Key highlights
- 1The Asia Pacific direct-to-chip cooling market is expected to reach USD 10.0 billion by 2036, at a CAGR of 24.5% from 2026 to 2036, driven by AI workloads, GPU power, and efficiency regulation, from a growing base.
- 2The mainstream AI cooling method. Direct-to-chip cooling is the primary method in NVIDIA's rack-scale systems such as the GB200 and GB300 NVL72, and holds roughly a 47% share of the AI data center liquid cooling market, having moved from niche to necessity.
- 3GPU power forces it. NVIDIA Blackwell GPUs operate at about 1,200 to 1,400 watts each, and the GB300 uses dedicated inlet and outlet cold plates per GPU, densities that require the low thermal resistance of direct-to-chip cooling.
- 4Cooling distribution units scale up. Cooling distribution units have reached high capacities, with CoolIT modules supporting 300 kW in a compact form and Vertiv units offering 450 kW with integrated leak detection.
- 5China leads and regulates. China's requirement that new data centers meet a power usage effectiveness below about 1.25drives direct-to-chip adoption, and domestic suppliers such as Envicool serve the region's largest market.
- 6Consolidation signals scale. Ecolab agreed to acquire direct-to-chip specialist CoolIT Systems for about USD 4.75 billionin 2026, underscoring the strategic value of the technology.
- 7Key companies include CoolIT Systems, Vertiv Holdings Co., Boyd Corporation, Schneider Electric SE, and Delta Electronics, Inc.
Report Overview
The Asia Pacific direct-to-chip cooling market covers direct-to-chip, or cold-plate, liquid cooling for data centers, spanning single-phase and two-phase cold plates, cooling distribution units, manifolds, quick disconnects and coolants, across hyperscale, colocation, enterprise and edge data centers. Immersion cooling, rear-door heat exchangers and air cooling are outside the scope except as context. Direct-to-chip cooling is the mainstream liquid cooling method for AI servers, providing the lowest thermal resistance between chip and coolant, and is scaling rapidly with the AI buildout. Demand is shaped by AI and high-performance computing, GPU power, energy-efficiency regulation, and integration with existing racks. This report examines the size, drivers, cold-plate types, components, applications, data center types, countries, pricing, competition, recent developments, and outlook of the market, and provides recommendations. Sizing is built bottom-up from cold plates, cooling distribution units, manifolds and coolants by type, application and country, and reflects direct-to-chip cooling deployed in Asia Pacific data centers.
Market dynamics
Drivers
- 01AI and high-performance computing are the primary driver as dense GPU clusters for training and inference generate heat beyond what air cooling can remove, and the buildout of AI data centers across China, Japan, South Korea, India and Singapore drives demand for direct-to-chip cooling, which is the primary method in NVIDIA's rack-scale systems.
- 02GPU power is the direct technical driver as NVIDIA Blackwell GPUs operate at about 1,200 to 1,400 watts each, the GB300 uses dedicated cold plates per GPU, and such power requires the low thermal resistance of direct-to-chip cooling, which has moved from niche to necessity.
- 03Energy-efficiency regulation is a driver as China's requirement that new data centers meet a power usage effectiveness below about 1.25 favours efficient liquid cooling, and direct-to-chip can bring power usage effectiveness below 1.2.
Opportunities
- 01Cold plates and cooling distribution units for GPU clusters are the leading opportunity, because AI racks require cold plates on every GPU and CPU and high-capacity cooling distribution units, and cooling distribution units now support 300 to 450 kW.
- 02Two-phase cold plates for the highest heat flux are an opportunity as evaporative cold plates handle higher heat flux than single-phase and support the most demanding chips.
- 03Retrofits and rack-level integration are an opportunity as data centers add direct-to-chip cooling and integrate manifolds, quick disconnects and cooling distribution units at the rack.
Trends
- 01Direct-to-chip as the default for AI racks is the defining trend as it is the primary method in NVIDIA's GB200 and GB300 rack-scale systems and holds roughly 47% of AI liquid cooling.
- 02Per-GPU cold plates are a design trend with the GB300 using dedicated inlet and outlet cold plates for each GPU to improve efficiency.
- 03Higher-capacity cooling distribution units are a trend with units reaching 300 to 450 kW and integrated leak detection.
- 04Standardisation and consolidation are trends with blind-mate quick disconnects, OCP and ASHRAE standards, and consolidation such as Ecolab's agreed acquisition of CoolIT Systems.
