AI heatwave pushes data centers to liquid cooling

The global liquid cooling market for data centers is surging, projected to leap from USD 870 million in 2024 to an impressive USD 10.7 billion by 2030, fueled by a staggering CAGR of 51.93%. This explosive growth is propelled by escalating rack power densities driven by AI and HPC workloads, alongside sustainability imperatives that prioritize energy-efficient thermal management. North America, led by the US, and China stand at the forefront of adoption, especially in greenfield hyperscale data centers, while regions with challenging climates and high electricity costs such as the Middle East and Latin America are increasingly investing in liquid cooling to capitalize on its energy-saving benefits.

Hyperscale data centers dominate the liquid cooling investment landscape, accounting for the majority share due to their intense computing and cooling demands, with colocation facilities capturing roughly 36% of the market. Enterprise data centers, however, remain cautious early adopters, hindered by high upfront costs and supply chain disruptions. Direct liquid cooling leads investment focus, aligning with global sustainability goals, while immersion cooling, though smaller at around 17%, is gaining traction to support AI-ready infrastructures. This dynamic underscores a market balancing rapid technological advancement with operational and financial challenges.

Regionally, Asia-Pacific is experiencing the fastest growth, exemplified by Japan’s data center market expanding at a CAGR of 20.42% through 2031, driven by hyperscale expansions in Tokyo and emerging hubs like Hokkaido and Kyushu. Pilot projects by Preferred Networks, IIJ, and JAIST in liquid-cooled AI servers highlight Japan’s push toward ultra-efficient AI infrastructure. Meanwhile, Southeast Asia’s powerhouse Singapore maintains its lead despite capacity constraints, with hyperscale giants AWS and Microsoft investing billions to expand cloud infrastructure. Sustainability requirements in Singapore are pushing development into neighboring Malaysia and Indonesia, where rapid AI-driven power density increases demand innovative cooling strategies amid challenges like green power availability and skilled talent shortages.

Europe’s data center market is accelerating, propelled by rising AI and HPC workloads that necessitate advanced liquid cooling solutions such as direct-to-chip and immersion cooling. Western Europe, particularly the UK and Nordic countries, leads in power capacity and investment, supported by government policies and major players like AWS, Google, and Spode Works Regeneration Ltd. Meanwhile, emerging European markets like Spain and Poland are attracting attention due to strategic locations and cost advantages. Despite geopolitical risks in the Middle East disrupting hyperscale operations, global investment flows remain robust, with Latin America also emerging as a significant growth region, expected to attract $31 billion between 2026 and 2031, led by Brazil, Chile, and Mexico.

By early 2026, traditional air cooling methods have become fundamentally inadequate for the soaring power densities of AI workloads, which now range from 50 to over 130 kW per rack—far surpassing the 10–20 kW design limits of conventional systems. These dense GPU clusters generate highly concentrated and volatile thermal profiles, producing localized hot spots and bursty spikes that overwhelm standard air distribution and containment, making liquid cooling or hybrid architectures essential for effective heat management.

The physical and economic constraints of air cooling are pushing data centers to a strategic inflection point where air is transitioning from a primary to a secondary cooling solution. Air’s low heat capacity and reliance on cooling entire data halls lead to disproportionate energy consumption and inefficiency, with practical limits reached around 30–40 kW per rack. Hyperscale operators like those deploying NVIDIA’s H100 and Grace Hopper chips, which have thermal design powers in the multi-kilowatt range, are increasingly adopting liquid cooling infrastructure to avoid excessive PUE and operational costs.

Liquid cooling offers roughly four times the thermal conductivity of air, enabling data centers to manage heat densities that air cooling cannot handle—such as NVIDIA’s GB200 NVL72 racks approaching 140 kW. This shift requires comprehensive redesigns of data center infrastructure including floor loading, piping, and leak detection, making retrofitting legacy air-cooled facilities prohibitively expensive and often resulting in stranded assets. Consequently, new builds increasingly integrate liquid cooling from the outset, fueling a market projected to grow from $5.5 billion in 2025 to $25.8 billion by 2035, driven primarily by AI infrastructure demands.

The traditional data center underwriting model, which assumes cooling as a fixed operational cost over a 20-30 year asset life, is misaligned with the rapid hardware refresh cycles of 12 to 18 months driven by AI workload evolution. As chip flux densities rise, liquid cooling becomes essential even at densities as low as 40 kW per rack, with manufacturers recommending liquid solutions for next-generation chips to ensure efficient thermal exchange. This paradigm shift underscores the necessity of planning liquid cooling early to sustain high-performance AI workloads and avoid costly retrofits.

