Liquid Cooling of Hyperscale Data Centers – Challenges and Opportunities

Presenter: Dereje Agonafer, Phd, University of Texas at Arlington
Location: ERF 1043

Abstract: As demand from High Performance Computing (HPC), AI, and memory-intensive workloads grows, air cooling can no longer keep up - especially beyond 30–50 kW per rack. Liquid cooling is now essential, with solutions like direct-to-chip, immersion, and two-phase systems offering superior thermal performance, higher density, energy efficiency, and heat reuse. Despite the performance benefits, liquid cooling introduces several engineering challenges. Direct-to-chip systems use microchannel cold plates with geometries often under 30 µm, which require stringent control of fluid cleanliness, mechanical tolerances, and thermal interface stability. Corrosion risk increases due to electrochemical reactions, pH imbalance, or contamination, especially in tight flow paths.

Two-phase liquid cooling, which uses phase change to enhance heat transport, brings additional complexity. These systems must maintain precise pressure regulation and vapor containment, and are vulnerable to issues such as material expansion, seal degradation, and mechanical stress from thermal cycling. Recently, our team embarked (with funding from ARPA-E) to develop a novel cooling technology which includes evaporative cold plate with hollow silicon micropillars for vapor-only heat dissipation, liquid metal composite thermal interface materials for improved conductivity, and gyroid heatsinks for efficient cooling of high-power components. Machine learning models accelerate design optimization, while system-level integration uses Supermicro hardware and software to enhance performance.

Despite all current challenges, the opportunities remain plentiful. As energy efficiency and thermal management become key design constraints in modern computing systems, liquid cooling is positioned to play a critical role in enabling scalable, high-density, and sustainable data center architectures. The market is projected to reach $48.42 billion by 2034 (24% annual growth rate), which indicates strong momentum toward adoption in next-generation thermal management.

Speaker Bio: Dr. Agonafer is a presidential distinguished professor in the Department of Mechanical and Aerospace Engineering at the University of Texas at Arlington. He heads two centers and is site director of NSF IUCRC in Energy Efficient Systems and director of Electronics, MEMS and Nanoelectronics Systems Packaging Center. He is now building a new center called RAHIS (Reliability Assessment in Heterogeneous Integrated Systems). After receiving his PhD at Howard University, he worked for 15 years at IBM, where he was awarded the IBM Outstanding Technical Achievement Award for Computer Aided Thermal Modeling. Since joining UTA in 1999, he has graduated over 260 graduate students (a record for the University) including 41 PhDs. His current research areas are in energy efficiency of data centers, heterogeneous integration, and 3D packaging and cooling. He has received several awards including the Semi-Therm Thermi Award, the InterPACK Excellence Award, the ITHERM Achievement Award, the NSBE Golden Torch Award for Golden Torch Legacy, and the Distinguished Alum Award from the University of Colorado. He is an ASME Honorary Member, and fellow of AAAS and the National Academy of Inventors. He has received the ASME prestigious Heat Transfer Memorial Award and the SEMI-THERM Thermal Hall of Fame for “Significant Contributions to Electronics Thermal Management.” In 2019, he was elected to the National Academy of Engineering, and in 2020, he was presented the Howard University Alumni Award for Distinguished Postgraduate Achievement in the field of Engineering.