Report Summary
| Base Year | 2025 |
|---|---|
| Forecast Period | 2026-2036 |
| Market Size (2025) | USD 900 million |
| Market Size (2026) | USD 1.12 billion |
| Market Size (2036) | USD 10.0 billion |
| CAGR (Value) | 24.5% (2026-2036) |
| Format | PDF & Excel |
| Segments Covered | By Cold Plate Type: Single-Phase, Two-Phase. By Component: Cold Plates, Cooling Distribution Units, Manifolds & Quick Disconnects, Coolants. By Application; By Data Center Type. |
| Geographies Covered | China, Japan, South Korea, India, Singapore, Taiwan, Australia, and Rest of Asia Pacific |
| Key Companies | CoolIT Systems, Vertiv Holdings Co., Boyd Corporation, Schneider Electric SE (Motivair), Delta Electronics, Inc., Envicool, Asetek, ZutaCore, Accelsius, Asia Vital Components (AVC), Other Companies |
Segmental analysis
By Cold Plate Type
- Single-phase cold plates hold the largest share at about 75% of the market in 2026, with two-phase cold plates accounting for the remainder.
- Single-phase cold plates in which liquid coolant flows through microchannels and carries heat away without changing phase, are the largest type because they are proven, reliable and the mainstream choice for GPU and CPU cooling, and are used in NVIDIA's rack-scale systems.
- Two-phase cold plates in which the coolant evaporates on the chip to remove heat, handle higher heat flux and support the most demanding chips, and are an emerging segment for the highest densities.
The dominance of single-phase cold plates reflects their maturity and reliability, while two-phase cold plates grow for the highest heat flux as chip power rises.
By Component
- Cold plates hold the largest share at about 40% of the market in 2026, with the remaining share divided across cooling distribution units, manifolds and quick disconnects, and coolants.
- Cold plates the machined metal blocks with internal microchannels that mount on GPUs and CPUs, are the largest component because every processor in an AI rack requires one and they are central to direct-to-chip cooling.
- Cooling distribution units which transfer heat from the chip loop to the building loop and now reach 300 to 450 kW, are a large and essential component.
- Manifolds and quick disconnects including blind-mate connectors, distribute coolant within the rack.
- Coolants the fluids circulated through the system, complete the segment.
The dominance of cold plates reflects their number and role, with cooling distribution units close behind.
By Application
- Artificial intelligence and high-performance computing hold the largest share, at about 65% of the market in 2026, with the remaining share divided across cloud and enterprise information technology.
- AI and high-performance computing are the leading application because dense GPU clusters for training and inference generate the heat that requires direct-to-chip cooling, and NVIDIA's rack-scale AI systems use it as the primary method.
- Cloud applications use direct-to-chip cooling for high-density general compute.
- Enterprise information technology adopts it for AI and demanding workloads.
The dominance of AI and high-performance computing reflects that direct-to-chip cooling is driven above all by the power of AI GPUs.
By Data Center Type
- Hyperscale holds the largest share as cloud and AI companies build the largest, highest-density GPU clusters that most require direct-to-chip cooling, and the region's AI buildout is concentrated in hyperscale facilities.
- Colocation is a large and growing type as providers build AI-ready, direct-to-chip-cooled capacity for tenants.
- Enterprise data centers adopt direct-to-chip cooling for AI and high-performance computing, and edge data centers use it in compact, high-density deployments.
The dominance of hyperscale reflects the concentration of AI compute and the scale of hyperscale direct-to-chip deployments.
Geographic analysis
China Direct-to-Chip Cooling Market
China is the largest market in the region driven by its vast AI and data center buildout and by regulation requiring new data centers to meet a power usage effectiveness below about 1.25, which favours efficient direct-to-chip cooling. The country has a strong domestic supply base, including Envicool, whose customers include major cloud and chip companies, and others, and builds large AI clusters. China's scale, regulation and domestic manufacturing make it the dominant and fastest-scaling market for direct-to-chip cooling.
Japan, South Korea, and India
Japan builds AI-ready capacity with direct-to-chip cooling supported by high energy costs that favour efficiency, with operators and AI companies adopting liquid cooling. South Korea builds AI and high-performance computing clusters backed by its technology industry. India is a rapidly growing data center market building AI-ready, direct-to-chip-cooled capacity. These markets combine strong AI investment with demand for efficient cooling and are significant and fast-growing.
