Retrofitting Air-Cooled Facilities for Liquid

Greenfield liquid is a design problem; retrofit is an archaeology problem. The building already exists, and it was sized for a different physics — 3 to 7 kW racks on a raised floor, air handlers tuned to a 1.4–1.6 PUE, an electrical bus engineered for a power density that a single GB200 NVL72 rack now exceeds by 20x. The retrofit question is never 'can liquid cool this hardware' (it can); it is 'what does this building let me do before one of its original constraints runs out.' Roughly 68% of enterprise data centers built before 2015 lack the power density and cooling capacity for modern AI workloads (Introl, 2025) — and many of those same halls have a decade-plus left on their leases, which is precisely why the retrofit market exists at all rather than everyone simply building new.

This chapter runs the decision in the order the building forces on you. First the brownfield assessment — the four fixed quantities you inventory before committing capital. Then the retrofit paths ranked by disruption, because in a live hall the cost that matters is not capex per kW but downtime per rack and risk to the racks already earning. Then phased migration through mixed air-and-liquid operation, where the hardest engineering is keeping a half-converted hall thermally and electrically stable while you commission charged piping over running equipment. We close on the stranded-capacity math — the power-versus-cooling-versus-floor reconciliation that tells you how much of the building you actually unlocked, and how much you paid to leave empty. Each fork carries a cost the building extracts for choosing wrong.

Every retrofit feasibility study reduces to four inventories, and a candidate building passes only if it clears all four — they are an AND, not an OR. Miss any one and that constraint becomes your density ceiling no matter how generous the other three are. The discipline is to find the binding constraint before design, because the binding constraint determines whether you are buying a liquid-capable AI hall or merely a more expensive air hall.

Floor loading is the constraint that fails most retrofits, and the one that is least negotiable, because it is structure, not equipment. A legacy raised floor is commonly rated around 150 lb/ft² (≈730 kg/m²); a GB200 NVL72 weighs ~1.36 t (3,000 lb) in a 600 mm × 48U footprint, which works out to a point load on the order of 1,800–2,000 kg/m² — well past what raised floors rated for 1,200–1,500 kg/m² will carry, let alone a 730 kg/m² legacy deck. The fixes — steel load-spreading plates, removing the raised floor and landing racks on the structural slab, or post-tensioned slab reinforcement at roughly $200/ft² with facility downtime — are costly and slow precisely because they are civil work in an occupied building. Wet weight (coolant-filled cold plates, manifolds, charged in-rack piping) and the seismic restraint of charged pipe push the number higher still. → the structural basis is engineered in Chapter 6.2.

Plenum and distribution volume is the second. A raised-floor plenum that was an air path is now a contested corridor: it must carry coolant supply/return mains, CDU connections, and leak-zoning drainage, while it is often already congested with decades of legacy power and data cabling. A slab-on-grade hall with no plenum forces overhead pipe racks and a different leak-containment geometry. The plenum question decides where the CDU lives (in-row, end-of-row, perimeter gallery), how the secondary loop is routed, and whether liquid distribution competes with the residual air cooling you still need for the ~15% of rack load that stays air-cooled. → spatial planning for liquid in Chapter 6.1.

Electrical headroom is the third, and it is frequently the true ceiling. A hall sized for 5 kW racks has a bus, transformers, switchgear, and PDU topology to match. Liquid does not reduce the IT load — it removes the chiller and CRAH penalty, but the racks themselves draw 20–100x more. If the upstream feed, the busway ampacity, or the available UPS capacity caps out at, say, 8 MW, then you can cool far more density than you can power, and your retrofit is power-bound the moment the cooling works. The worst version of this is the retrofit that succeeds thermally and then strands half the floor because there is no megawatt left to energize it. → the power chain in Chapter 4.5; the binding-constraint framing in Chapter 1.1.

