
- Microsoft, Google, Meta and Amazon’s hyperscale data centre buildout — with over $200 billion in combined announced CAPEX for 2025–2027 — is creating one of the fastest-growing segments in the industrial valve market, as liquid cooling replaces air cooling as the dominant thermal management technology.
- Direct liquid cooling (DLC) and immersion cooling systems require a completely different valve specification from conventional HVAC — operating at higher pressures, with dielectric fluids or deionised water, and with reliability requirements calibrated to the commercial cost of server downtime rather than industrial process interruption.
- The AI computing revolution is driving unprecedented power densities in server racks — from 5–10 kW per rack in conventional data centres to 50–100 kW per rack for AI training clusters — making effective liquid cooling not just advantageous but physically necessary, and transforming the cooling infrastructure valve market in the process.
The Data Centre Cooling Revolution and Its Valve Engineering Implications
Data centres are the infrastructure of the modern economy — and they are being built at a pace, scale and density that was unimaginable even five years ago. The artificial intelligence computing revolution, the continued growth of cloud services, and the streaming and gaming infrastructure demands of a digitally dependent population are combining to drive a capital investment wave in data centre construction that is reshaping the global industrial infrastructure landscape. And at the heart of this revolution is a thermal management challenge that is fundamentally changing the valve and flow control requirements of data centre engineering.
Traditional data centres cooled their servers with forced air — computer room air conditioning (CRAC) units that pushed chilled air through raised floors or overhead ducts, across rack-mounted servers, and returned warm air to the cooling units for recirculation. This approach worked adequately when server power densities were modest (5–15 kW per rack), but it becomes physically unworkable as AI training servers push rack densities to 50, 75 or 100 kW. You simply cannot move enough air through a rack at those power densities to remove the heat effectively — the air would need to be flowing at velocities that would destroy the electronics it’s trying to cool.
The solution — liquid cooling in various forms — has been understood by the computing industry for decades but has been slow to achieve mainstream adoption due to the complexity and cost premium versus air cooling. That calculation has now fundamentally changed, and the liquid cooling market is experiencing the fastest growth in its history.
Microsoft’s AI Data Centre Programme: The Valve Procurement Scale
Microsoft has committed to $80 billion in data centre investment in its fiscal year 2025 alone, with a substantial proportion targeted at AI-capable infrastructure. The company’s Azure cloud platform serves as the foundation for its OpenAI partnership, and the computational demands of large language model training and inference are driving an aggressive programme of facility construction and upgrade across the US, Europe and Asia.
Microsoft’s approach to data centre cooling has evolved significantly in recent years. The company’s Project Natick — a research initiative exploring underwater subsea data centres — demonstrated the thermal advantages of seawater cooling but was not pursued at commercial scale. The production approach focuses on direct liquid cooling (DLC) for high-density AI compute racks, using server-level cold plates bonded directly to CPU and GPU heat spreaders, with chilled water or a suitable coolant fluid circulated through the cold plates at temperatures of 25–45°C.
The valve requirements for a DLC cooling system in a hyperscale data centre are demanding in ways that differ from conventional industrial applications. The primary coolant circuit — delivering chilled water from the building’s central cooling plant to the row-level distribution manifolds — operates at moderate pressures (typically 6–10 bar) with deionised water or a glycol-water mixture. Valves in this circuit must be compatible with deionised water, which is extremely aggressive toward common copper alloys due to its low ionic content and high copper-dissolving capability. Stainless steel or plastic-lined valves are standard for primary DI water circuits.
At the rack level, distribution manifolds with individual row isolation valves provide the granularity of control needed to service individual racks without interrupting the entire row’s cooling. These manifold valves — typically 1/2-inch to 1-inch ball valves or needle valves with position indicators — need to maintain reliable function despite the physical access constraints of operational data centre rack aisles and the expectation that any valve work will be performed without disruption to adjacent operational equipment.
Google’s Data Centre Cooling: Setting the Industry Standard
Google has been at the forefront of data centre efficiency innovation since its earliest days, and its cooling infrastructure design philosophy has heavily influenced industry practice. Google’s data centres routinely achieve Power Usage Effectiveness (PUE) values of 1.1 or below — meaning 90% or more of facility energy input is used for computing, with only 10% consumed by cooling and other facility services. Achieving these PUE values requires extremely efficient cooling infrastructure, including sophisticated valve control systems that continuously optimise coolant flow distribution in response to real-time server load variations.
Google’s cooling valve infrastructure uses automated pressure-independent control valves (PICVs) — valves that combine flow-limiting and flow-balancing functions in a single body, maintaining constant flow rate regardless of system pressure variations. PICVs from manufacturers including Belimo, Siemens and Danfoss are widely used in hyperscale data centre cooling distribution, providing the hydraulic stability needed for effective temperature control in systems with thousands of parallel cooling circuits.
Immersion Cooling: The Next Frontier and Its Radical Valve Implications
Single-phase immersion cooling — where servers are submerged in a thermally conductive, electrically insulating dielectric fluid (typically engineered fluids from 3M‘s Fluorinert range or synthetic hydrocarbon fluids from Shell‘s Diala range and specialist manufacturers including Submer and LiquidCool Solutions) — is gaining commercial traction for the highest-density AI compute applications. The thermal advantages are substantial: immersion cooling can remove heat directly from chip surfaces without the thermal resistance of a cold plate interface, enabling higher sustained computing power from a given rack footprint.
The valve requirements for immersion cooling systems are quite different from DLC. The dielectric fluid — which has specific compatibility requirements with elastomers and sealants used in valve construction — operates in a sealed tank system where leak detection and prevention is paramount. A dielectric fluid leak in a data centre is not just an operational problem (loss of expensive coolant) but a potential safety issue (risk of fluid contact with electrical systems) and an environmental problem (engineered fluorochemical fluids are persistent in the environment). Valve selection for immersion cooling systems prioritises zero-leakage performance: typically all-stainless or PVDF body construction with metal-seated or PTFE-seated designs that avoid the elastomers that dielectric fluids can degrade.
Amazon, Meta and the Continued Hyperscale Buildout
Amazon Web Services (AWS) and Meta are executing data centre construction programmes of comparable scale to Microsoft and Google. AWS’s commitment to expanding its global data centre footprint — with major investments in Northern Virginia (the world’s largest data centre concentration), Dublin, Singapore, Sydney and a growing list of emerging markets — involves liquid cooling upgrades across its custom-designed server generations as well as new facility construction designed around DLC from the outset.
Meta’s data centres, which serve Instagram, WhatsApp, Facebook and the compute infrastructure for its AI research programmes, have adopted an open-source infrastructure philosophy through the Open Compute Project (OCP). Meta’s OCP-published cooling specifications — which define the interfaces and performance requirements for liquid cooling infrastructure across its facilities — have become an informal industry standard that influences valve and cooling component specifications across the hyperscale sector.
The data centre valve market is a genuinely exciting emerging opportunity for valve engineers and suppliers who understand that the technical requirements — while different from traditional industrial applications — are no less demanding. The companies building and operating these facilities are sophisticated buyers who understand the commercial cost of reliability failures and who will invest appropriately in high-quality, well-specified valve infrastructure. For the valve industry, the AI computing revolution is not an abstract technological development — it is a growing and highly valuable segment of the flow control market that will be worth billions of dollars annually within this decade.