Designing Data Centers for 100kW+ Racks: How Liquid Cooling is Reshaping Facility Layout in 2026

The era of 100+ kW racks is here. AI and high-performance computing demands are driving unprecedented power densities in data centers, forcing a fundamental rethinking of cooling and facility design. NVIDIA’s latest GPUs (Blackwell Ultra and the upcoming Rubin architecture) are pushing 1,800+ watts per chip, with some future accelerators rumored to reach 4.4 kW *per chip (www.tomshardware.com). The result: individual racks now pack 100–240 kW of IT load, far above the 10–30 kW per rack that traditional data centers were built to handle. At these densities, air cooling is physically incapable of removing heat fast enough – beyond roughly 30–50 kW per rack, fans and chilled air simply can’t carry away the thermal energy (www.techradar.com). This has made liquid cooling *mandatory for cutting-edge deployments.

Market indicators reflect this rapid shift. The data center liquid cooling market is projected to reach $38.4 billion by 2033 (growing ~28.7% CAGR) (www.globenewswire.com). Why? Because traditional air-cooled approaches are hitting a wall as rack densities escalate (www.globenewswire.com). Direct liquid cooling (pumping fluid directly to server components) has quickly moved from niche to necessity. Industry reports show direct-to-chip cooling is now the dominant approach for high-density deployments, thanks to its easier integration into existing form factors (www.globenewswire.com). Immersion cooling – where entire servers submerge in dielectric fluid – remains a smaller slice of deployments but is the fastest-growing segment as organizations grapple with the most extreme heat loads (www.globenewswire.com). In fact, many “AI factories” are experimenting with fully-immersed servers to wring out every last bit of cooling performance when space and power are at a premium.

But here’s what most discussions miss: adopting liquid cooling doesn’t just swap out the cooling system – it changes everything about facility design. When you introduce liquid-cooled 100kW+ racks into a data center, it ripples through the entire engineering playbook. Let’s dive into the specific design implications of this liquid cooling revolution, from new piping infrastructure and power distribution to floor layouts and structural design. We’ll then explore how ArchiLabs Studio Mode helps engineering teams navigate this complexity in the 2026 landscape of AI-driven data centers.

Beyond CRACs and Hot Aisles: New Cooling Infrastructure

Moving to liquid cooling means introducing an entirely new coolant distribution infrastructure alongside (or replacing) traditional air HVAC. High-density racks now require chilled liquid delivered directly to the rack and even to each server chassis. This involves deploying:

• Coolant Distribution Units (CDUs) – These are pumping and heat exchange stations that transfer heat from the server loop to the facility’s primary chilled water system. CDUs regulate coolant temperature, pressure, and flow. They’re often deployed as in-row or end-of-row appliances serving a cluster of liquid-cooled racks. Each 100kW+ rack might consume tens of liters of coolant per minute, so CDUs must be engineered with sufficient pump capacity and redundancy to avoid any single point of failure in cooling supply.
• Pipes, Manifolds, and Heat Exchangers – Facility piping must route coolant supply and return lines to every rack (and sometimes to individual servers). This means adding insulated pipework either under a raised floor or overhead in the rack aisle. Manifold assemblies at each rack distribute coolant to server inlets (for direct-to-chip cold plates) or to rear-door heat exchangers. All piping components need to withstand the pressure and flow rates of high-volume cooling – large-diameter supply lines, valves, and gauges become as critical as CRAC units once were.
• Drip-Free Quick Disconnect Fittings – Maintenance on liquid-cooled gear requires connectors that can be safely disconnected without spills. Industry-standard dry-break quick disconnect couplings (like the OCP’s Universal Quick Disconnect) allow staff to service or swap a server blade without fluid leaks (nationalhose.com). These leak-free, tool-free connectors use spring-sealed valves that automatically close when a line is unplugged, preventing even a drop of coolant from escaping. Designing the facility means planning for these fittings on every liquid-cooled server tray and providing drip containment just in case. Technicians must be trained in using locking quick-connects so that removing a node doesn’t turn into a slip-and-slide on the data center floor.