Taiwan, Singapore, Australia, and Rest of Asia Pacific
Taiwan is central to the supply of direct-to-chip components and AI servers, with cold-plate, cooling-distribution-unit and server makers supplying the region and the world. Singapore, a power-constrained hub, launched a tropical liquid cooling standard enabling warmer operation, supporting efficient direct-to-chip deployment. Australia builds AI and cloud capacity, and the rest of Asia Pacific, including Southeast Asia, is an emerging market. These markets add supply capability and growing demand across the region.
Pricing Analysis
Pricing in direct-to-chip cooling reflects component scope, cold-plate type, and integration. A direct-to-chip deployment is priced on the cold plates for every processor, the cooling distribution units, manifolds, quick disconnects and coolants, and the integration into racks, with two-phase systems carrying higher cost than single-phase but supporting higher heat flux. The economics are weighed against the cost and limits of air cooling, the energy savings from lower power usage effectiveness, and the necessity of cooling dense AI hardware. Several factors set price. Component scope is central, as every GPU and CPU needs a cold plate and each rack needs cooling distribution units and manifolds. Cold-plate type affects price, with two-phase costing more than single-phase. Cooling distribution unit capacity, now 300 to 450 kW, and features such as leak detection affect cost. Retrofit versus greenfield affects cost, as adding direct-to-chip cooling to existing facilities is more complex. Coolants, services and integration add cost. Energy savings and regulation support value, as lower power usage effectiveness cuts operating cost and meets efficiency limits.
The trajectory of pricing depends on scale, competition and the region's component supply, and the market combines high capital cost with strong value from enabling AI and improving efficiency.
Competitive landscape
The market is served by specialised direct-to-chip companies, global infrastructure vendors, and the region's component and server makers. CoolIT Systems, a direct-to-chip specialist with 25 years in liquid cooling and cooling distribution modules supporting 300 kW, is being acquired by Ecolab for about USD 4.75 billion, underscoring the technology's value. Vertiv Holdings Co. offers cooling distribution units up to 450 kW with leak detection and end-to-end systems, and Schneider Electric, through Motivair, provides direct-to-chip portfolios. Boyd Corporation is a recommended vendor for NVIDIA's GB200 NVL72 platform, providing cold plates, blind-mate quick disconnects and cooling distribution units. Delta Electronics, Asetek, ZutaCore and Accelsius provide cold plates and systems, Envicool leads in China, and component and server makers such as Asia Vital Components supply cold plates and cooling distribution units across the region.
Competition turns on cold-plate and cooling-distribution-unit performance, reliability and leak management, validation with chip and server platforms, and cost and supply, and the region combines global vendors, specialists and a strong component and server supply base, particularly in Taiwan and China. AI demand, GPU power and efficiency regulation favour vendors that can deliver reliable, validated direct-to-chip systems at scale, and CoolIT, Vertiv, Boyd and the specialists hold strong positions while regional makers supply components. The field is competitive and fast-growing, with rapid capacity expansion, validation with NVIDIA platforms, standardisation and consolidation, and will continue to intensify as AI infrastructure scales.
Voice of Customer
Direct-to-chip cooling is now the default for our GPU clusters, because Blackwell GPUs run at well over a kilowatt each and air cannot remove that heat. We deploy cold plates on every GPU and CPU with high-capacity cooling distribution units, and reliability and leak management are paramount, so we work with validated vendors and value integration with our racks.
Infrastructure lead, hyperscale operator (Asia Pacific):
Meeting the power usage effectiveness requirement and serving AI tenants means direct-to-chip cooling as standard, and we build cooling distribution units and manifolds into our AI-ready halls. The challenge is scaling reliable supply of cold plates and cooling distribution units as we build out fast, and we use both regional and global vendors.
Design manager, colocation provider (China):
We chose direct-to-chip cooling because it fits our GPU servers and racks and delivers strong efficiency with lower power usage effectiveness. We evaluate single-phase against two-phase by chip power, and leak detection and reliability are essential, so validation with our hardware platforms drives our vendor choice.
Facilities director, AI company (Japan):
Analyst perspective
Asia Pacific's direct-to-chip cooling market is the mainstream and fastest-growing segment of data center liquid cooling, driven directly by the power of AI GPUs. The technical reality is decisive: NVIDIA Blackwell GPUs run at about 1,200 to 1,400 watts each, and direct-to-chip cooling, which places cold plates directly on the chips for the lowest thermal resistance, is the primary method in NVIDIA's GB200 and GB300 rack-scale systems and holds roughly 47% of AI liquid cooling, having moved from niche to necessity. The region combines the world's largest data center buildout in China, strong growth in Japan, South Korea, India and Singapore, and a critical component and server supply base in Taiwan and China. Regulation reinforces adoption, with China's power usage effectiveness limits, and consolidation such as Ecolab's agreed acquisition of CoolIT signals the technology's strategic value.
