Cornelis and Supermicro advanced a joint infrastructure model that targets mounting power and networking limits in AI and HPC data centers. The companies validated Supermicro’s liquid-cooled FlexTwin server platforms with Cornelis’ CN5000 400Gbps fabric, creating a combined system designed to handle larger AI training and simulation workloads without increasing power draw or cluster complexity. The integration focuses on predictable performance at scale, addressing a growing trend in which network congestion—not compute capacity—dictates real workload throughput.
The collaboration arrives as operators push against tight thermal envelopes and rising energy costs. Supermicro’s FlexTwin architecture captures up to 95% of system heat through liquid cooling, while Cornelis’ CN5000 fabric provides deterministic, congestion-free communication between nodes. Together, the platform aims to improve workload throughput, enable higher rack-level efficiency, and allow clusters to scale while maintaining consistent performance. Both companies emphasized that power, cooling, and network scalability now define the upper bounds of AI and HPC deployment strategy.
Across manufacturing, physics, life sciences, and climate modeling benchmarks, the validated FlexTwin + CN5000 system produced measurable efficiency gains: up to 1.5× higher application performance versus competing 400Gbps fabrics, a 29% average increase in performance per networking dollar, and up to 2.3× higher performance per watt in liquid-cooled environments. The companies said these improvements allow customers to run larger problems on fewer nodes while staying within fixed energy budgets.
• Integration validates Cornelis CN5000 with Supermicro FlexTwin liquid-cooled platforms
• Targets AI and HPC environments constrained by power, cooling, and network congestion
• CN5000 delivers deterministic 400Gbps communication and congestion-free data movement
• FlexTwin captures up to 95% of heat removal via direct liquid cooling
• Supports both air- and liquid-cooled CN5000 switches, plus air- and conduction-cooled SuperNICs
• Up to 1.5× higher application performance vs. competitive fabrics
• 29% improvement in performance per networking dollar
• Up to 2.3× higher performance per watt
• Improved scaling efficiency for larger clusters
• Solution available now through Supermicro and Cornelis partners
“We’re giving our customers the tools they need to accelerate the most demanding HPC and AI workloads without compromise,” said Vik Malyala, President & Managing Director EMEA, SVP Technology & AI, Supermicro.
🌐 Analysis
This collaboration underscores how AI and HPC system design is shifting toward power-efficient networking and dense liquid-cooled compute architectures. Cornelis has steadily positioned CN5000 as an alternative to Ethernet and InfiniBand fabrics in large-scale simulation environments, while Supermicro continues to expand its portfolio of liquid-cooled platforms aimed at hyperscale and enterprise AI deployments.
