DataChill

Liquid Cooling for AI: Is Higher Flow Rate the Only Answer? The rapid rise of high-performance AI processors—such as NVIDIA’s GB300 (1400W TDP) and AMD’s MI355X (1100W TDP)—is pushing thermal management to its limits. In response, liquid cooling companies are showcasing cold plates that handle over 2kW, or even 4kW TDP. But how is this achieved? A deeper dive into recent technical developments reveals that many solutions simply increase the coolant flow rate to manage higher heat loads. Is this truly a breakthrough—or just brute force engineering? ?? The Reality of High-TDP Liquid Cooling Liquid cooling, particularly single-phase direct liquid cooling (DLC), has become the de facto solution for AI and HPC workloads. With air cooling unable to handle rising TDPs, single-phase DLC removes heat via sensible heat transfer—a mechanism that requires an increase in coolant flow rate to maintain temperature differentials. However, simply pumping more liquid into a system does not represent a fundamental advancement in cooling technology. It raises system-level inefficiencies due to: ? Higher power consumption from increased pump workloads. ? Greater pressure drop, leading to distribution and sealing challenges in dense server racks. ??? Beyond Flow Rate: Smarter Cooling Innovations To effectively manage high-TDP chips without just increasing flow rates, the industry must innovate. Key advancements include: 1?? Advanced Channel Geometries Cold plates with novel fin structures, or jet impingement designs enhance turbulence and heat transfer at lower flow rates. This improves efficiency without excessive liquid movement. 2?? High-Conductivity Materials Copper and hybrid metallic-composite cold plates reduce thermal resistance, improving heat transfer without needing higher fluid velocity. 3?? Two-Phase Cooling Liquid-to-vapor phase change significantly boosts heat absorption per unit volume, reducing the need for bulk liquid movement. This approach is gaining traction in high-density AI racks. ?? The Future of High-Wattage Cooling While increasing flow rate is a quick fix, it is not a sustainable long-term strategy. Future cooling designs must focus on: ?? Efficient cold plate architectures with enhanced heat dissipation. ?? System-wide thermal optimization, including fluid routing and distribution. ?? Hybrid cooling methods (combining direct liquid, immersion, and two-phase). ?? Final Thought The AI revolution is driving exponential compute demands, and thermal bottlenecks will dictate scalability. Cooling efficiency is now as critical as processing power. The companies that innovate beyond simply increasing flow rates will lead the future of sustainable AI infrastructure. Are we on the verge of a cooling revolution, or are we just turning up the pumps? #LiquidCooling #AIInfrastructure #DataCenters #SustainableCooling #HPC #TwoPhaseCooling #ThermalManagement #AdvancedCooling #ColdPlate #DataCenterCooling #AIThermalChallenges #EnergyEfficiency

SmartSense Coolant Distribution Unit (CDU) seamlessly integrates into AI-ready data centers, delivering peak performance when paired with JetCool’s SmartPlate? cold plates to accelerate AI time-to-solution. Designed for GPU-dense, high-power racks, it cools up to 300kW per rack and scales to 2.1MW in a row-based configuration. As a Flex company, JetCool provides end-to-end liquid cooling solutions—from design and integration to products, services, and global support. Backed by Flex’s worldwide footprint, we meet any power, any scale, anywhere.

Is growth in the data center sector set to continue? Our latest report reveals the top trends that will shape the sector in 2025 and beyond. From the impact of #AI on power demands to the rise of rural data center development, discover what's next for the industry: https://cbre.co/4briliW

Novva Lines up $2 Billion for Utah Campus Expansion The Novva Data Centers campus in Salt Lake City is about to get a lot bigger. Novva and CIM Group have arranged a $2 billion construction loan from JPMorgan Chase and Starwood Property Trust. The financing was highlighted by The Wall Street Journal, which called it “the latest sign of the market’s enormous appetite for facilities that provide the backbone of artificial intelligence.” Novva completed the first building in 2022 on the 100-acre Utah campus, which can support three data center buildings and about 1 million SF of space. The 175-megawatt project is already fully-leased, the Journal reported. Novva Data Centers was founded by Wes Swenson, who previously built C7 Data Centers into a regional network. The company is currently expanding across the Western U.S., with projects in Nevada, Colorado, Arizona and California. The WSJ story: https://lnkd.in/eghhfeRp