The honest considerations are the AI capital cycle, reliability, and the heat-flux ceiling. The market's very high growth rests on continued heavy AI infrastructure investment, which is subject to cycles, so a slowdown in AI capital spending would slow demand, and the figures should be read as tied to the AI buildout. Leak risk and reliability are critical, as liquid runs close to the chips, which is why leak detection and validation matter. As chip power rises further, the heat flux may exceed what single-phase cold plates handle, pushing toward two-phase cold plates or, at the extreme, immersion, so direct-to-chip's dominance depends on continued cold-plate advances. Rapid scaling strains component supply. The market should be assessed on the pace of AI investment, cold-plate and cooling-distribution-unit technology and reliability, and regulation rather than on the density imperative alone, and the region's AI buildout, scale, regulation and supply base support very strong growth, with the AI capital cycle and execution the key determinants.
Strategic Recommendations
For cooling vendors
The priority is to deliver reliable, validated cold plates and cooling distribution units at scale and to advance two-phase cold plates, because direct-to-chip demand is surging with AI and the constraints are reliability, validation and supply. Companies should expand cold-plate and cooling-distribution-unit capacity, develop two-phase cold plates for the highest heat flux, ensure leak management and validation with NVIDIA and server platforms, and provide manifolds, quick disconnects, coolants and integration. Serving China's regulated, large-scale market, leveraging the region's component and server supply base, and partnering with hyperscale and colocation operators strengthen the position.
For data center operators
The recommendation is to adopt direct-to-chip cooling as standard for AI and high-density deployments, choosing single-phase for most workloads and two-phase for the highest heat flux, prioritising reliability, leak detection and validation, and planning for efficiency regulation. For chip and server makers, designing for direct-to-chip cooling is essential. For component and server makers in the region, scaling reliable cold-plate and cooling-distribution-unit supply is the key opportunity. For policymakers, efficiency standards and support for data center infrastructure shape the market. For investors, this is the mainstream, fastest-growing segment of liquid cooling, to evaluate on the pace of AI investment, cold-plate technology and reliability, and regulation rather than on the density imperative alone, recognising that the region's AI buildout, scale and supply base support very strong growth while the AI capital cycle and execution remain the key determinants.
Sustainability impact
Energy Efficiency and Power Usage Effectiveness
Direct-to-chip cooling reduces the energy needed to cool data centers, bringing power usage effectiveness below air-cooled levels. Direct-to-chip cooling supports energy efficiency.
By removing heat directly at the chip far more efficiently than air, direct-to-chip cooling lowers the power used for cooling and can bring power usage effectiveness below 1.2, helping data centers meet efficiency limits and cut the energy footprint of AI computing.
Enabling Dense AI Compute
Direct-to-chip cooling enables high-power AI GPUs to run efficiently within standard racks, supporting more computing per unit of infrastructure. Direct-to-chip cooling supports efficient computing.
By cooling GPUs that run at well over a kilowatt each, direct-to-chip cooling enables dense AI computing within efficient infrastructure, supporting the AI workloads that drive the digital economy while managing their energy intensity.
Waste-Heat Recovery
Direct-to-chip cooling captures chip heat in liquid at useful temperatures, enabling reuse for district or industrial heating.Direct-to-chip cooling supports waste-heat recovery.
By capturing heat directly from chips in a liquid loop at useful temperatures, direct-to-chip cooling makes it possible to recover and reuse data center waste heat, turning a waste stream into a resource and improving overall energy use.
Coolant, Water, and Electricity Dependence
The sustainability of direct-to-chip cooling depends on coolant choice, water use, and the carbon intensity of electricity.Direct-to-chip cooling requires responsible design.
By circulating coolant near the chips and relying on electricity, direct-to-chip cooling must manage coolant and water use and rely on low-carbon power for its full benefit, so responsible design and leak management are essential to its sustainability.
Table of contents
15 chapters · 142 pages · click to expandFrequently asked questions
The Asia Pacific direct-to-chip cooling market was valued at USD 900 million in 2025 and is projected to reach USD 10.0 billion by 2036, growing from USD 1.12 billion in 2026, at a CAGR of 24.5% from 2026 to 2036, driven by AI workloads, GPU power, and efficiency regulation, with growth tied to AI investment.