PUBLISHER: Mordor Intelligence | PRODUCT CODE: 2063861
PUBLISHER: Mordor Intelligence | PRODUCT CODE: 2063861
According to Mordor Intelligence, the asia pacific gPU immersion cooling market size is expected to grow from USD 0.44 billion in 2025 to USD 0.55 billion in 2026 and is forecast to reach USD 2.07 billion by 2031 at a 30.38% CAGR over 2026-2031.

This report is Segmented by Immersion Type (Single-Phase Immersion Cooling, and Two-Phase Immersion Cooling), Solution Type (Immersion Cooling Tanks/Systems, Dielectric Fluids, and More), Deployment (Hyperscale and Cloud, Enterprise, and Government and Research HPC), GPU Power Density (Below 300W, 300W-700W, and More), and Region. The Market Forecasts are Provided in Terms of Value (USD).
Hyperscalers are acquiring record volumes of accelerators, pushing rack power beyond 100 kW and forcing a rapid pivot to immersion solutions. ByteDance committed USD 14 billion for GPU procurement in 2025, while Alibaba Cloud earmarked USD 100 billion for AI infrastructure, both sizing their halls around 700 W-class devices. Naver's 4,000-unit NVIDIA B200 roll-out and SK Telecom's 1,000-unit deployment show that sustained operation at nameplate performance requires liquid capture of chip heat. Japan's USD 12 billion GMI Cloud campus explicitly mandates immersion tanks for Blackwell-generation clusters. Because NVIDIA roadmaps indicate devices surpassing 1,000 W TDP within two product cycles, immersion is no longer a niche efficiency play, it is the thermal pre-requisite for competitive AI training throughput.
Governments are weaponizing PUE caps to curb data-center electricity demand. Japan requires existing sites to hit 1.4 by 2030 and new builds to meet 1.3 from 2029, backed by subsidies up to JPY 114.6 billion (USD 1.0 billion) for immersion projects. Singapore limits new facilities to PUE more than 1.3 and is drafting a dedicated liquid-cooling code for 2026. Beijing, Shanghai, and Guangdong already enforce thresholds between 1.2 and 1.3. Demonstrated immersion fields routinely deliver PUE near 1.02, eliminating CRAH fans and ductwork while slashing cooling power by up to 90%. The policy trajectory suggests air cooling will be non-compliant across tier-1 metros well before the end of the decade.
Total cost of ownership favors immersion above 50 kW per rack, yet many enterprise budgets still focus on initial price tags. Indian operators pay 10-20% more at commissioning because tanks, pumps, and imported fluids inflate procurement outlays. Fluid alone can add USD 6,000-60,000 per rack, depending on formulation and tank volume. Although Shell and Keppel showed 40% life-cycle savings in Singapore, enterprises operating on 18-month payback horizons hesitate to approve the switch. Reliance on imported components further widens the gap through tariffs and logistics, particularly in South Asia.
Other drivers and restraints analyzed in the detailed report include:
For complete list of drivers and restraints, kindly check the Table Of Contents.
Single-phase designs held 79.45% of Asia Pacific GPU immersion cooling market share in 2025. They remain popular because mineral-oil and synthetic-ester fluids let operators submerge standard servers without sealed lids, easing maintenance and retrofit projects. Asperitas attracted USD 55.5 million in Series C funding and began rolling out such systems in Thailand's tropical climate. Yet next-wave GPUs topping 700 W are eroding the headroom of convection-only loops. Two-phase solutions, which vaporize low-boiling-point fluids for superior heat flux, are scaling as hyperscalers chase rack densities above 150 kW. Sabey Data Centers is integrating OptiCool refrigerant loops portfolio-wide, bypassing chilled water entirely.
Looking ahead, NVIDIA's Blackwell parts are expected to push per-device TDP near 1,000 W, at which point two-phase will mainstream among cloud builders. SLiquid's C8000 cabinet already claims 200 W/cm2 dissipation and 220 kW loads. Nevertheless, single-phase will not disappear; enterprises with <700 W accelerators still favor its lower complexity, and edge nodes prize direct technician access to submerged boards. The Asia Pacific GPU immersion cooling market therefore bifurcates: convection loops dominate mid-power deployments, while phase-change tanks anchor ultra-dense AI training farms.
Tanks and ancillary systems represented 55.34% of Asia Pacific GPU immersion cooling market size in 2025 because they are the starting point for any liquid build. Green Revolution Cooling's CarnotJet units, for example, are now shipping under a Samsung C&T alliance across multiple APAC campuses. OEMs, however, are unveiling boards purpose-engineered for submerged use. Fujitsu, Supermicro, and Nidec co-developed fan-less chassis that shed heatsinks and blowers, cutting bill of materials by up to 15%. Wiwynn invested in LiquidStack manufacturing lines to embed immersion readiness at design time.
Such integrated offerings eliminate warranty disputes about fluid compatibility and support higher GPU counts per node, translating to better performance per rack. As a result, immersion-optimized servers are projected to outgrow tanks at a 30.85% CAGR. Tanks will persist as the delivery vehicle, but value capture migrates up-stack toward pre-validated compute modules that slash integration risk for hyperscale and enterprise buyers.