Structural & Civil Engineering for Dense Liquid-Cooled Halls

Cooling chapters tell you a GB200 NVL72 draws ~120 kW and needs ~80 L/min of warm water. They rarely tell you it also weighs 3,000 lb empty and 6,000–8,000 lb wet-and-cabled, concentrated over roughly ten square feet, and that this single fact quietly decides whether your hall can host it. The structural and civil basis of a dense liquid-cooled hall is the substrate every other Part 6 decision sits on — and unlike a transformer, a CDU, or a fabric switch, you cannot swap a slab mid-life. Get the floor-loading basis wrong and you do not have an inefficient building; you have an air-cooled-only building committed to a liquid-cooling roadmap it can never deliver.

This chapter owns the civil and structural basis for dense halls as a series of decisions and their downstream costs: slab-on-grade versus elevated structural slabs, the long-span steel and column grid that determine whether the hall is workable, the multi-story complications that liquid introduces (live load, deflection, vibration, and the dead weight of coolant and pipe), the crane and rigging provisions that make equipment moves possible, and the seismic and anchorage strategy for racks that are heavy, plumbed, and energized at once. Floor loading for 3,000+ lb racks — point loads, distributed loads, and the limits they impose — is owned here as the civil basis the rest of the guide inherits. The geotechnical and soils interface, and seismic hazard at the site level, tie back to Chapter 3.8.

The first and least-reversible structural decision is what the racks stand on. There are two answers, and they pull the entire building in different directions.

Slab-on-grade rests the floor directly on prepared subgrade, so the load path is short and the achievable capacity is governed by the soil and the concrete, not by beams and girders. This is why nearly every purpose-built dense AI hall in 2026 is single-story slab-on-grade: it is the cheapest way to reach the 600–1,000+ psf the racks demand. A dense liquid-cooled slab is poured thick — commonly 12–18 inches versus the 6–8 inches of a legacy air-cooled hall — with post-tensioning and epoxy-coated rebar to hold deflection near zero under heavy point loads. The premium is real but bounded: on the order of $50–75 per square foot of added construction cost for the upgraded slab. The constraint it imports is geotechnical, not structural: the slab is only as good as the subgrade beneath it, which is exactly why the soils basis in Chapter 3.8 is on the critical path, not a footnote.

An elevated structural slab — a suspended concrete deck on a steel or concrete frame — is what you pour when you go multi-story to conserve land, or when you need a service level below the white space for piping and power. Now the load path runs through beams, girders, and columns, and every one of them must be sized for the wet-rack live load plus the dead weight of the coolant, pipe, and equipment it carries. Capacity is no longer cheap: doubling the design live load on an elevated deck cascades into bigger members, more steel, more foundation, and tighter deflection control. The fork is therefore not stylistic. Slab-on-grade buys the highest density per dollar but spends land; the elevated slab buys land efficiency but pays for every pound in structure — and makes the multi-story liquid problem below unavoidable.

Structural engineers design floors to a uniform live load in psf or kN/m², but racks deliver their weight as concentrated point loads under each foot or caster path — and the two are reconciled, not equal. Walk the arithmetic, because the whole civil basis turns on it. A loaded ~1,500 kg compute rack on a 0.6 m × 1.2 m footprint imposes about 20.8 kN/m² (~435 psf) averaged over its own footprint. But racks do not tile the floor edge-to-edge; once you account for aisles, service clearance, and CDU/manifold lanes, the same rack pattern resolves to an equivalent uniform design live load near 10 kN/m² (~210 psf) over the hall — which is why a hall can be specced at a deceptively modest psf and still carry severe local point loads. The binding check is the worst-case concentrated load on a single caster path during rigging, not the tidy hall average.

This gap between footprint load and hall-average load is the source of most floor-loading mistakes. Design only to the average and you under-provision the slab against the point loads of a rack being rolled into place; design the whole hall to the worst-case footprint load and you over-build steel and foundation by 2x for area that is mostly aisle. The discipline is to specify both: an equivalent uniform live load for the hall structure, and a concentrated point-load rating for the rack positions and the rigging paths between them. NVIDIA, notably, does not publish an official floor-loading spec for the NVL72 — the ~440 psf figure that circulates is a back-calculation from rack mass over footprint — so the load basis is the structural engineer's to set per site, and the wet-and-cabled weight (not the datasheet empty weight) is the number that governs.

A hall is not just a floor — it is a volume that equipment has to move through for two decades. The column grid and roof span decide whether that volume is workable. Dense AI halls trend toward long-span steel with wide column bays for one blunt operational reason: NVL72-class racks ship pre-integrated at 3,000 lb and must be rigged into position, CDUs and manifolds need clear runs, and overhead cable tray and busway want unobstructed paths. Every column in the white space is an obstacle a forklift, a rigging skate, or a future row reconfiguration has to route around. A tight column grid saves steel up front and costs you flexibility and move-time for the life of the building — a classic capex-now-versus-opex-forever trade that almost always resolves in favor of the wider bay for a building expected to absorb three GPU generations.

Roof and overhead structure carry their own load case that is easy to miss: in liquid halls a great deal of the mechanical distribution — pipe racks, manifolds, and sometimes overhead cable tray bearing 500–800 lb of copper per rack — hangs from the structure above the white space. That is suspended dead load the long-span steel must carry in addition to its own weight and any crane provision, and it interacts with deflection limits because sagging structure pulls on rigid pipe joints. The decision to run liquid and power overhead (common where racks sit on slab) versus in a service level below (common in multi-story) is therefore simultaneously a structural-loading decision and a maintenance-access decision — see the hall-layout treatment in Chapter 6.1.

