Air Cooling at the Limit

Almost every AI operator is migrating off air, and almost none has left it entirely. A 2026 AI campus that runs GB300 NVL72 racks (or the GB200 fleet before them) on direct-to-chip liquid still rejects 10–15% of every rack's heat to air, still cools its entire networking spine and storage tier with moving air, and, if it is honest about its fleet, still runs the majority of its floor area, if not its megawatts, on air-cooled CPU and modest-density inference. Air still handles the load liquid does not take, the brownfield you cannot re-pour, and the regime where the cheapest correct answer is a fan and a coil.

The forks here are mechanical, and the consequences show up in PUE points, fan watts, and stranded floor. Four decisions structure the chapter: raised floor versus slab (the distribution substrate), containment and the air-handling plant (CRAC vs CRAH vs in-row vs fan-wall), how warm you run the supply air (the ASHRAE envelope as an economic dial), and how far you can actually push a rack before air loses on cost. We close on the load that stays on air for good. The density wall itself — the heat-flux physics and the cooling hierarchy — is established in Chapter 5.1; this chapter is what you do on the air side of it.

Cold air goes under the IT or over it — a distribution choice that predates the AI era and constrains everything after it. The raised access floor (perforated tiles over a pressurized plenum, the iconic 600 mm void) was the industry default for thirty years because it hid cabling and let a CRAC unit pressurize the whole room. It is also a recirculation trap and a structural ceiling. A loaded 48U liquid-ready rack runs 1,200–2,300 kg — and a wet NVL72 approaches ~1,360 kg / 3,000 lb before you add the CDU and coolant — heavy enough that the OEM foot/wheel and rolling reactions must be checked against the complete access-floor assembly and move route. That verification, together with airflow and distribution needs, is why many dense AI halls use a slab.

Slab-on-grade with overhead air distribution — supply ducted down from above or pushed by perimeter/in-row units, return through a ceiling plenum — removes the structural limit and the under-floor obstruction that wrecks plenum pressure once you fill it with liquid pipe and busway, and it is the only sane substrate for racks heavy enough to need liquid. What you give up is the plenum as a free distribution path: every kilogram of air now has to be moved by a unit you can see, which turns airflow management into an active engineering problem rather than a passive room-pressure one. For a greenfield AI hall this is the right trade — you were going to plumb the floor for liquid distribution anyway, and a raised floor full of coolant manifolds is a leak you cannot see. For a brownfield air hall, the existing raised floor is exactly the constraint that caps your retrofit, a thread picked up in Chapter 5.10.

The single highest-return air-side intervention is the cheapest: stop hot and cold air from mixing. Uncontained, a room's supply and return blend, you over-cool to compensate, and your fans and chillers work against your own bypass. Containment — sealing either the cold aisle (a roof and end-doors over the supply aisle) or the hot aisle (capturing the exhaust and ducting it back) — is the difference between a PUE in the 1.5s and one approaching the low 1.2s on the same plant. The two choices differ operationally: cold-aisle containment keeps the working aisle comfortable and the room hot, is cheaper to retrofit, but bakes the rest of the floor (and the people in it) in return air; hot-aisle containment keeps the room at supply temperature and pipes the hot exhaust away, which is better for warm-supply operation and mixed liquid/air halls but costs more and complicates the ceiling. For a hall that will host rear-door heat exchangers or a liquid retrofit, hot-aisle (or chimney) containment is the forward-compatible choice, because it keeps the room cold while the racks run hot — see Chapter 5.3.

The next choice is the plant that moves and cools the air, and it turns on how close the cooling sits to the load. The terminology trips people up, so be precise: a CRAC (computer-room air conditioner) has a compressor and refrigerant in the unit — it is a packaged DX air conditioner; a CRAH (computer-room air handler) has only a chilled-water coil and fans, with the refrigeration done remotely by a chiller plant. CRAH is the hyperscale default because chilled water decouples heat rejection from the room and rides the economizer; CRAC survives where there is no chilled-water plant (smaller rooms, edge, some retrofits). As density climbs, perimeter units of either kind lose: the air has too far to travel, recirculation grows, and you spend fan power fighting your own room. In-row coolers sit between the racks and shorten the air path; fan-wall (close-coupled) units replace the wall of perimeter CRAHs with a high-efficiency fan array right at the aisle. The closer the cooling, the higher the density it supports and the lower the fan energy per kilowatt — at the cost of putting water (CRAH/in-row) nearer the IT.

Containment and plant selection set the ceiling; airflow management decides how much of that ceiling you actually reach. The failure modes are mundane and they compound: bypass (cold air that returns to the cooler without doing work — leaking tiles, gaps under racks, mis-placed perforated tiles in a hot aisle), recirculation (hot exhaust looping back into the inlet over the top or around the ends of a rack), and open U-space (a missing blanking panel that lets the two streams short-circuit straight through the rack). Each one forces you to drop supply temperature to protect the worst inlet, which spends chiller energy and erases the free-cooling hours you were trying to bank. Blanking panels, brush grommets, floor-leakage sealing, and correct tile placement are the cheapest PUE money in the building, and the first thing an audit finds missing.

The metric that captures this is the approach temperature — the gap between what the cooler supplies and what the chip's inlet actually sees. A well-managed contained aisle holds that gap to a few degrees; a leaky uncontained room can lose 5–10 °C to mixing, which is 5–10 °C you have to claw back at the chiller. The qualified air envelope is set as much by delivery, pressure, containment and recirculation as by the coil schedule itself. The thermal-metric vocabulary (approach, NTU/effectiveness) is established in Chapter 5.1; live-floor instrumentation of inlet temperatures, ΔT, and bypass belongs to operations in Chapter 14.2.

