BlogSite-owner offtake

Data Center Cooling Options: A Practical Site Owner Guide

If you hold unused or stalled grid capacity, cooling is the question that turns a power letter into a site a GPU buyer can plan around. Here is what to understand, what to verify and how to present it.

Picture a landowner with a stalled industrial project and a signed utility feed sitting idle. A GPU buyer calls, likes the power story, and then asks three questions in a row: how will you reject the heat, where does makeup water come from, and who owns the cooling interface at the pad? The owner has good answers on megawatts and none on cooling. The call ends politely, and the buyer moves on to a site that came prepared.

This guide is for site owners, land holders and anyone with grid capacity they want to package for AI compute offtake. You do not need to become a mechanical engineer. You need to understand data center cooling well enough to describe your site honestly, separate facts from assumptions and know which questions a buyer will ask before they commit time to a site visit.

1. Why cooling decides what your site can host

Every watt delivered to a GPU becomes heat that has to leave the building or module. Modern accelerator systems pack a lot of power into a small footprint, so the cooling approach determines how densely compute can be placed, how much space it needs and what utilities beyond electricity the site must supply.

For an offtake conversation, cooling sits right next to power. A buyer will use your cooling story to judge which hardware generations your site can realistically support, how much of your electrical capacity is usable for compute versus overhead, and how much engineering they will have to fund before anything is energized. If you are still firming up the electrical side, start with the basics of feed, connection point and timing, then layer cooling on top.

  • Cooling choice shapes usable density, which shapes how much compute fits on your land.
  • Cooling equipment draws power too, so it affects how much of your feed reaches the servers.
  • Cooling may require water, which brings its own permits, sources and disposal questions.
  • Cooling interfaces define who builds what, which drives cost and schedule risk for the buyer.

2. Air cooling: what it is and where it fits

Air cooling moves heat from servers into room air, then removes that air's heat through computer room air handlers, chillers, economizers or evaporative systems. It is familiar, widely supported and simpler to maintain. Many enterprise and storage workloads still run comfortably on air.

The limit is density. As per-rack power climbs with current GPU systems, air struggles to carry heat away fast enough without very large airflow volumes, wide aisles and significant fan energy. For the highest-density accelerator platforms, air alone usually becomes the bottleneck.

  • Good fit: storage, networking gear, lower-density compute and support systems.
  • Watch for: fan and chiller energy, floor space needs and hot-weather derating.
  • Ask your engineer: what climate data says about free-cooling hours and peak ambient conditions at your site.

3. Liquid cooling and direct-to-chip

Liquid carries heat far more efficiently than air. In direct-to-chip cooling, cold plates sit on the processors and memory, and a coolant loop pulls heat straight off the hottest components. A coolant distribution unit separates that clean technology loop from a facility water loop, which then rejects heat outside through dry coolers, cooling towers or other heat rejection equipment.

Direct-to-chip is the common path for dense GPU systems today. Most designs are hybrid: liquid handles the processors while a smaller air system covers remaining components, power supplies and networking. Immersion cooling, where whole servers sit in a dielectric fluid, is another option, though buyers may treat it as a more specialized choice that narrows hardware compatibility.

  • Facility water loop: supply and return temperatures, flow capacity, pipe routing and redundancy.
  • Heat rejection: dry coolers use little or no water; evaporative towers use more water and can reject heat more efficiently in hot weather.
  • Residual air load: even liquid-cooled halls need some air handling, so do not drop it from your plan.
  • Leak detection, water quality and maintenance access all become part of the operating interface.

4. Thermal headroom in offtake conversations

Thermal headroom is the margin between the heat your cooling system can reliably reject and the heat the planned compute will produce, across the full range of weather and operating conditions. A buyer cares less about nameplate cooling capacity and more about what the system can do on the hottest day with one unit down for maintenance.

Be specific about what you know. If you have an engineering study, share its design conditions and assumptions. If you have only a concept, say so. Overstating headroom tends to surface during diligence and costs you credibility on everything else. For a deeper walkthrough of what to share and how, see our guide on how to disclose cooling and thermal headroom for GPU offtake.

  • State the design ambient conditions your cooling estimate assumes.
  • Note redundancy: what happens when a chiller, cooler or pump is offline.
  • Separate existing equipment from planned equipment.
  • Flag any seasonal or permit limits on heat rejection or noise.

5. Water makeup: what buyers will ask

Water makeup is the water a cooling system consumes and must replace, mostly through evaporation and blowdown in evaporative designs. Closed-loop dry coolers use very little makeup water, while cooling towers rely on a steady supply. Either way, buyers want to know the source, the allowed volume, the water quality and how discharge is handled.