Available water is the fourth, and it is a siting fact you inherit, not a design choice. A full liquid-to-liquid retrofit needs a facility water loop and heat rejection sized to the new load — and many legacy halls were sited where water was never a planning input. The path around it (closed-loop L2A, which rejects to the existing air system) trades water independence for a hard density ceiling, because you cannot reject more heat to the room than the room's air handlers can carry away. → water sourcing and climate-driven rejection in Chapter 3.7 and Chapter 5.8.

There are three real paths from an air hall to liquid, and the right way to rank them is not by peak density — it is by how much they disturb a building that is already running revenue. In a live hall, the dominant cost is downtime per rack and risk to the neighbors, not dollars per kW. The paths form a disruption ladder: closed-loop liquid-to-air sidecars at the bottom (drop-in, no facility plumbing), rear-door and air-assisted liquid in the middle (room-level water, no chip plumbing), and full direct-to-chip liquid-to-liquid at the top (facility water loop, CDUs, in-rack manifolds, the works).

L2A — closed-loop liquid-to-air (the drop-in). The cold plates and in-rack loop are real direct-to-chip liquid cooling, but the heat is rejected back into the room air through a liquid-to-air heat exchanger inside the rack or an adjacent sidecar, with no facility water piping at all. Google's open-sourced Brazos sidecar (2025–26) is the canonical example: a sidecar rack that cools an adjacent ~60 kW rack inside an existing air-cooled hall, rejecting into the hot aisle, letting an operator add dense racks incrementally without re-plumbing the building. The trade is a hard ceiling — you can only reject as much heat to the room as the room's air handlers can ultimately carry away — and it pushes the air-side harder, but it is the only path that needs neither facility water nor a slab-up rebuild, which is why it dominates the early, low-disruption end of the retrofit market.

RDHx / AALC — rear-door and air-assisted liquid (the bridge). Rear-door heat exchangers replace the rack's back door with a water (or refrigerant) coil that intercepts the hot exhaust, carrying roughly 15–30 kW per rack passive and up to ~50–100 kW with active fans. This needs a room-level water loop and a CDU, but it leaves the servers themselves untouched — no cold plates, no quick-disconnects at the chip — so it is the natural bridge for a hall ramping density gradually, or for a mixed fleet where some racks are dense and some are not. It is the brownfield-friendly middle: more capacity than L2A, far less surgery than L2L. → engineered in full in Chapter 5.3.

L2L — full direct-to-chip, liquid-to-liquid (the destination). Cold plates on the hot silicon, in-rack manifolds, ~150–200 quick-disconnects per rack, a CDU isolating the technology-cooling loop from the facility water loop, and heat rejection sized to the new load. This is the only path that reaches GB200/GB300-class density (120–142 kW/rack) and beyond, and it is the only one whose economics make sense for training-shaped halls. It is also the most disruptive: it requires the facility water loop and heat rejection that the brownfield assessment may have told you the building never had, plus the largest civil and electrical works. → engineered in Chapter 5.4; CDUs and the secondary loop in Chapter 5.6.

The table is a disruption ladder with a density tax attached at the bottom. L2A buys you incremental, drop-in liquid with no negotiation with the building's water and slab — but it caps near 60 kW and pushes the air system, so a hall that goes all-L2A is still ultimately air-reject-bound. RDHx buys more headroom for a door swap and a room loop, but it tops out before training density. Only L2L reaches the racks that pay for training-shaped facilities, and it demands exactly the facility water and structural capacity the brownfield assessment told you the building may not have. The honest reading: the path is mostly chosen for you by the binding constraint, not by your ambition. If the slab and the water are there, you can reach for L2L; if they are not, L2A and RDHx are not lesser versions of the same thing — they are the ceiling.

You almost never convert a live hall in one outage; you convert it in phases, and for a long, awkward middle period the hall runs mixed — some rows on air, some on liquid, sharing a room, a power chain, and an air-handling plant that was balanced for neither. The engineering difficulty of a retrofit lives here, in the mixed state, not at either endpoint.

