Data Center Liquid Cooling Boom: The Next Growth Engine for Heat Exchanger Industry
The global data center industry is undergoing a thermal inflection point. For most of the past two decades, removing heat from servers meant moving air — bigger fans, tighter hot-aisle containment, more kilowatts spent on computer room air conditioning. That paradigm is collapsing under the weight of AI workloads. And the infrastructure layer that stands to benefit most is not the server OEM, not the cooling tower vendor — it is the industrial heat exchanger manufacturer.
This article examines the mechanics of that shift, the engineering requirements it creates, and why ASME-certified shell-and-tube heat exchangers are becoming a specification item in data center primary and secondary cooling loops.
The Numbers Behind the Shift
The global data center liquid cooling market was valued at approximately $3.5 billion in 2024. By 2030, industry estimates place that figure at $12.8 billion, a compound annual growth rate of roughly 24%. To put that in context: the broader data center infrastructure market grows at roughly 10–12% CAGR. Liquid cooling is growing at more than twice that rate.
The driver is not incremental. AI accelerator chips — GPU clusters used for training large language models and inference workloads — generate heat densities that air systems fundamentally cannot manage at scale. Two years ago, a high-density AI rack might dissipate 300 watts per rack unit. Today's H100 and Blackwell-architecture GPU nodes push 1,000 watts per rack unit or more. A single 42U rack running fully populated AI accelerators can generate 40–80 kW of heat — more than a small apartment building's peak electrical draw.
Traditional CRAC/CRAH-based air cooling, even with in-row and rear-door supplements, hits a practical ceiling around 20–25 kW per rack for sustained loads. Beyond that, you are either over-provisioning white space (expensive, inefficient) or accepting thermal throttling (unacceptable for billable GPU compute). Liquid cooling is not optional for the AI era — it is load-bearing infrastructure.
Three Liquid Cooling Architectures — And Where Heat Exchangers Sit
Understanding the heat exchanger opportunity requires mapping the three dominant liquid cooling architectures now being deployed at scale.
1. Direct Liquid Cooling (DLC) with Rear-Door Heat Exchangers
In DLC deployments, coolant flows through cold plates attached directly to CPUs and GPUs, then returns warm to a rear-door heat exchanger (RDHx) mounted on the rack. The RDHx transfers heat from the server-side fluid loop to a building chilled water loop.
For facility engineers, the rear-door HEX is a relatively accessible retrofit path — existing racks can be retrofitted without touching servers. The HEX itself is typically a brazed plate type for the rear-door application, but the facility-side cooling distribution infrastructure — the CDU that serves an entire row or hall — requires shell-and-tube or brazed plate heat exchangers at a much larger scale.
2. Immersion Cooling
In single-phase and two-phase immersion systems, servers are submerged in dielectric fluid. The fluid absorbs heat directly from all components and is circulated to an external heat exchanger that dumps heat to facility water. This is the highest-density application: immersion tanks can handle 100+ kW per unit, and the heat exchanger connecting the dielectric loop to the building's cooling water loop is a critical pressure-boundary component.
Two-phase immersion in particular involves fluorocarbon or engineered fluid at controlled temperature and slight positive pressure — a service condition that plate-and-frame HEX can handle at small scale but where shell-and-tube designs begin to offer meaningful advantages in pressure robustness and cleanability.
3. Cooling Distribution Units (CDUs)
The CDU is the hydronic heart of a liquid-cooled data hall. It receives chilled water from the building plant (or an adiabatic cooler / dry cooler), conditions it to the precise supply temperature required by the server-side loop, and circulates it to rack-level distribution manifolds.
The heat exchanger inside a CDU — or between the CDU and the building chilled water system — is where industrial heat exchanger manufacturers find their most natural entry point. This is not a consumer or IT product. It is a pressure vessel with industrial service conditions: DI water or glycol on one side, chilled water on the other, operating at pressures that can reach 10–16 bar, with strict fouling requirements, corrosion-resistant materials, and in many cases third-party certification requirements.
Engineering Specification: What Data Center HEX Actually Requires
This is where the technical depth matters. A data center heat exchanger is specified differently from a petrochemical or marine HEX — but the engineering discipline is the same.
Material Selection: Why 316L Is the Starting Point
Deionized water (DI water) used in server cooling loops is one of the most corrosive common fluids in industrial service. Its very low conductivity means it aggressively leaches ions from metal surfaces — copper, carbon steel, and even standard 304 stainless steel are attacked over time by high-purity DI water circuits.
The baseline material specification for data center CDU heat exchangers is 316L stainless steel — the low-carbon variant of 316, selected for improved resistance to sensitization and intergranular corrosion in welded construction. For more aggressive applications (high-velocity DI water, elevated temperature), duplex stainless or titanium may be considered. On the secondary (building water) side, 304SS or carbon steel with appropriate treatment is typically acceptable depending on local water chemistry.