In short, a liquid-cooled data hall starts to resemble a hybrid of data center and mechanical plant. You’ll see chilled water manifolds and pipe headers alongside the rows of racks, and every rack is now part of the building’s plumbing system. Cooling distribution is no longer just air handlers and overhead ducts – it’s pumps, filters, pipe hangers, and heat exchangers woven through the facility. This requires close coordination between mechanical and IT design teams. For example, pipe routes must avoid blocking IT equipment access and maintain bending radius rules, much like cable trays, but with the added need for valves and expansion loops to handle thermal growth of pipes.

Rethinking Floor Layout and Structural Design

The shift to liquid cooling also forces a re-examination of the data center’s physical layout and structural support:

• Heavier Racks and Floor Loading: Liquid-cooled racks are significantly heavier than their air-cooled counterparts. A 48U rack filled with servers, plus the weight of water manifolds, cooling plates, and fluid volume, can easily weigh hundreds of kilograms more than a standard rack. If an immersion cooling tank is used, the weight can double (coolant fluid is ~1 kg per liter, and a full tank can hold several hundred liters). Data center floors – especially raised floor systems – must be evaluated for these point loads. Many next-gen facilities for AI choose slab concrete floors or reinforced pedestals under liquid-cooled racks to safely support the weight. Structural engineers need to account for dynamic loads too (e.g. the shifting weight if a coolant reservoir quickly drains or in seismic events where sloshing could occur). In practical terms, this means checking that floor ratings (lbs/sqft) aren’t exceeded and possibly adding steel plates or additional floor stands under heavy racks. Facility layout might cluster liquid-cooled racks in areas with enhanced floor support. If certain rows are earmarked for 200kW racks, those may sit directly on slab or have a shortened raised floor span, whereas lighter air-cooled IT can occupy standard raised floor sections.
• Rack Clearance and Aisle Spacing: With liquid cooling, the “guts” of cooling (pipes and hoses) often attach at the rear or top of racks. For example, a rack with a rear-door heat exchanger (RDHx) will have a radiator-like door on its back plus supply/return hoses coming out. Aisle widths may need to increase to accommodate these rear protrusions and to allow technicians to maneuver with tools to disconnect lines. Likewise, top-of-rack overhead piping might require higher ceiling clearances or careful positioning so that pipe manifolds don’t obstruct cable trays or maintenance ladders. Designers must ensure there’s adequate service clearance around racks for things like swinging open a rear-door cooler or replacing a pump box on top of a rack. This can impact the floor plan density – you might sacrifice one rack position per row to make room for plumbing assemblies or wider aisles, slightly reducing the rack count in exchange for maintainability.
• Raised Floor vs. Overhead Distribution: Traditional air-cooled data centers often use a raised floor to distribute cold air. With liquid cooling, that raised floor space could be repurposed for running coolant pipes (and leak trays), or facilities may abandon raised floors altogether. Many modern designs opt for overhead coolant distribution, mounting insulated supply/return lines above the racks and dropping flexible hoses down to each rack’s manifold. Overhead routing keeps liquids away from any sensitive electrical gear at floor level and makes leak detection easier (since any drip is immediately evident). It also frees up underfloor space for power cabling or allows a slab floor design. However, overhead piping must be engineered with proper hangers and seismic bracing – a ruptured overhead pipe could spray equipment below, so some designs include catch trays or double-walled piping for extra security. Underfloor piping, on the other hand, keeps the liquid lines out of sight but demands an impeccable leak detection system under the floor and easy access panels to reach valves. Many facilities will use a combination: for example, primary coolant loops run overhead across the room, then drop to underfloor manifolds that feed each rack from below, combining the benefits. The key is that routing of liquid lines becomes a first-class element of layout design, just like power busways or cable ladders.
• Hot Aisles, Cold Aisles… and Liquid Aisles? In air cooling, we worry about hot vs. cold aisle containment. With liquid cooling, the concept shifts: much of the heat is captured in the fluid and carried away before it ever warms the air. This can actually simplify airflow management – for instance, a rack with direct-to-chip cooling might only reject 10–20% of its heat to air (from components like DIMMs or VRMs that aren’t liquid-cooled). You might not need elaborate contained aisles for liquid-cooled zones, since the air heat is minimal. However, you still need some air cooling as backup and to handle residual heat. Many designs end up as hybrid air/liquid setups, where liquid removes the lion’s share of heat and a smaller CRAC or in-row cooler manages the rest. These hybrid layouts require careful planning: you don’t want the HVAC system overcooling an area that doesn’t need it, or worse, undercooling if the liquid cooling portion fails and suddenly the air system must pick up the slack. Some data centers are designing “liquid pods” – self-contained groups of racks with their own coolant distribution and backup air cooling – separated from traditional air-cooled rows. This modular approach can localize any mess (both thermally and in case of a leak) to the liquid-cooled pod.