The first problem is thermal balance. As liquid rows come online, they pull heat out through water instead of dumping it into the room — which sounds purely good, but it changes the air-handling load non-uniformly. The remaining air-cooled rows still need their full airflow and supply temperature; meanwhile the room's total air heat load drops, CRAH units that were sized for the old load now short-cycle or over-cool, and hot/cold-aisle containment that was tuned for a uniform air hall now has to cope with rows that contribute almost no exhaust. Containment, airflow rebalancing, and CRAH/CRAC setpoint changes have to track the conversion row by row, or you get hot spots on the surviving air rows even as the liquid rows run cold. → hybrid containment in Chapter 5.2.

The second problem is electrical sequencing. Each liquid row is a step-change in rack power on a bus that was provisioned for far less, and the conversion order has to respect the upstream feed, PDU loading, and UPS capacity so that energizing a new dense row does not starve or trip the rows already running. This is where the electrical-headroom inventory from the assessment becomes an operational schedule, not just a feasibility number.

The third — and the one that keeps facility managers awake — is commissioning charged piping over live equipment. Flushing, filling, and pressure-testing a coolant loop a few feet from running, revenue-bearing racks means a first-fill leak is not a maintenance event, it is an incident with the potential to take down neighbors. The mitigations are procedural and physical: leak-zoning so a failure is contained to its row, leak detection wired to fast-acting shutoffs, blind-mate and quick-disconnect couplings that minimize open-fitting time, hydrostatic acceptance done on isolated sections before they go near live load, and a commissioning sequence that proves each loop dry-then-wet before any GPU depends on it. The retrofit's defining risk is that you are doing wet work in an occupied building. → leak detection, containment, and the commissioning sequence are engineered in Chapter 5.11 and accepted in Chapter 13.5.

The deliverable of a retrofit is not 'liquid cooling installed' — it is usable megawatts of IT load on the floor. And the trap is that a retrofit reshuffles the three quantities that gate usable load — power capacity, cooling capacity, and floor area you can structurally and physically populate — so that they almost never line up. Whichever is smallest is your real capacity; the gap between it and the others is stranded capacity you paid to create and cannot use.

Work it as three independent budgets for the same hall, then take the minimum:

• Power-limited load = the megawatts the upstream feed, busway, switchgear, and UPS can actually deliver to IT after the new density. Liquid frees the chiller/CRAH share of the electrical budget, but the racks consume the freed headroom many times over.
• Cooling-limited load = the heat the chosen path can reject. For L2A and RDHx this is bounded by what the room air system can ultimately carry away; for L2L it is bounded by the facility water loop and heat-rejection plant you installed.
• Floor-limited load = the racks you can structurally place (slab/floor rating after reinforcement) and physically fit once CDU galleries, pipe racks, leak-zoning, and service clearance consume white space.

Usable load = min(power, cooling, floor). The other two are stranded. A hall that retrofits cooling to 12 MW but is fed 8 MW has 4 MW of stranded cooling; a hall powered to 12 MW whose reinforced slab only carries dense racks across 60% of the area has stranded power sitting over a floor it cannot load. The pre-retrofit hall was typically cooling-bound (air ran out first); a successful liquid retrofit usually moves the binding constraint to power or floor — which is exactly why the assessment has to inventory all four quantities up front, and why the stranded number, not the density headline, is the figure that decides whether the retrofit earned its capital.

The strategic point underneath the arithmetic: a retrofit's return is decided by the smallest of its three budgets, and the disruption and capital are spent on all three. The operators who win at brownfield treat the assessment as a constraint-ranking exercise, pour irreversible substrate only where a committed density ramp justifies it, choose the lowest-disruption path that clears the actual ceiling, and report success in usable megawatts and stranded megawatts — not in 'liquid-ready' press releases. A retrofit that unlocks 10 usable MW with 1 MW stranded is a triumph; one that installs 142 kW cooling over an 8 MW feed and a slab that loads 60% of the floor is a cautionary tale, however liquid-ready it looks.