Building Typologies & Data-Hall Layout

The legacy data center building was a solved problem dressed up as an architecture brief: a single-story tilt-up shell, a raised floor, a perimeter of CRAH units, and as much white space as the developer could lease. The IT was the point and the building was the wrapper. The AI factory has demolished that model. When a GB300 NVL72 rack carries a facility design basis of up to 142 kW, the cooling plant, the electrical rooms, and the heat-rejection yard swell until they can dominate the developed site footprint, and the data hall, the part everyone still calls "the data center," becomes a minority tenant inside its own mechanical and electrical envelope. The building is now the plant; the data hall is what is left over. VR200 NVL72 carries a 330 kW cabinet facility design basis, and NVIDIA has announced Kyber-class racks at ~600 kW for H2 2027.

This chapter turns that inversion into layout decisions, each with a downstream cost, and they resolve in sequence. First the greenfield-vs-retrofit and powered-shell-vs-integrated split that decides who owns the density risk; then the single-vs-multi-story choice driven by power, land, and structural economics; the data-hall layout itself (white space, gray space, adjacencies, plant rooms, and the yard); the spatial planning for liquid that a 2026 hall depends on (CDU galleries, pipe racks, leak-zoning, service access); and the repeatability and future-flex discipline of bay sizing, knockouts, and reserved headroom that separates a building you can ramp from one you have to abandon. The civil and structural basis for these layouts (slab loading, long-span steel, seismic anchoring) is owned in Chapter 6.2; the envelope and site civil works in Chapter 6.3. This chapter owns the shapes.

Before any layout, decide what you are actually building. Greenfield is a clean sheet on raw land — maximal control over column grid, slab loading, ceiling height, and plant adjacencies, at the price of the longest schedule and the deepest capital commitment. Retrofit reuses an existing shell, and inherits its constraints as hard physics: a hall poured for 5–10 kW air-cooled racks can inherit inadequate floor loading, no reserved route for liquid distribution, insufficient electrical headroom, and no facility water — the same cooling cliff treated in Chapter 5.1, now expressed as a building you cannot bend. Retrofit buys schedule and capital; it cannot buy density past what the slab and the structure already allow.

Cutting across that is the delivery model: do you build one integrated facility, or split it into a base building (the powered shell) and a fit-out? The powered-shell model separates the slow 2026 building program from the tenant’s faster IT selection. A developer delivers the envelope, the structure, the utility power to the boundary, and the primary distribution (the slow, irreversible, capital-heavy substrate) and hands a weather-tight, energized shell to a tenant who installs the IT-specific fit-out (racks, CDUs, the technology-cooling loop, the network) on their own schedule and to their own generation. The split moves the fast-moving, generation-specific risk to whoever is best placed to carry it, and it lets the shell start two years before anyone has decided which accelerator goes inside. JLL’s 2026 capacity outlook frames the demand pressure behind this split (Chapter 0.1); a tenant can select its accelerator later while the developer advances the shell, but the integrated schedule must still prove that the split advances service.

The downstream cost of the split is an interface. Someone must own the boundary between base building and fit-out — the busway tap-off points, the facility-water headers and valved stubs, the floor-loading guarantee, the knockouts for future risers. Get the interface wrong and the shell is energized but un-fillable: the water stub is in the wrong bay, the busway is rated for last generation, the shell has only an area-average floor letter while OEM foot/wheel reactions, rolling loads, the complete floor assembly, and the rigging path remain unverified. The powered shell preserves the option to choose your IT generation later; it does so only if the irreversible substrate was over-provisioned to the ramp, not to the lease that signed it.

The default data center is single-story, and for good reason: heavy equipment sits on grade (on a slab that is still accepted only after project-specific geotechnical, thickness/reinforcement, joint, concentrated-reaction, punching, rolling-load and anchorage checks); pipe runs are horizontal; and a leak goes to a floor drain rather than onto the hall below. Single-story is the low-risk shape, and where land is cheap and abundant it remains the right answer — most US greenfield AI campuses are sprawling single-story halls precisely because the constraint is power and land, not vertical area.

Multi-story is what you build when land is the binding constraint — dense metros, sovereign sites, water-adjacent or grid-adjacent parcels you cannot enlarge — or when stacking shortens the electrical and fiber runs enough to matter at gigawatt scale. The 2026 generation has made multi-story harder, not easier, because the things you are now stacking are heavy and wet. A multi-story liquid-cooled hall must carry 2,000–3,000+ lb wet racks on an elevated structural slab rather than slab-on-grade, which drives a far stiffer (and costlier) structure to control deflection and vibration; it must route the technology-cooling loop and facility water vertically, adding risers, riser cores, and a leak path that now points downward at the IT below; and it must solve heat rejection on a roof or yard that serves several stacked halls at once. The downstream cost of stacking is paid in structure, in vertical liquid distribution, and in the leak-cascade risk that single-story simply does not have.