Power Distribution: 800V DC and 100kW per Rack Changes the Game

Beyond cooling, power distribution architecture undergoes a radical change at 100+ kW per rack. Pushing hundreds of kilowatts through a rack isn’t as simple as beefing up a few PDUs – it demands new strategies for delivering power efficiently and safely:

• From 480V AC to 800V DC: Hyperscalers are converging on high-voltage DC distribution (±400V DC bus) to feed these ultra-dense racks (www.techradar.com). In legacy facilities, a typical rack might be fed by dual 208V or 415V AC circuits, and inside each server, power supplies step that down to DC. But at 200kW+ per rack, the copper losses and conversion losses of AC distribution become untenable. Instead, many next-gen designs are adopting 600–800V DC busways that deliver DC right to the rack power shelf. For example, Google, Meta, and Microsoft have jointly backed an Open Compute design (the “Mt. Diablo” initiative) standardizing ~+/-400V DC (nominal ~800V) distribution for AI racks (www.techradar.com). NVIDIA also recently partnered with power systems vendors to roll out 800V DC rack power architectures supporting its megawatt-scale AI deployments (www.tomshardware.com). High-voltage DC offers big benefits: it cuts out one or two conversion stages, improves efficiency (no need for large AC UPS in the middle), and drastically reduces bus bar current for a given power level. That means thinner (or fewer) cables can carry the load, and voltage drop is less of a concern over distance. The tradeoff: new equipment is required at the rack level to convert 800V DC down to the various voltages used on motherboards (48V, 12V, etc.) – essentially moving the AC-DC conversion out of each server and into centralized rack-level rectifiers or bus converters. Facility designers must allocate space for these DC power shelves, which often include their own liquid cooling or advanced thermal management (because an 800V, 250kW power converter itself generates a lot of heat!). Also, safety and redundancy need careful thought – 800V DC is lethal and doesn’t have natural zero-crossings like AC, so emergency off switches, interlock systems, and arc-flash protection become critical in design. In many cases, the high-voltage DC is supplied from centralized rectifiers backed by battery systems (essentially creating a DC UPS plant), distributing DC down the line to each row of racks.
• Busways and Overhead Power Feeds: With such massive currents, many facilities are moving away from traditional whips and PDUs to overhead busway systems. Busways can carry a high amperage three-phase (for AC) or high-voltage DC along the length of a row, and drop-down taps feed each rack. These busways are now being adapted for HVDC. The benefit is that you can avoid running multiple thick cables underfloor to each rack – instead a rigid copper or aluminum bus bar system runs above the racks, and each rack connects via a tap-off box. For a 150kW rack, instead of six 30A 208V cables, you might have a single 800V DC hookup at maybe 200A. The layout needs to accommodate these bus ducts and the clearance they require. It also means the power distribution units (PDUs) in the room, if still used, are different – they are more like DC switchgear than the breaker panels of old. Data centers embracing liquid cooling often simultaneously upgrade their power rooms with larger transformers or rectifiers (to feed HVDC), smart switching gear, and advanced monitoring (to track per-rack power at unprecedented granularity and respond very quickly – for instance, if a liquid cooling pump fails and a system starts to overheat, you might need to throttle power or shut down servers in milliseconds).
• Power Density and Heat Rejection: One often overlooked aspect is that all this power distribution equipment itself generates heat. Electrical losses in busbars, conversion electronics, and switchgear at these power levels are non-trivial. For example, a 1 MW power feed at 97% efficiency still dumps 30 kW of heat – which needs cooling. That’s why some emerging designs actually liquid-cool the power infrastructure too. We’re seeing experimental water-cooled PDUs and transformers, where coolant loops run through busbars or around conversion electronics to draw off heat. This keeps the electrical rooms smaller and reduces air conditioning loads. In a fully liquid-cooled facility, not only the servers, but also the rectifiers, backup generators, and even the battery systems might use liquid cooling. The facility layout might then include secondary coolant loops for the power equipment, separate from the IT cooling loops. It’s a holistic approach: you cool the servers and the power delivery chain with water, capturing essentially all heat at high efficiency. In 2026, only the most cutting-edge hyperscalers are doing this, but it’s a design direction to watch.