Lmart routinely fabricates shell-and-tube heat exchangers in 316L stainless steel per ASME Section II material requirements, with complete material traceability documentation — a requirement that data center operators with quality management systems will recognize as foundational.
TEMA Type Selection: Cleanability and Serviceability
For data center primary/secondary loop service, TEMA BEM and BEW configurations are the workhorses. The key consideration is cleanability: tube-side fouling from DI water is lower than in process service, but the shell side (building chilled water) can accumulate biological growth or mineral deposits over years of operation.
TEMA type B front head provides bolted removable channel covers for easy tube-side access without disturbing piping. For immersion cooling applications where the dielectric fluid side requires isolation and occasional flush, a BEW (floating head) or AES design allows tube bundle removal for inspection.
TEMA fouling factors for data center service:
- DI water (server loop): R = 0.000044 m²·K/W (treated water per TEMA Standards)
- Facility chilled water (conditioned): R = 0.000088 m²·K/W
- Glycol-water mixtures: R = 0.000088–0.000176 m²·K/W depending on glycol concentration
These are substantially lower than typical process industry fouling factors, meaning the HEX runs cleaner — but also that any deviation from expected fouling rates (due to inadequate DI water treatment, for example) will show up quickly as thermal performance degradation.
Nozzle Configuration for Modular Data Halls
Modern AI data center construction is increasingly modular — data halls are built in standardized blocks (typically 1–4 MW per module) that can be deployed rapidly. This modularity extends to the mechanical systems, and CDU heat exchangers need to accommodate nozzle configurations that align with the module's piping layout.
Custom nozzle orientation (rotated nozzles, bottom-entry/top-exit configurations for low-point drain and high-point vent) is standard practice for shell-and-tube fabricators. The critical detail for data center applications is that nozzle sizing must account for the relatively low-viscosity, high-flow DI water circuits — velocity limits of 1.5–2.5 m/s at tube inlets are appropriate to prevent erosion on 316L thin-wall tubes while maintaining adequate turbulence for heat transfer.
Pressure Rating and Code Compliance
Data center CDU heat exchangers typically operate at modest pressures relative to petrochemical service — 6–16 bar design pressure is common for facility-side chilled water loops. However, ASME Section VIII Division 1 U-stamp certification is increasingly specified by hyperscalers with engineering procurement standards that require third-party pressure vessel certification regardless of operating pressure.
The logic is sound: ASME certification ensures documented design calculation, material traceability, NDE, and hydrostatic testing under an authorized inspection body. For a facility that will operate for 15–20 years with minimal scheduled downtime, specifying ASME-certified pressure equipment on the cooling infrastructure is a risk management decision, not an over-specification.
Market Context: Who Is Building, and Where
The liquid cooling buildout is concentrated in two waves.
Wave 1 — Hyperscale (Now): Microsoft, Google, Meta, and Amazon are actively deploying liquid-cooled AI clusters. Microsoft's announced investments in AI infrastructure include liquid-cooled GPU clusters in Asia-Pacific; Google has disclosed liquid cooling deployments across its TPU/GPU cluster generations. These operators have internal engineering standards that increasingly specify liquid cooling for any rack exceeding 20–30 kW.
Wave 2 — China AI Compute Parks (2025–2026): China's national AI infrastructure push is generating a wave of new data center construction. Industry estimates suggest 30+ new AI compute parks are under construction or in late-stage planning in China through 2025–2026, concentrated in Inner Mongolia, Guizhou, Sichuan (cheap power), and near coastal industrial clusters. These facilities require the full stack of liquid cooling infrastructure, including CDU heat exchangers that meet Chinese GB standards as well as international certifications for export-format supply chains.
Southeast Asia and Japan: Singapore's moratorium on new data centers was partially lifted; Malaysia (Johor) is seeing major investment from US hyperscalers; Japan's national AI strategy includes significant data center capacity expansion. These markets frequently specify ASME or European pressure vessel codes for plant equipment.
Lmart's Position: The Infrastructure Layer
Lmart is not a data center IT equipment company. We do not manufacture cold plates, rack manifolds, or CDU control systems. What we manufacture is the industrial heat exchanger at the heart of the CDU — the pressure-boundary component that makes the primary-to-secondary loop heat transfer happen safely, reliably, and traceably over a 15–20 year facility life.
This is the infrastructure layer, and it has characteristics that favor manufacturers with industrial fabrication capability over IT-adjacent cooling vendors:
ASME U-stamp certification: Required by hyperscalers and institutional data center operators for pressure equipment. This is a non-negotiable specification item for operators with serious engineering governance. Lmart holds current ASME U-stamp authorization.