Single-story is the default; multi-story is a deliberate trade of structural and liquid-distribution cost for land efficiency and run-length. Make it because land or grid adjacency forces your hand, not because the renders look impressive. And if you stack, the structural basis (elevated-slab live loads, deflection, vibration, coolant-weight) is owned in Chapter 6.2, and the rack-as-civil-load verification in Chapter 6.7 — this chapter only sets the shape that forces those problems.

Inside the building, two terms govern everything. White space is the conditioned floor where the IT lives — racks, network, and the immediate cooling at the rack. Gray space (or back-of-house) is everything that supports it: the electrical rooms (switchgear, UPS, transformers), the mechanical plant (CDUs, pumps, heat exchangers), the air-handling, the BMS room, and the staging and storage. In a legacy enterprise hall, white space dominated and gray space was a thin perimeter — the building was mostly IT floor, and the developer leased it by the square foot.

The AI factory has inverted that ratio, and that inversion is the layout fact that governs everything downstream in 2026. When the rack draws 132 kW instead of 5 kW, the power chain feeding it and the cooling plant rejecting its heat both swell by more than an order of magnitude — so the indoor gray space (electrical rooms, UPS, switchgear, CDU galleries, mechanical plant) can claim as much of the building's gross floor area as the white space or more, depending on which plant is indoors. The heat-rejection yard, substation and generator field take site area on top of that, and the white space shrinks to a dense, valuable core. Those are two different denominators — building area and developed site area — and a plan that mixes them under-sizes one of them. A hall planned on legacy white-to-gray ratios runs out of plant room and electrical-room space long before it runs out of IT floor: you have a beautifully proportioned data hall and nowhere to put the megawatts of switchgear and the acres of heat rejection it needs. Plan the gray space first. Size the electrical rooms and the cooling plant to the megawatts, then fit the white space into what remains — the opposite of the legacy instinct.

The adjacencies follow from that. Electrical rooms want to sit close to the racks they feed to keep busway and cable runs short and losses low, but they must be fire-compartmented from the hall. The cooling plant and CDU gallery want to be close to the white space to keep the technology-cooling loop short, low-pressure-drop, and fast to drain — but the heat-rejection equipment (dry coolers, cooling towers, chillers) belongs outside in the yard, where it competes for setback-limited space with the substation, the generators, the fuel, and the BESS (the yard layout itself is owned in Chapter 6.3). And every layout must reserve a laydown and staging area: dense racks arrive factory-integrated and are rolled in on rigging paths, so a hall with no laydown space and no clear move route cannot actually be populated on schedule. The construction-sequencing and rigging consequences are owned in Chapter 6.6.

Air-cooled halls had a simple spatial logic: hot aisle, cold aisle, containment, and a plenum. Liquid changes the problem from airflow to plumbing, and plumbing is far less tolerant about space. A direct-to-chip hall must find room for four things air never needed, and each one is a layout decision with a service-access consequence.

CDU galleries. The coolant distribution unit isolates the clean, controlled technology-cooling loop (the secondary loop, to the cold plates) from the facility water loop (the primary) and provides the pumping, heat exchange, filtration, and dew-point control for the rack-side fluid — its internal architecture and sizing are owned in Chapter 5.6. Spatially, CDUs are large, heavy, serviceable machines that need a home: either in-row (a CDU sitting in the rack line, close-coupled to a few racks, easy to drain but consuming white-space floor) or in a dedicated CDU gallery along the hall perimeter or in adjacent gray space (centralized, serviceable without entering the IT rows, but with longer secondary-loop runs). The choice trades white-space density against service access and loop length, and it must be made before the column grid is fixed, because the gallery's footprint and the manifold routing depend on it.

Pipe racks and risers. Liquid distribution needs a physical route: overhead pipe racks carrying supply and return headers down the hall, drop legs to each rack-row manifold, and — in multi-story — vertical riser cores connecting floors. This is volume that air-cooled halls spent on a deep plenum and now spend overhead and in dedicated chases. Under-size the pipe-rack space and you cannot add capacity without re-routing live plumbing; forget the riser cores in a multi-story shell and you have stranded the upper floors. The detailed hydraulic design of these loops lives in Part 5 — this chapter only insists that the space for them be reserved at layout time.