Operational and Hybrid Challenges (Air + Liquid)

Introducing liquid into a data center doesn’t eliminate all the traditional concerns – it adds new ones. Day-to-day operations and maintenance routines must be adapted:

• Maintenance Protocols: With air-cooled servers, swapping hardware is straightforward – pull the server out, replace it, standard ESD precautions, done. In a liquid-cooled environment, maintenance might involve draining a server’s coolant loop or using blind-mate quick disconnects to remove a blade without spilling. Technicians need training on handling liquid-cooled gear: verifying valve positions, checking for trapped air in lines, and wearing appropriate PPE (face shields are wise when dealing with pressurized coolant!). Procedures for leak detection and response become part of standard operating procedure. For example, if a leak sensor trips, do you know which valve to shut off first? Can you isolate a single rack’s cooling loop without impacting others? These questions should be answered in the design phase by including isolation valves and segmented loops. Routine maintenance on the cooling system itself (pumps, filters, coolant quality checks) also enters the picture – facilities may need to schedule periodic fluid testing, pump refurbishments, or heat exchanger cleanings, none of which were relevant in air-cooled sites. This is why designs often include redundant pumps and bypass loops so maintenance can occur without shutting down the IT load. From a layout perspective, you’ll allocate floor space for things like coolant expansion tanks, filtration units, and pump control panels, ideally in an accessible spot.
• Leak Detection and Containment: While modern liquid cooling components are highly reliable (many use double O-ring seals and undergo helium leak testing (amphenol-industrial.com)), the facility must be designed assuming a leak will eventually happen. Leak detection cables or sensors typically run under each rack and beneath any overhead piping. If even a few drops of liquid hit the sensor, it triggers an alert (and can automatically shut valves). Containment pans or sloped floors can direct any spilled coolant to a drain or holding tank rather than letting it drip onto active equipment. Engineers sometimes design the slab with a slight slope in liquid-cooled areas, so any leaked fluid flows away from server rows and toward a safe collection point. Additionally, coolant choice matters here – many data centers use water-based glycols for direct-to-chip cooling, which are electrically conductive (so a spill on electronics is bad news). Others use dielectric fluids for immersion, which won’t short-circuit electronics but can still ruin components by seeping into connectors. In either case, cleaning up a spill is a specialized task (involving vacuum pumps or absorbent mats), so having clear access around racks is important. Spacing, as mentioned earlier, should allow a technician to get to the rear of a rack with a cart or containment vessel if they need to quickly address a failure.
• Hybrid Cooling Zones: During this transition era, many facilities operate with a mix of air-cooled and liquid-cooled equipment. This hybrid situation can be tricky. You might have one row retrofitted with direct liquid cooling while the adjacent row is legacy air-cooled servers. The environmental requirements differ: the liquid-cooled row can run with higher inlet temperatures (since liquid removes most heat), whereas the air-cooled row might still need cooler airflow. Balancing the room’s cooling system to handle both is an art. One strategy is to physically separate zones – e.g. put liquid-cooled racks in their own enclosed pod with a dedicated coolant loop and an air handler in standby mode, while the rest of the room remains on CRAC units. Another strategy is to use rear-door heat exchangers for the transitional racks: these radiator doors allow the rack to be cooled by facility water without needing to modify the servers themselves. Rear-door coolers can remove ~70-80% of heat at the rack, dramatically easing the load on AC for that row. This technology is seeing rapid adoption as a “bridge” solution where full direct liquid integration isn’t feasible immediately (www.globenewswire.com). A data hall might upgrade 10 racks with rear-door cooling as a pilot, before committing to full direct-to-chip cooling for the next phase.

