Lenovo expands AI and HPC infrastructure with new servers and services

New infrastructure targets compute and data bottlenecks

The expanded portfolio combines Lenovo ThinkSystem servers powered by 6th Generation AMD EPYC SP7 processors with high-performance storage, interconnect options and Lenovo Neptune liquid cooling. The company said the systems are intended to help organizations turn large volumes of data into scientific discovery, engineering advances and business innovation as workloads move toward continuous inference, agentic execution and more complex scientific applications.

The Lenovo ThinkSystem SC755 is a dual-socket, Neptune liquid-cooled server. According to Lenovo, it is designed to remove up to 100% of system heat through water and deliver up to 40% higher sustained performance than similar air-cooled systems. Its Lenovo Dynamic Partitioning capability allows customers to convert two dual socket systems into single sockets, and back, to allocate compute resources more flexibly.

Lenovo also announced the ThinkSystem SR755, designed for high-bandwidth workloads including large-scale virtualization, engineering, financial modeling, electronic design automation and agentic AI infrastructure. The company said it is engineered for high compute density and memory bandwidth, with lower power consumption and better economics than dual-socket configurations for demanding workloads.

Storage and services extend the portfolio

Lenovo's ThinkSystem DE6800 and DE6500 storage systems are designed for high-throughput, latency-sensitive block workloads. The DE6800 is intended for applications that require high throughput and reliable storage, while the DE6500 is positioned as an economical high-performance system with simplified deployment and management. Lenovo also announced validation for the Cornelis CN5000 fabric, providing another interconnect option for AI and HPC environments.

The company is expanding services for enterprise and gigawatt-scale Hybrid AI factories. New GPU deployment, configuration and optimization capabilities are intended to help customers build and scale hybrid AI factories. Multi-node support across Red Hat AI Enterprise and NVIDIA AI Enterprise with NVIDIA Run:ai is designed to provide more flexibility for deploying, governing and scaling workloads including agentic AI and physical AI.

For larger deployments, Lenovo is adding gigafactory services that include NVIDIA Infrastructure Services expertise for specialized deployment and cluster validation. The services cover concept and build through deployment and optimization, according to the company.

Examples span research and commercial production

Lenovo cited deployments at research institutions and commercial organizations as examples of AI and HPC workloads in use. At the University of Florida, the fourth generation of HiPerGator, built with Lenovo, NVIDIA and AMD technology, delivers 17.49 PFLOPS and processed 33 million research requests in the past year, according to Lenovo.

DreamWorks Animation uses Lenovo Neptune liquid-cooled servers in its data center. Lenovo said the deployment delivers 20% greater performance and faster rendering, helping the studio accelerate creative iteration.

Lenovo also said it can facilitate access to Digital Realty's global colocation network for organizations that need additional data center capacity. Customers can use the Digital Realty Innovation Lab environment to validate AI infrastructure, applications and workload requirements, while Lenovo provides a single point of contact and offers scalable infrastructure, modular services and flexible financing through Lenovo TruScale Infrastructure as a Service.

For users, the announcement means more options for assembling and operating AI and HPC environments across owned, hybrid and colocation infrastructure. For the market, it reflects Lenovo's effort to address the full system, from compute and storage to cooling, deployment and ongoing optimization. The performance figures cited for Lenovo's products are based on company testing or analysis, and the source says actual results may vary.