Custom fabrication to project specifications: Data center CDU heat exchangers are not catalog items. They are designed to a specific process duty, nozzle configuration, and material specification for each project. Lmart's engineering-to-order model is aligned with this requirement.
316L stainless steel construction: Lmart fabricates in 316L as a standard offering, with full material traceability per ASME Section II and customer QA requirements.
Low fouling factor design: Clean DI water service is a favorable application for shell-and-tube HEX. Lmart's standard tube layout and baffling design can be optimized for low-fouling, high-NTU service without special tooling.
GB 151 and ASME simultaneous compliance: For Chinese domestic data center projects, GB 151 shell-and-tube HEX certification may be required alongside or in lieu of ASME. Lmart designs and fabricates to both standards, providing flexibility for domestic and export supply chains.
Manufacturers like Kelvion and Alfa Laval are developing specialized liquid cooling product lines for data centers, often built around brazed plate or gasketed plate-and-frame architectures. These products excel at the rack-level and small CDU scale. Where shell-and-tube becomes the preferred specification is in high-pressure applications, large-duty CDUs (500 kW and above), and where tube-bundle cleanability or removal is an operational requirement. Lmart's differentiation is in that space.
Technical Checklist: Specifying a Data Center CDU Heat Exchanger
For engineers and procurement teams evaluating shell-and-tube heat exchangers for data center CDU applications, the following checklist captures the key specification points:
Process Design
- [ ] Define primary loop fluid (DI water, glycol-water mixture, dielectric fluid) and concentration
- [ ] Define secondary loop fluid (facility chilled water, adiabatic cooler water) and treatment chemistry
- [ ] Establish inlet/outlet temperatures for both sides (typical: server loop 40–55°C supply, 25–35°C return; facility side 15–25°C supply)
- [ ] Calculate or specify heat duty (kW) and allowable pressure drop (typically 50–100 kPa per side for CDU applications)
Materials
- [ ] Specify tube material: 316L SS minimum for DI water service; consider titanium for aggressive water chemistry
- [ ] Specify shell material: 304SS or 316L for potable/treated water; carbon steel acceptable for conditioned chilled water with inhibitor
- [ ] Specify tubesheet material: match to more aggressive of the two fluid services
- [ ] Confirm weld filler metal compatibility (308L for 304/304 joints; 316L for 316/316 joints)
Mechanical Design
- [ ] Select TEMA type: BEM (fixed tubesheet, removable channels) preferred for cleanability; BEW or AES for bundle removal
- [ ] Specify design pressure both sides (minimum 10 bar for CDU applications; confirm with facility mechanical engineer)
- [ ] Specify ASME Section VIII Div. 1 certification if required; confirm U-stamp and CRN requirements for jurisdiction
- [ ] Define nozzle orientation and size per CDU P&ID
- [ ] Specify fouling resistance: R=0.000044 for DI water; R=0.000088 for treated facility water
Quality and Documentation
- [ ] ASME U-stamp datasheet and stamping
- [ ] Material test reports (MTRs) for all pressure-boundary components
- [ ] Hydrostatic test certificate at 1.3x design pressure
- [ ] NDE reports (RT or UT on seam welds; PT on tubesheet-to-shell welds)
- [ ] Dimensional inspection report
Looking Ahead: 2026 and Beyond
The liquid cooling market is entering its second phase. The first phase (2022–2025) was driven by early adopters: GPU cloud providers, semiconductor fabs building on-premise AI clusters, and HPC research facilities. The second phase (2026–2030) will be driven by enterprise adoption — corporate data centers, financial services firms, and regional cloud operators upgrading aging facilities to handle AI inference workloads.
This broader adoption will pull demand for CDU heat exchangers further down the market, from the largest hyperscale builds toward mid-size data centers in the 5–50 MW range. This is territory where industrial heat exchanger manufacturers compete on lead time, flexibility, and technical support — not just on catalog price.
For heat exchanger manufacturers with the right certifications, materials capability, and engineering-to-order experience, data center liquid cooling represents a market expansion that is both structurally durable (AI thermal density will not reverse) and geographically distributed (China, SE Asia, and Japan are all building now).
The inflection has already happened at the chip level. The infrastructure buildout is following.
About Lmart
Lmart (苏州利玛特能源装备) designs and manufactures custom shell-and-tube heat exchangers, pressure vessels, and modular process skids. Our products carry ASME Section VIII U-stamp, PED 2014/68/EU, and ISO 9001 certification.
For data center CDU heat exchanger applications, contact our engineering team at jnlmart.net to discuss your project specifications.
Last reviewed: May 14, 2026 · Technical accuracy verified by Lmart Engineering Dept.