Managing hybrid environments requires dynamic monitoring. If an air-cooled server row starts running hot, you might temporarily boost chilled water flow to the neighboring liquid-cooled racks’ rear doors, effectively using them as heat sinks to absorb some extra heat from the room – a creative but effective trick. Computational fluid dynamics (CFD) modeling and digital twins are increasingly used to tune such scenarios. In fact, design teams are now using simulation tools early in the process to model both air and liquid cooling performance together, ensuring one doesn’t create hotspots for the other (www.techradar.com). The complexity of mixed cooling drives the need for integrated design platforms that can handle thermal, mechanical, and electrical trade-offs simultaneously (more on that shortly). Everyone from mechanical engineers to IT architects must collaborate when a facility has both an air plenum and a liquid loop intertwined.

Designing for Complexity: How ArchiLabs Studio Mode Simplifies Liquid-Cooled Data Center Design

When you’re dealing with 100kW racks, dual cooling systems, high-voltage power, and all the nuances we’ve discussed, designing a data center becomes a massively interdisciplinary puzzle. This is where ArchiLabs Studio Mode comes into play. ArchiLabs Studio Mode is a web-native, code-first parametric CAD platform built specifically for the modern era of AI-centric infrastructure. Unlike legacy desktop CAD tools (which often bolt on scripting as an afterthought), Studio Mode was designed from day one for automation and AI-driven design. In practice, this means complex data center layouts – with all their power, cooling, and structural constraints – can be rapidly prototyped, validated, and iterated in a single unified environment. Below, we highlight how ArchiLabs Studio Mode addresses the key challenges of liquid-cooled facility design:

