceable exit protection, credible resource supply, and site-specific limits. A responsible “not yet” or “no” can follow when those protections are absent, the forecast is speculative, or local impacts cannot be mitigated. The evidence should choose the position.
Sources and further reading

Where the numbers and frameworks come from

FacilityFlexential Parker specifications; Town of Parker FAQ; CoreSite DE3 construction page; QTS Aurora-Denver campus.

UtilityXcel’s description of its proposed large-load tariff; Colorado PUC newsletter. Xcel’s proposal is context, not Parker’s governing tariff.

Local reportingColorado Public Radio on QTS backup generators; Colorado Sun on Aurora’s utility scale and local economic case.

MethodDavid MacKay, Sustainable Energy—without the hot air, for quantitative balance sheets; Andy Masley, “What a data center is”, and “Data centers & electricity—part 1”, for counterfactual and denominator questions. Their framing informs questions here; it does not substitute for Colorado evidence.