Disclose Cooling and Thermal Headroom for GPU Offtake Buyers

A site owner tells a GPU offtake buyer that the campus has 20 MW of cooling. The interconnect is real. The unused capacity is real.

Then the buyer asks the next questions. Is that 20 MW IT load or facility plant? At what ambient design condition? Is the hall air-cooled, liquid-ready, or neither? Where does the heat go after it leaves the white space?

At that point the megawatt number stops being a product. GPU offtake buyers underwrite a thermal envelope: how much compute heat the site can reject, under which conditions, with which plant, in which phase. Unused interconnect without that envelope is still a development story.

Publish an explicit IT-load thermal envelope

Do not lead with “cooling available.” Lead with the IT load the site can thermally support.

State design IT load in megawatts or kilowatts of compute heat, separately from mechanical plant capacity, electrical infrastructure, and building square footage. Then state the ambient design condition behind that number: dry-bulb, wet-bulb, and any altitude or water-temperature assumption. A plant that rejects 20 MW on a mild day is not the same product as a plant that rejects 20 MW at the site’s summer design wet-bulb.

Name the redundancy the envelope assumes. N, N+1, and 2N are different remaining capacities when a tower, dry cooler, or heat exchanger is down. If the offtake block can only be sold at N, say so rather than quoting the redundant plant as if every ton were tenant capacity.

Buyers comparing sites against compute demand will read this the same way they read power: contracted right, currently usable quantity, and remaining headroom. For context on how capacity buyers think about deployable infrastructure, see Pacific's reserved GPU capacity offering.

Separate air-cooling readiness from liquid-cooling readiness

Dense GPU offtake often needs liquid at the rack even when the building was designed for air. Those are two different readiness states. Publishing one “cooling ready” flag hides the gap.

Air-cooling readiness means the hall can reject the stated IT load with the existing CRAH, CRAC, or air-side plant, including raised-floor or aisle containment as it actually exists. Liquid-cooling readiness means CDUs, primary and secondary loops, heat exchangers, leak detection, water treatment, floor loading, and the space to place that equipment are either installed, designed, or still conceptual.

Disclose each state separately. Installed liquid plant, designed but unbuilt liquid plant, air plant with spare floor for CDUs, and air-only halls are not equivalent. A buyer placing high-density GPU systems will price those four conditions as four different schedules.

You do not need to complete a full rack design on page one. You do need to stop implying that an air-cooled office or light-industrial hall is liquid-ready because a GPU buyer asked about offtake.

Trace the complete heat-rejection path

Heat does not disappear at the rear door. Show the path from IT load to atmosphere or permitted discharge.

A complete disclosure runs chip or rack → CDU or air handler → heat exchanger or condenser water → cooling towers, dry coolers, or other rejection plant → the air, water, or sewer destination that actually receives the heat. Include any utility, environmental, or neighbor constraint on that last step: water rights, blowdown limits, noise, plume, or seasonal operating restrictions.

Nearby water is not liquid readiness. A river, lake, or municipal connection is an input to a rejection design. Until intake, treatment, return temperature, discharge permits, and plant are specified, the site has a water adjacency, not a liquid-cooled offtake product.

If Pacific is evaluating GPU capacity deployment against a site, this path is what makes a cooling claim diligencable.

Show headroom by phase and operating condition

Thermal headroom is leftover rejectable IT load after what is already committed, under a named operating condition, in a named phase. It is not the nameplate of the largest tower on the drawing.

Break the envelope into phases if the plant arrives in stages. Phase one may be air-only at a lower density. Phase two may add liquid loops and additional rejection. Do not present the ultimate campus number as day-one offtake.

Then show at least two operating conditions: design summer and a more typical condition, plus the degraded condition if a redundant unit is offline. Buyers would rather model a smaller summer number than discover that the attractive megawatt figure only exists in winter.

If electrical interconnect is available before the thermal plant, say so. Power without heat rejection is not GPU offtake. If the opposite is true — spare thermal plant waiting on transformers — disclose that pairing as well, the same way you would disclose a schedule gap that might need bridge capacity on the compute side while site infrastructure catches up.

Package evidence without overstating certainty

Buyers do not require a fully commissioned plant before the first conversation. They require an honest evidence grade.

Separate installed and operating plant, equipment on order, engineered but unissued packages, vendor quotes, and conceptual studies. A cooling-load calculation, a PE-stamped design, and a commissioned trend of actual rejectable load are different artifacts. Presenting them as the same “20 MW cooling” claim is what destroys trust.

Attach the few documents that support the envelope: design basis, ambient condition, one-line of the thermal loop, plant list with status, water or discharge constraints, and the phase table. Leave detailed CFD and rack layouts for diligence. The screening package should make the remaining uncertainty visible enough to schedule.

Pacific Intelligent Technologies, Inc. works across compute capacity and the physical infrastructure required to deploy it. Site owners who can describe a thermal envelope, not just an interconnect, can review Pacific's infrastructure and capacity platform or schedule a 30-minute discussion with Harper about GPU offtake.

FAQ

What do GPU offtake buyers ask first about cooling?

How much IT load the site can reject, at what ambient design condition, with what redundancy, and whether that load is air-cooled or liquid-ready today. The facility-plant megawatt number is a follow-up, not a substitute. Buyers screening sites against available and planned GPU capacity need the thermal envelope on the same page as power.

Can I claim cooling capacity before engineering is complete?

You can describe a direction of travel if you grade the evidence: study, design, procured, installed, or commissioned. Do not present a conceptual load as operating headroom. Buyers can underwrite a future phase. They cannot underwrite a rounded number with no design basis.

Does nearby water mean the site is liquid-ready?

No. Water adjacency is an input. Liquid readiness requires specified loops, CDUs or equivalent, heat-rejection plant, leak and water-quality provisions, and the permits or rights to use and discharge that water. Until those exist or are clearly designed, keep liquid readiness off the ready list.

How should a site owner state thermal headroom?

State remaining rejectable IT load by phase, at a named ambient condition, after committed load, with the redundancy assumption visible. Example language: “Phase 1 can support a stated IT MW at summer design wet-bulb with N+1 towers; liquid loops are designed but not installed.” That sentence is more useful than “20 MW cooling.” If you want help packaging that envelope for a buyer conversation, book 30 minutes with Harper.

A credible GPU offtake pitch therefore connects unused interconnect to an explicit IT-load thermal envelope, separate air and liquid readiness, a traced heat-rejection path, headroom by phase and operating condition, and evidence graded by certainty. That is what turns a cooling claim into infrastructure a buyer can underwrite.