• Smart Components Carry Intelligence: In Studio Mode, every object in the model can encapsulate data and rules. For example, a rack component “knows” its own power draw, weight, and cooling requirements. Place a 200kW liquid-cooled rack into your layout, and it brings along clearance rules for its rear-door coolant hoses, its maximum floor loading, and the requirement that it be within X meters of a CDU. Likewise, a cooling manifold component can enforce how much flow it can handle or what diameter piping it needs. These smart components drastically reduce guesswork – the expertise of your best engineers (e.g. “don’t put more than 4 of these 50kW servers on one coolant loop or it’ll exceed flow capacity”) is embedded as properties of the component. As a designer, you snap together the model like LEGO, and the rules ensure those pieces make sense together.
• Integrated Power and Cooling Modeling: Studio Mode doesn’t treat electrical and mechanical design as separate silos. Because components can have multiple attributes, you can model both the power distribution and the cooling loop in the same parametric model. For instance, you might define a 600V DC bus object that runs above a row of racks, and simultaneously overlay a coolant supply pipe model following that same route. The platform can then check clearances – are the bus duct and piping colliding? – and even do basic co-analysis, like ensuring the heat from the bus doesn’t warm the coolant pipe above acceptable limits. By having a single source of truth model that includes IT equipment, cooling circuits, and power feeds, ArchiLabs lets you catch spatial and functional conflicts early. No more getting to a construction drawing review and realizing the electrical ladder tray blocks the cooling manifold valves – Studio Mode would have flagged that in the 3D model when you attempted to route them through the same space.
• “Recipes” for Automated Liquid-Cooled Layouts: One of the most powerful features of ArchiLabs Studio Mode is its Recipe system. Recipes are essentially parametric scripts or workflows – they can be written by domain experts in Python (or even generated by AI from plain English instructions) to automate design tasks. For liquid cooling, imagine a recipe that takes an input row length and power density, and auto-generates a liquid-cooled pod layout: it could place the required number of racks, insert a CDU at the end of the row, auto-route the supply/return piping to each rack with proper bends and support brackets, and even populate cable trays or power bus connections. All following best-practice rules encoded in the recipe. For example, a recipe could enforce: “If row has more than 10 racks at >50kW each, insert a second CDU for redundancy,” or “Use 2-inch diameter pipe for the first 5 racks, then 3-inch main line thereafter.” These recipes are version-controlled and reusable, so your team’s hard-earned design standards become repeatable automation – no more starting from scratch for each new project. ArchiLabs’ approach means you can quickly generate multiple design alternatives: one recipe run might configure rear-door coolers, another might try direct-to-chip cold plates, each with different pipe routing. You can then compare which design uses less floor space or meets the cooling capacity with more headroom.
• Real-Time Rule Checking and Validation: Designing complex systems often leads to errors slipping through (like forgetting a clearance, or overloading a floor section). Studio Mode makes validation proactive: as you build the model, it continuously checks against a set of validation rules. These rules can be anything from simple (“no more than 30 kW per air-cooled rack in this zone”) to complex (“pipe bends must not reduce flow below X GPM given the pump spec, otherwise flag it”). If you try to place a heavy immersion tank on a raised floor section that can’t support it, the platform will flag a structural loading violation immediately – before you issue drawings or, worse, before something fails on site. Thermal capacity checks are similarly integrated: the model can calculate total BTU removal for each cooling loop and compare it against the CDU capacity. If you add another rack to a cooling loop that exceeds what the heat exchanger can handle, a warning notification pops up. This kind of multi-domain validation (covering thermal, structural, electrical clearances, etc.) is usually done manually across spreadsheets and meetings – but ArchiLabs does it in-software, ensuring that by the time a design is finalized, all the big errors are already caught. As an example, consider mixing cooling methods in one room: Studio Mode could have a rule that requires any row with both air and liquid-cooled racks to have a certain minimum AC cooling backup. If you violate that by removing a CRAC unit, it’ll alert you to the cooling shortfall before any equipment is purchased or installed.
• Parametric Design and Rapid Iteration: Because the platform is fully parametric (supporting all the usual CAD operations like extrude, sweep, boolean, fillet, etc., but driven by code or constraints), making changes to try new ideas is fast and reliable. Let’s say you want to explore different cooling strategies for a 20-rack pod: You can model a baseline with direct-to-chip cold plates, then branch the model (like a git branch) and swap those out for immersion tanks. The geometry and metadata update consistently – perhaps the immersion scenario needs fewer CRAC units, different power feed configuration, and a changed floor layout. Studio Mode’s git-like version control means you can branch layouts, compare diffs, and merge improvements. Maybe one engineer works on a branch adding overhead busways while another tweaks the pipe routing – the platform can merge their changes intelligently, because all design decisions are recorded as parametric operations (with an audit trail of who changed what and when). This level of traceability is a game-changer for large engineering teams: it brings software development rigor to facility design. Mistakes and changes are no longer a mystery; you can pinpoint which parameter change caused, say, a clearance issue and roll back if needed.
• AI-Driven Design and Integration: ArchiLabs Studio Mode was built with an API-first mentality, making it ideal for AI-driven workflows. You can literally ask an AI agent to generate a design – for example, “Layout a 5MW data hall with 40 liquid-cooled racks and 20 air-cooled racks, optimize for minimal piping length,” – and because the platform’s interface is code, the AI can interact with it as naturally as a person would click buttons. The result is not a black-box output; it’s a parametric model you can inspect and tweak. Moreover, ArchiLabs doesn’t lock you into its world – it acts as a hub for your existing tools. Through connectors, it keeps your design data in sync with everything from Excel spreadsheets to legacy CAD like Revit, to DCIM databases. For instance, you could have a live link that pulls the latest asset list from your DCIM software and updates the model (adding any new servers to racks), or conversely, have the CAD model auto-generate a bill of materials and push it to your procurement system. Industry interoperability is built in: need to deliver IFC models for BIM coordination? Studio Mode can export to Industry Foundation Classes (IFC) format for architects (en.wikipedia.org). Working with a consultant who uses AutoCAD? You can generate DXF drawings (en.wikipedia.org) on the fly from the 3D model. All changes remain under version control, so the “single source of truth” is always maintained. In one unified platform, you have geometry, data, logic, and integrations, which greatly streamlines the complex dance of data center design and construction.
• Domain-Specific Content Packs: Designing a data center is very different from designing a house or a car. ArchiLabs recognizes this by providing content packs – libraries of components, rules, and workflows – tailored to specific domains like Data Center / MEP, as well as others for architecture or industrial facilities. The beauty is these aren’t hard-coded features; they’re modular content. If your team has particular standards (say, a custom liquid cooling skid design or an in-house procedure for commissioning tests), you can extend the platform without waiting on a software update – essentially, teach the platform new “skills” by adding to the content pack. This means ArchiLabs Studio Mode stays flexible and future-proof: as new cooling technologies emerge (perhaps two-phase immersion or on-chip microfluidics from research labs), you can model and incorporate them by updating your component library and rules. It’s a system that evolves with the industry. In contrast, legacy CAD or DCIM tools often lag years behind emerging tech – but a code-first platform can adapt immediately as your best engineers codify the new best practices. For example, when OCP introduced that new Universal Quick Disconnect (UQD) standard, an ArchiLabs user could create a UQD component with its metadata (pressure rating, size, clearance) and drop it into all designs, ensuring any model going forward uses the approved connector and flags any old component that should be replaced.