Supply-air temperature is the dial most operators under-exploit. ASHRAE TC 9.9's thermal guidelines define a recommended band (18–27 °C inlet, all classes) and progressively wider allowable envelopes: A1 15–32 °C, A2 10–35 °C, A3 5–40 °C, A4 5–45 °C. Running warmer supply air is a lever, not a concession: every degree you raise the supply (and therefore the chilled-water and condenser temperatures) buys more hours of free / economizer cooling and lowers compressor energy. A hall designed and operated to A2–A3 in a temperate climate can run hundreds to thousands more economizer hours per year than one pinned to the conservative recommended band, dragging mechanical PUE down materially.

Warmth costs you in three currencies. First, server fan power: above roughly 25–27 °C inlet, on-board fans ramp non-linearly, and past a point the rack's own fans eat the chiller savings — the warm-air optimum is a curve with a minimum, not a monotonic win. Second, reliability margin: warmer silicon means a higher failure rate and less thermal headroom for a cooling excursion, which matters more for synchronous training (a thermal trip that throttles or drops a node restarts the job) than for stateless inference. Third, acoustics and the human floor: ramped fans are loud, and a hot-aisle in an A3 hall is a workplace-safety question. The right answer is climate- and workload-specific: a cool-climate inference hall should run as warm as the fan-power curve allows; a tropical training hall has less free-cooling headroom to harvest and tighter reliability stakes. The downstream effect of supply temperature on plant selection and economizer mode lives in Chapter 5.8; how PUE and ITUE are defined and reported is canonical in Chapter 15.1.

Air-cooling feasibility has no universal rack-kW ceiling; the scoping decision hinges on the complete airflow, inlet-temperature, pressure, containment, acoustics, fan-power, heat-flux, and future-density envelope. The cited source does not publish that figure; it reports 30–40 kW typical RDHx capacity and more than 50 kW with active rear-door fans, which are door-specific reference conditions rather than a product-independent boundary. A perimeter-cooled raised-floor room closes only when its scheduled airflow, pressure, inlet map, containment, recirculation, acoustics, and design-day rejection satisfy the named racks. With aggressive close-coupling — fan-wall or active in-row units right at the aisle, full containment, and disciplined airflow management — qualified close-coupled air systems may support materially higher duties, but only at their declared airflow, pressure, inlet, containment, acoustic, and design-day conditions. The transition is often economic and operational rather than a single physical cutoff: as airflow demand rises, the fan energy per kilowatt removed climbs steeply enough that the marginal kilowatt of cooling costs more in fan power and floor space than the same kilowatt would cost on a rear-door heat exchanger or a direct-to-chip cold plate. No economic conclusion follows from rack kW alone; compare modeled fan and plant energy, floor area, water/refrigerant interfaces, service, redundancy, and expansion cost for the qualified candidates.

The cooling-selection region is an overlap. For every candidate, record the rack heat split and flux, OEM airflow/inlet limits, TCS/FWS or entering-water conditions, residual-room rejection, design climate, service/redundancy case, and named refresh profile. The cited GB200 NVL72 is resolved by its OEM record—roughly 115 kW to liquid and 17 kW to residual air—not by subtracting a generic air limit from ~132 kW. A design is stranded when its qualified envelope cannot support the next named profile, regardless of which rack-kW band that profile occupies.

Workload labels are useful for locating likely equipment, but they do not select cooling. Treat the following as starting hypotheses and close each named rack against its heat split and flux, airflow/inlet limits, water interfaces, room rejection, climate, service/redundancy case, and refresh tail.

• Networking and storage. Many current profiles remain air-cooled, but record the actual switch/server airflow direction, inlet limits, fan power, acoustics, and room duty; do not infer suitability from the equipment category.
• Inference fleets. Air, RDHx/AALC, hybrid, and DLC can all occur at the same rack kW. Model, precision, batching, server packaging, OEM heat split, and the site's thermal envelope—not the word inference—determine the candidate.
• Edge and micro-sites. Air or sealed/modular systems are common, but ambient range, filtration, acoustics, rejection, remote service, redundancy, and equipment inlet limits must close for the named site.
• Batch or curtailable work in an inherited hall. Retry tolerance changes the outage consequence, not heat removal. Use the existing air service only where the named rack and roadmap profile fit its qualified envelope.

Separate zones when their airflow, inlet, heat split, water, rejection, service, or roadmap requirements differ. The split follows the equipment and facility evidence, not permanent-air and not-yet-liquid workload labels.

Three air-side mis-scopes recur, each from reasoning about the room instead of the load:

• Using rack kW as the selector. A 45 kW air profile, a 45 kW cold-plate/AALC profile, and a 45 kW rack that fits neither can all exist. Record heat split/flux, airflow/inlet and water conditions, room rejection, climate, service/redundancy, and the named refresh tail before choosing.
• One zone for incompatible envelopes. Split zones where named equipment has materially different inlet, airflow, heat split, water, rejection, service, or fault-response requirements; do not split them merely by workload label.
• Chasing the widest ASHRAE envelope as a setpoint. Running A3/A4 to bank free-cooling hours, then watching server fans and acoustics eat the savings and reliability margin past the fan-power knee. A4 is a survivability envelope for a cooling excursion, not a steady-state target.