NVIDIA, Dell Technologies, Supermicro, Hewlett Packard Enterprise, Vertiv, and Schneider Electric are each shipping AI-ready liquid-cooled platforms
Decision Focus
According to a July 2026 Precedence Research forecast, the global data center liquid cooling manifolds market reached USD 1.05 billion in 2025 and is projected to reach approximately USD 15.99 billion by 2035, at a compound annual growth rate of 31.3%. Direct-to-chip cooling held 58% of the technology mix in 2025, with supply manifolds leading by configuration at 38% market share. For Global Heads of Data Center Energy, the operational signal is not the hardware spend — it is what accelerating liquid cooling penetration does to the power infrastructure equation across a multi-GW portfolio.
90-Second Brief
Now, a decade-long, 15x expansion in liquid cooling infrastructure is underway, driven by AI workload density that conventional air cooling cannot reliably sustain. Direct-to-chip and immersion configurations are being deployed at hyperscale pace: AWS built and installed a custom liquid cooling system in under 11 months for its next-generation AI infrastructure. NVIDIA, Dell Technologies, Supermicro, Hewlett Packard Enterprise, Vertiv, and Schneider Electric are each shipping AI-ready liquid-cooled platforms. The energy leadership implication is direct, as liquid cooling penetrates the installed base, the ratio of IT power to total facility power changes, altering how demand is modeled, how electrical infrastructure is sized, and what volumes you commit to in a long-term PPA.
What Is Really Happening?
AI GPU clusters are running at thermal loads that air-cooled infrastructure cannot reliably sustain. The gap is not marginal and is widening with each GPU generation. Liquid cooling manifolds solve the thermal problem at the rack level, but they do so by changing the facility’s power topology.
When direct-to-chip cooling replaces mechanical air cooling, the compressor-heavy chiller load drops substantially. A facility that previously required significant electrical overhead for cooling plant operation begins to carry a structurally different load profile. The AI and HPC segment held 46% of the liquid cooling manifolds market in 2025 and is forecast at 36.8% annual growth — meaning the transition is dominant and accelerating, not marginal.
North America held 41% of the 2025 market, anchoring the shift in the highest-density data center geographies. Asia-Pacific, where semiconductor manufacturing and AI compute investment are concentrated, is projected at 35.7% CAGR through 2035 — the fastest regional rate — introducing new infrastructure configuration questions for operators with global footprints.
Why It Matters for Global Heads of Data Center Energy
The energy procurement model most operators built assumes a stable relationship between IT load and total facility power draw. That relationship is being disrupted. As liquid cooling penetrates the fleet, PUE baselines fall and the mechanical overhead that historically shaped procurement volumes becomes a less reliable guide to forward demand.
This creates specific risk in long-term PPAs. A 10- or 15-year agreement sized against a chiller-heavy load profile may be structurally oversized if a facility retrofits to liquid cooling at scale mid-contract. The inverse risk is equally real: if liquid cooling enables substantially higher rack densities within the same physical footprint, total IT power per site could increase enough to strain existing grid capacity commitments rather than relieve them.
Substation and transformer sizing decisions face the same uncertainty. AWS deploying a full custom liquid cooling system in under 11 months signals that deployment cycles are compressing below typical interconnection or transformer procurement lead times. An energy leader working against a three-year transformer lead time is effectively planning against a cooling technology mix that may have already shifted by equipment delivery.
For colocation operators — who held 26% of the manifolds market in 2025 — the pressure differs but is equally concrete. Multi-tenant facilities cannot enforce cooling uniformity across tenants, yet hyperscale customers are increasingly specifying liquid-cooled configurations. The result is a mixed load profile that complicates shared infrastructure sizing and energy cost allocation across the lease term.
Forward View
If the 31.3% CAGR holds through the late 2020s, liquid cooling moves from hyperscale-specific to standard configuration in new builds. Three dynamics follow.
First, cooling overhead as a share of total facility power declines fleet-wide. Energy demand models anchored to historical PUE benchmarks will need recalibration against site-specific cooling configurations rather than portfolio averages — changing how procurement volumes are justified at board level.
Second, the Asia-Pacific growth trajectory means that emerging data center markets in Southeast Asia and India are likely to build with liquid-cooled assumptions from the outset. Operators entering those markets now are setting energy infrastructure baselines that will govern procurement commitments for a decade.
Third, immersion cooling — currently 14% of the market but projected at 39.8% CAGR — represents a more disruptive scenario. Full-immersion deployment eliminates conventional cooling plant requirements almost entirely, which would fundamentally reshape the electrical infrastructure profile of an affected facility. The timeline is longer and less certain, but the scenario belongs in 10-year planning models now.
What Is Still Uncertain
The Precedence Research forecast is a single-vendor estimate without published methodology or independent validation. The 31.3% CAGR assumes sustained AI infrastructure capital expenditure at current rates; any moderation from hyperscalers would compress the trajectory meaningfully.
The energy efficiency impact at portfolio level also depends on deployment depth in ways the market data does not resolve. A facility that liquid-cools 20% of its racks achieves modest PUE gains; full retrofit achieves a different outcome entirely. Neither the market figures nor the vendor announcements quantify what proportion of installed capacity is being fully converted versus partially upgraded.
Finally, heat reuse from liquid-cooled facilities — whether captured for district heating or industrial processes — carries regulatory and contractual implications in several European jurisdictions that remain largely unquantified for global operators.
One Question for Your Team
If our highest-density sites complete a full transition to direct-to-chip liquid cooling over the next five years, what does that do to the total facility load assumptions embedded in our current PPA commitments — and do we have contractual flexibility to respond if the power profile shifts materially in either direction?
Sources
- Precedenceresearch — Data Center Liquid Cooling Manifolds Market Size to Hit USD 15.99 Billion by 2035 (Link)