Leak zoning. Liquid in the white space is a fault mode air never had. Good layout treats the hall as a set of leak zones — bounded regions with their own leak detection, bunded/curbed containment, and isolation valves — so a failed quick-disconnect or a manifold breach wets one zone and is isolated, not the whole hall, and (critically in multi-story) does not cascade onto the floor below. Leak zoning is a spatial discipline: it dictates where isolation valves sit, where containment curbing runs, and how the drainage is graded. It is cheap to design in and very expensive to retrofit after a hall is energized.

Service access. Every one of the above must be reachable without shutting the hall down. Concurrent maintainability — the Tier III and IV site-infrastructure property that required capacity components and distribution paths can be removed for planned maintenance while the critical load remains served (Uptime Institute) — is a layout property before it is an electrical one: if the only path to a CDU runs through a live IT row, or a manifold valve sits behind a rack, the hall is not concurrently maintainable no matter what the single-line diagram claims. Reserve the access aisles, the valve-pit clearances, and the CDU pull-out space at layout time, or pay for them in downtime later. Commissioning these loops and proving the leak response is owned in Chapter 13.5.

At gigawatt scale, no one designs a hall — they design a module and pour it many times. The unit of repetition is some standardized block (a data hall of a fixed size, a power-and-cooling skid, a pod) with frozen interfaces, replicated across a campus to a master plan. Vendor reference designs crystallize this: Vertiv's 2024 360AI reference block with NVIDIA is a roughly 7 MW data hall at 132 kW/rack with prefabricated power and cooling modules, sized so the electrical, mechanical, and structural interfaces repeat identically; current-generation blocks size to GB300 densities and above. The modular-construction tradeoffs — factory build vs stick-built, transport limits, commissioning-in-factory — are owned in Chapter 6.4; the master plan's job is to make the module tileable.

The trade-off in standardization is sharp. The upside compounds: the first module is slow and expensive to design and commission, but every subsequent one reuses the same grid, the same interfaces, the same commissioning scripts, and the same supply-chain order — schedule and quality both improve as the campus ramps. The downside also compounds: a mistake in the standard unit scales with the unit. An under-sized CDU gallery, a riser core that is one bay short, a floor-loading spec that just misses the next wet rack — replicated forty times across a campus, a small per-module error becomes a campus-defining liability. The master plan must therefore fix not just the module but the seams between modules: shared yard and substation capacity, inter-module fiber and busway, phased construction that lets one module be commissioned while the next is still steel, and expansion bays held in reserve for a generation not yet announced. The siting and yard context for the campus is owned in Chapter 3.1 and Chapter 6.3.

The defining condition of 2026 is that the density ramp is steeper than any building's depreciation. A hall scoped for today's GB300 NVL72 facility design basis of up to 142 kW will, within its structural life, be asked to host racks at 2x to 4x that draw. VR200 NVL72 is the intervening fit-out with a 330 kW cabinet facility design basis, and the Kyber-class ~600 kW rack is already announced for H2 2027. You cannot build for 600 kW on day one, and you should not try; the IT, the power chain, and the cooling plant for it do not exist at sane cost yet. What you can do is make the irreversible substrate accommodate the ramp while keeping the reversible fit-out matched to the current generation — that balance is what future-flex means.

Concretely, future-flex is a short list of cheap-now-or-impossible-later provisions. Bay sizing generous enough to absorb deeper, heavier racks and wider service aisles without re-gridding. Floor loading signed to a declared future wet-rack mass and reaction envelope, because strengthening a slab under operating racks requires a disruptive decant and cannot be priced from kilowatts. Knockouts — pre-formed openings, sleeves, and reserved chases in walls and slabs for future power busway, water headers, and fiber, so the next capacity step is a tap-off rather than a demolition. Reserved space for the CDU gallery, the pipe racks and risers, and additional electrical rooms, held empty (or lightly used) against a density step-up. And a documented reservation register that records, bay by bay, what headroom is committed and what is held — so the next team does not unknowingly fill the space the ramp needed.

The economics are lopsided in favor of reserving. Separate the two halves of that list before pricing it: knockouts, sleeves and reserved chases are interface reservations that cost almost nothing at design, while over-sized bays and a stronger slab are priced structural scope the engineer has to quantify per bay against the local baseline. Price the structural and service-route alterations together with the cooling retrofit scoped in Chapter 5.1, the stranded capacity, and the downtime from opening the occupied hall. Reserve the headroom you cannot retrofit; defer the spend you can. This is the same reversible-vs-irreversible logic derived in Chapter 5.1, applied to the floor plan.

Choose the building shape from separate indoor and site area accounts, then reserve the CDU gallery, riser, yard and rack replacement paths against a named equipment envelope. Spending that space twice buys a full-looking plan and a hall that cannot accept its next load.