The surge in AI chip thermal design power (TDP), exemplified by NVIDIA’s GB200 and GB300 chips reaching up to 2,700W, has outpaced the cooling limits of traditional air systems capped around 700W, making liquid cooling indispensable for modern AI data centers. This necessity has driven a paradigm shift towards high-temperature liquid cooling, with NVIDIA CEO Jensen Huang highlighting that 45 °C warm water suffices to cool chips operating above 80 °C, enabling supercomputer-scale cooling without energy-intensive chillers. Such warm water liquid cooling systems employ dual-loop architectures that transfer heat from server components to external cooling towers, facilitating free cooling and significantly enhancing energy efficiency.

As AI workloads push rack densities beyond 40 kW, liquid cooling technologies like direct-to-chip loops and immersion cooling have become essential, with two-phase immersion cooling emerging as the go-to solution at ultra-high densities exceeding 600 kW, as seen in platforms like Vera Rubin Ultra. This evolution demands entirely new data center designs since retrofitting air-cooled facilities is impractical due to fundamental infrastructure changes including floor loading, piping, and electrical distribution. Industry leaders such as BDC have pioneered direct-to-chip cooling since 2022, currently operating 90 megawatts of such capacity alongside immersion cooling deployments, underscoring the rapid maturation and scaling of these technologies.

Full-immersion cooling is gaining traction as a highly efficient and operationally advantageous liquid cooling method, demonstrated by Ready Server’s customized pilot in Singapore that submerges entire servers in non-conductive fluids to achieve superior heat transfer and higher rack densities. This approach reduces maintenance complexity, lowers coolant leak risks compared to direct-to-chip methods, and supports continuous 24/7 operation, while requiring hardware modifications and strategic risk management such as selecting servers nearing warranty end. The operational insights from such pilots are informing future rack designs optimized for compute-intensive AI workloads, positioning providers to meet escalating demand for AI-ready infrastructure.

The liquid cooling ecosystem is rapidly expanding beyond GPUs and CPUs to encompass critical rack components like network interface cards, busbars, and optical modules, driven by AI chipsets with TDPs exceeding 1 kW and rack-scale power consumption in the hundreds of kilowatts. Companies such as NVIDIA, AMD, and Google are spearheading this transition, with Google deploying liquid cooling in over 80% of its AI servers and AMD’s Helios platform fully embracing liquid cooling at rack scale. Innovations in materials and system design, including NVIDIA’s move to a one-piece heat spreader to address substrate warpage, and the integration of advanced semiconductor thermal management technologies like Samsung’s zHBM, are collectively enhancing cooling efficiency, reliability, and scalability, thereby cementing liquid cooling as the backbone of next-generation AI data centers.

The rapid escalation of AI workloads in regions like Indonesia has dramatically increased rack power densities from traditional 3 kW levels to upwards of 100 kW, creating urgent demands for new cooling architectures and operational expertise. However, this surge is compounded by a critical shortage of skilled personnel capable of managing multiple projects and ongoing operations simultaneously, forcing operators to rely on cross-border talent transfers within APAC to bridge the gap and sustain growth.

Retrofitting existing air-cooled data centers to support high-density liquid cooling is often impractical and cost-prohibitive, requiring comprehensive redesigns of structural elements such as floor loading, piping, leak detection, and electrical distribution. With retrofit costs reaching two to three million dollars per megawatt, legacy facilities face the risk of becoming stranded assets as AI workloads demand thermal densities far beyond their original design, pushing operators toward either costly rebuilds or ceding market share to hyperscalers who build liquid-native facilities from the ground up.

Operational resiliency under extreme AI workloads hinges on advanced liquid cooling technologies—ranging from rear-door heat exchangers and direct-to-chip loops to full immersion cooling—that are now mandatory at densities exceeding 120 kW per rack. Companies like ASUS and Ready Server emphasize rigorous system commissioning, including full-rack scale testing, accelerated aging, and careful hardware preparation, to ensure reliability and maintain warranty coverage, while also balancing coolant temperatures to optimize energy efficiency without compromising component longevity.