Water is increasingly a community and permitting topic, so it can affect schedule as much as engineering does. A clear water story, including when you have none, lets a buyer pick a cooling design that matches your site from the start. Our post on disclosing water makeup for GPU offtake covers the documents that help.

  • Source: municipal connection, well, reclaimed water or none.
  • Limits: permitted or contracted volumes and any drought restrictions.
  • Quality: treatment needs that affect equipment and operating cost.
  • Discharge: sewer capacity, blowdown handling and any local rules.

6. Verify versus assume: utility, landlord, engineering

The most useful habit you can build is keeping verified facts apart from assumptions. A power letter, an access route and a drawing each answer different questions, and none of them answers a cooling question on its own. Build a simple table with three columns: what is confirmed in writing, who confirmed it, and what still needs confirmation.

  • Utility: confirm the available feed, connection point and timing, plus whether cooling loads count against the same capacity.
  • Water provider or permitting body: confirm source, volume limits and discharge permissions.
  • Landlord or land records: confirm rights to place heat rejection equipment, run pipe, handle noise and access the site for heavy deliveries.
  • Engineering: confirm pad loading, heat rejection sizing, pipe routing and climate assumptions.
  • Local authority: confirm zoning, noise ordinances and any setbacks for outdoor equipment.

Anything not confirmed belongs in the assumptions column, labeled clearly. Buyers are comfortable with open items they can see. They are uncomfortable with surprises.

7. How cooling interfaces map to module placement

Factory-produced modules change the cooling conversation because much of the cooling hardware arrives integrated. What remains is the interface: where facility water, power and network connect, and who is responsible for each side. That interface looks different depending on where the modules land.

  • Prepared pad: you or your partner provide the foundation, power connection, and possibly heat rejection space and water connections. Clarify who supplies and maintains outdoor heat rejection equipment.
  • Colocation location: the colo operator usually provides facility water and heat rejection. Confirm that their loop temperatures, capacity and redundancy match the module's needs.
  • Customer site: the host's existing systems may supply cooling, so confirm spare capacity, operating responsibilities and access during maintenance.

In every case, write down the demarcation points. A one-page interface sheet listing supply and return connections, power hand-off, network entry and responsible parties will save weeks of back-and-forth later.

8. Where Pacific's module reference fits

Pacific develops factory-produced AI infrastructure for unused or stalled power. Its module reference includes direct-to-chip liquid cooling alongside networking, storage and bare-metal compute, so cooling is planned together with the rest of the stack instead of bolted on later. Modules land on a prepared pad, at a colocation location or at a customer site, followed by site commissioning and integration.

Pacific evaluates site readiness case by case, and deployment time is a site-specific target. A viable site begins with verified power, a workable placement plan and clear operating interfaces, which is why the verify-versus-assume discipline above matters. You can read how this works on the Pacific AI factory sites page for unused or stalled power. To see the kind of capacity that ends up running on these sites, browse Pacific's catalog of dedicated bare-metal GPU nodes, which lists GPU types from H100 SXM through GB300 NVL72 systems.

  • Bring your confirmed power facts and the source of each one.
  • Bring what you know about water, heat rejection space and noise limits.
  • Bring a site plan showing access routes and candidate pad locations.
  • Be ready to mark which items still need utility, landlord or engineering confirmation.

If you hold unused or stalled grid capacity and want an informed read on how cooling and placement could work at your site, Pacific Intelligent Technologies, Inc. reviews sites case by case. Book a 30-minute site conversation and bring whatever documents you have, even if some answers are still open.

FAQ

Do I need to choose a cooling system before talking to a GPU buyer?

No. You need to describe the constraints that shape the choice: climate, water availability, space for heat rejection equipment, noise limits and how much of your power feed can serve cooling loads. A clear picture of those constraints lets a buyer or module provider propose a design that fits.

Is direct-to-chip liquid cooling possible without much water?

Often, yes. The technology loop is closed, and heat can be rejected through dry coolers that use little or no makeup water. The trade-off usually shows up in larger equipment footprints and more cooling energy on hot days, so your engineer should check local climate data before you rely on it.

What drives cooling cost at a site?

The main drivers are climate, the heat rejection method, water treatment and supply, redundancy level, pipe routing distance, pad and structural work, and who owns each side of the interface. Sites with confirmed water rights, open space near the pad and clear demarcation points tend to need less custom engineering.

Continue on the mothership

This satellite stops at the playbook. Transactions, specs, and comparisons live on pacificmachines.com. If the next step is a human, book 30 minutes with Harper.

Book 30 min