Going vertical to conserve land is increasingly common in power- and land-constrained metros, and liquid cooling makes it materially harder than a stacked air-cooled building ever was. Four structural problems compound floor on floor:

• Live load on every deck. The wet-rack live load that slab-on-grade passes straight to soil must now be carried by beams, girders, and columns on each suspended floor — and the columns accumulate the load of every floor above. The lowest columns and the foundation see the full stack, so a four-story dense hall is not four times a single floor; it is a load pyramid that drives oversized lower members and deep foundations.
• The dead weight of coolant and pipe. Air weighs nothing structurally; water does not. An NVL72 rack carries ~200 L of coolant (~200–300 lb), and the facility distribution — risers, mains, CDUs, and full pipe — adds permanent dead load on every deck that an air-cooled building never had to carry. This is load that is present 100% of the time, which makes its long-term deflection contribution worse than transient live load.
• Deflection and the rigid-pipe conflict. Elevated decks must hold tight deflection limits (commonly L/360 for the deck, tighter where rigid pipe is supported) because a floor that sags under load stresses the very pressurized coolant joints running across it. A leak from a deflection-driven joint failure is a structural problem that becomes an electrical-safety problem one floor down. Deflection control is therefore a coupled mechanical-structural requirement, not a comfort spec.
• Vibration. CDU pumps, large pipe flow, and rotating mechanical plant introduce vibration that a stiff slab-on-grade simply absorbs but a framed floor can amplify or transmit. Pump and pipe vibration is held to ISO 10816/20816-class criteria, and the structure must be stiff enough that resonance does not loosen connections or fatigue supports over years of continuous operation.

The net consequence: a multi-story liquid hall trades land savings for a structure that is heavier, stiffer, more deflection-controlled, and more expensive per square foot than its slab-on-grade equivalent. That can absolutely be the right trade in a land- or power-bound metro — but it is a trade, and the structural premium belongs in the siting math, not discovered after the frame is designed.

Pre-integrated racks and multi-ton CDUs do not levitate into place. The hall has to be designed for how heavy equipment arrives, moves, and gets replaced — and that is a structural provision, not an afterthought. Three load cases matter and are routinely under-scoped.

Rigging paths and dynamic loads. A 3,000 lb rack rolled on skates concentrates its full weight on a narrow caster path that moves across the floor, and a rack being lifted imposes a dynamic load (an impact factor above the static weight) at the lift point. The slab and any elevated deck along the rigging route — from the loading dock, through corridors, into the hall — must be rated for these moving concentrated loads, which are frequently higher than the resting load at the final position. A hall whose resting positions are fine but whose dock-to-row path crosses an under-rated deck cannot actually receive the equipment it was built to host.

Crane and hoist provisions. Overhead monorails, jib cranes, or gantry provisions for setting racks and swapping CDUs impose suspended dynamic loads on the roof structure and must be designed in from the start; retrofitting a crane rail into an occupied dense hall is effectively impossible. Knockout panels, oversized doors, and clear-height for equipment ingress are cheap at design time and effectively impossible to add later. Reserving the rigging corridor, the crane load path, and the ingress envelope is the structural expression of the same density-ramp discipline that governs floor loading: you cannot retrofit the move-in path, so you provision it. Construction sequencing and the equipment-set phase that exercises all of this live in Chapter 6.6.

An AI rack in a seismic zone is a uniquely unforgiving structural object: it is heavy, plumbed with pressurized coolant, and energized at 415 VAC or 800 VDC, all at once. An earthquake that merely tips a server rack is an annoyance; an earthquake that ruptures a coolant line above an energized power bus, or shears a quick-disconnect under a moving rack, is a flood-into-electrical event. Seismic protection is therefore not only life-safety and equipment-protection — it is the thing that keeps the liquid and the electricity apart when the ground moves. Increasingly it is also an insurability requirement: insurers in California, Japan, and other active zones now mandate seismic provisions for high-value compute, which links this decision directly to Chapter 2.6.

The design ladder runs from cheapest-and-stiffest to most-protective-and-costly. Anchorage ties racks and equipment rigidly to the slab so they ride with the building — adequate in low-to-moderate hazard, but it transmits the full ground acceleration into the racks and, critically, into the rigid coolant piping connected to them. Stiffened/braced structure raises the building's own seismic capacity. Base isolation — ball-bearing or elastomeric platforms that decouple the equipment from ground motion, permitting on the order of 12 inches of horizontal travel while holding vertical stability — is the high-hazard answer, and it changes the plumbing problem entirely: a base-isolated platform that moves a foot relative to the building demands flexible coolant connections, slack power whips, and seismic loops in every pipe crossing the isolation plane, or the isolation itself shears the very services it was meant to protect.

Every load path discussed above ultimately terminates in the ground, and the ground is the part of the structure you did not get to design. A slab-on-grade is only as strong as its subgrade bearing capacity; an elevated frame's deep foundations are sized to the soil's resistance and settlement behavior; a base-isolation system's performance assumes a defined site-response spectrum. This is why the structural basis cannot be set independently of the geotechnical basis — the two are one decision viewed from two ends. A site with poor bearing soils, high water table, expansive clays, or a stiff seismic site class can force a fundamentally different and more expensive foundation system, or push a planned slab-on-grade toward piles and pile caps, materially changing the civil cost and schedule.

The practical sequencing consequence: the geotechnical investigation is a critical-path input to the structural design, not a parallel workstream. You cannot finalize slab thickness, reinforcement, foundation type, or seismic detailing until the soils report and the site-specific seismic hazard are in hand. Skipping or compressing the geotech to save schedule is a classic way to discover, after the frame is half-designed, that the floor-loading basis the whole density roadmap depends on rests on soil that cannot carry it. The full treatment of land, soils, seismic hazard, and flood diligence — the secondary screen that nonetheless gates this entire chapter — lives in Chapter 3.8.