In essence, ArchiLabs Studio Mode brings software agility to hardware design. Designing around 100kW racks and liquid cooling is no easy feat – it involves juggling mechanical, electrical, and operational considerations simultaneously. Studio Mode’s AI-first, automation-first approach means you can iterate through solutions far faster and with greater confidence in their correctness. Teams at neocloud providers and hyperscalers are using these kinds of tools to stay ahead of the capacity planning curve, letting them prototype a 100 MW facility (with hundreds of liquid-cooled racks, power systems, and miles of pipe) and test “what-if” scenarios in hours – something that used to take weeks of siloed effort across different departments.

Conclusion

The rise of 100+ kW racks is fundamentally reshaping data center design. What started as a quest to cool ultra-hot chips has cascaded into new paradigms for facility layout: mechanical plant systems running right into the white space, power delivery more akin to an electrical substation than a traditional server room, and new operational protocols merging IT with facilities management. In 2026, designing a cutting-edge data center means designing an integrated ecosystem of power, cooling, and technology, all tightly bound. Liquid cooling is at the heart of this shift – not only enabling the latest AI hardware to run at full tilt, but also driving engineers to rethink how we build and operate the digital infrastructure of the future.

For data center designers and planners, the mandate is clear: embrace the tools and methodologies that let you manage this new level of complexity. Whether it’s adopting liquid cooling hardware or adopting advanced design automation platforms, the goal is the same – to deliver reliable capacity for ever-denser compute, on tight timelines and budgets, without letting the complexity overwhelm you. Platforms like ArchiLabs Studio Mode are stepping up to assist in this new era, providing a means to capture institutional knowledge as code, automate the grunt work, and ensure that before a single pipe is laid or server installed, your design is holistically validated against the demanding criteria of high-density, liquid-cooled operation.

The data center is evolving rapidly under the pressures of AI and cloud growth. By understanding how liquid cooling changes everything – and harnessing the right design approaches – engineering teams can turn what seems like a daunting challenge into a competitive advantage. A facility built for 200kW racks is fundamentally a different animal than yesterday’s data centers. As we’ve explored, it’s not just cooling – it’s the entire nervous system of the building that changes. With knowledge, preparation, and the aid of next-gen design platforms, we can architect these new digital engines to run safely, efficiently, and sustainably for the next decade and beyond.

Liquid cooling isn’t just a trend; it’s a transformative force reshaping data center architecture from the ground up. The companies that design with this in mind – from piping to power to software – will be the ones that successfully ride the next wave of the cloud revolution. The ones that don’t? They’ll be left trying to jam 10 pounds of heat in a 5 pound sack, and that’s not a strategy any of us can afford in the age of AI.