Is Liquid Cooling Enough to Keep HPC Sustainable in the AI Era?

I recently visited the Cray Museum and HPE's facility in Wisconsin as part of their AI Investor Day. During the visit, I spent time with Antonio Neri, Fidelma Russo, and Neil MacDonald, gaining deeper insights into HPE's innovations. A key focus of our discussions was on HPE’s pioneering fanless cooling solutions, which are integral to their next-generation supercomputing systems. The trip highlighted for me HPE's commitment to energy efficiency and sustainability in high-performance computing and AI infrastructure and how this will increasingly be a key part of the AI Infrastructure narrative..

The LRZ, part of the Gauss Center for Supercomputing, unveiled its plan for "Blue Lion," a advanced high-performance computing (HPC) system. Blue Lion, according to the details shared combines HPE Cray supercomputing infrastructure with next-generation NVIDIA accelerators, aiming to support both AI-centric and traditional scientific simulation workloads. The system introduces HPE’s Slingshot interconnect for high-speed data transfer, a feature that is designed to ensure scalability and minimizes latency during large-scale computations. Additionally, it employs 100% direct liquid cooling with warm water (up to 40°C), designed to reuse waste heat for operational cost savings and reduced emissions.

According to the press release a unique aspect of Blue Lion is its focus on hybrid workloads, which combine AI-driven modeling with conventional simulation techniques. This capability is particularly relevant for researchers addressing turbulence, fluid mechanics, or climate scenarios, where AI’s ability to recognize patterns complements physics-based modeling. Alongside the hardware, LRZ and HPE are launching workshops to prepare users for programming and optimizing workloads in this hybrid environment.

Blue Lion's announcement underscores a major shift in how supercomputing systems are designed, funded, and deployed. The growing demand for integrating AI into HPC workloads is redefining architectural approaches, with accelerators like NVIDIA GPUs becoming central components. This trend marks a divergence from traditional HPC systems focused solely on raw simulation power, moving toward AI-enhanced hybrid capabilities. 

The integration of liquid cooling and waste heat reuse reflects an industry-wide response to sustainability concerns. With energy costs and carbon footprint becoming critical metrics, solutions like Blue Lion are a necessary evolution. However, sustainability in HPC isn't just a matter of cooling efficiency. The ability to deliver high-performance computing while maintaining a manageable energy footprint will likely become a differentiator for supercomputing vendors in an era of escalating environmental scrutiny.

Blue Lion also highlights a broader issue: accessibility and usability in HPC. The emphasis on workshops and programming support indicates that hardware innovation alone is not sufficient for a leadership position. The complexity of hybrid workloads demands skillsets that blend traditional HPC expertise with AI and data science capabilities. This need for upskilling is a bottleneck, especially as governments and organizations push for sovereign AI initiatives and more decentralized research infrastructure. Skills and the lack thereof could be a key barrier to adoption as AI workloads move to maturity.

Another interesting development is Blue Lion’s use of NVIDIA accelerators in tandem with HPE Cray solutions. This partnership illustrates how major players in the supercomputing space are aligning their strategies to capture market share in both HPC and AI-driven workloads. The competitive landscape is shifting toward integrated solutions that combine hardware, software, and operational efficiency.  A key takeaway from my Wisconsin trip was that every system on the production line had some form of NVIDIA GPU in the build.

While the announcement focuses heavily on research and scientific exploration, the commercial implications of such systems shouldn’t be overlooked. Blue Lion’s architecture—especially its focus on scalable interconnects and hybrid AI capabilities—could serve as a prototype for future systems tailored to industries like healthcare, financial modeling, and autonomous systems development.