Infrastructure considerations extend beyond cooling technology to encompass cabinet design, water management, and integrated power-cooling architectures. Robust cabinet engineering ensures structural integrity and airflow isolation critical for off-site integration and rapid deployment, while evolving water strategies address challenges posed by heat waves and droughts that threaten cooling water availability. Strategic acquisitions by companies like Eaton and Schneider highlight the industry's move toward unified power and cooling solutions, reflecting the necessity of holistic planning to balance performance, efficiency, and operational risk in high-density AI data centers.

Singapore’s stringent sustainability framework, exemplified by the DC-CFA2 standards requiring a PUE below 1.25, Green Mark Platinum certification, and at least 50% green energy sourcing, has transformed sustainability from a regulatory hurdle into a strategic advantage that filters and concentrates the market among well-capitalized incumbents like Equinix, Microsoft, and Singtel’s Nxera. This regulatory environment is complemented by ambitious projects such as the 700MW Jurong Island green data center park, which reserves 10% of its land for emerging energy technologies and heat reuse strategies, underscoring Singapore’s integrated approach to renewable energy and climate-adapted infrastructure expansion.

The tropical climate and resource constraints in Singapore have necessitated innovative cooling solutions tailored to local conditions, as demonstrated by Ready Server’s pioneering full-immersion liquid cooling pilot. By submerging servers in a non-conductive fluid, Ready Server not only boosts rack density and energy efficiency but also aligns with regional sustainability imperatives to reduce cooling energy consumption and operational costs. Their self-built, customized system reflects a strategic emphasis on in-house expertise and capital cost reduction, enabling adaptability to the challenges of tropical data center operations while supporting Singapore’s broader goals of energy-efficient, climate-resilient AI infrastructure.

Beyond Singapore, climate extremes such as record heat waves in Korea highlight the critical role of advanced liquid cooling technologies in maintaining AI data center reliability and sustainability. Traditional air cooling falters under soaring temperatures, risking GPU shutdowns above 30 degrees Celsius, while water-based direct-to-chip systems and liquid cooling methods offer more efficient thermal management. However, these climate-driven adaptations also raise sustainability concerns, as simultaneous heat waves and droughts threaten water availability for cooling, emphasizing the urgent need for climate-conscious energy transitions and resource-efficient cooling strategies in data center operations across diverse geographies.

By early 2026, the escalating power densities of AI workloads—from traditional 10–20 kW racks to cutting-edge configurations exceeding 120 kW—have transformed cooling from a mere operational expense into a fundamental architectural constraint. Data centers unable to adopt advanced liquid cooling technologies risk exclusion from the most lucrative AI market segments, as air cooling cannot efficiently support these thermal demands. This shift compels investors and operators to undertake costly mechanical redesigns or face the prospect of losing high-return tenants to more technologically equipped facilities.

The financial risks tied to inadequate cooling infrastructure are multifaceted, encompassing margin compression, stranded legacy assets, and heightened outage probabilities that can severely undermine profitability. For instance, retrofitting air-cooled facilities with liquid cooling can cost $2–3 million per megawatt, a stranding expense rather than a simple upgrade, compressing returns just as demand for 100-kilowatt racks surges. Furthermore, water scarcity and regulatory constraints—highlighted by the U.S. data center water use reaching 17 billion gallons in 2023—are increasingly factored into credit risk assessments, making sustainable and resilient cooling solutions a strategic imperative for operational continuity and expansion.

Recognizing the strategic necessity of liquid cooling, leading data center providers like Aligned Data Centers are pioneering scalable, on-demand liquid cooling solutions that allow enterprises to future-proof their infrastructure without premature capital outlays. This approach aligns with enterprise preferences to defer spending until densities surpass 35 kW per cabinet, balancing financial risk management with the flexibility to scale. As one expert noted, liquid cooling is becoming ubiquitous across AI infrastructure, underpinning hybrid cloud strategies where workload placement demands adaptable, high-efficiency cooling architectures.

The complexity of liquid cooling systems—exemplified by HPE’s innovations in warm water cooling to meet regional demands—underscores the critical importance of meticulous installation and maintenance to avoid costly leaks and downtime that could jeopardize multimillion-dollar AI infrastructure investments. Moreover, sovereign AI initiatives emphasize end-to-end infrastructure ownership and data transparency, reflecting broader strategic considerations that extend beyond cooling technology to encompass legal, operational, and territorial controls essential for trust and compliance in AI data center operations.