XConn Technologies and MemVerge Demonstrate CXL Memory Pool for KV Cache using NVIDIA Dynamo for breakthrough AI workload performance at 2025 OCP Global Summit

Joint demo highlights CXL memory pool (up to 100TiB) is the right scale up memory solution for solving AI workload memory wall issue, from both performance and TCO point of views

XConn Technologies (XConn), the innovation leader in next-generation interconnect technology for high-performance computing and AI applications, and MemVerge®, the leader in Big Memory software, announced a joint demonstration of Compute Express Link® (CXL®) memory pool for breakthrough AI workload memory scale-up at the 2025 OCP Global Summit, October 13–16, in San Jose, California.

As AI applications continue to surge in scale and complexity, the industry faces an urgent challenge—the memory wall. To power the next generation of intelligent computing, a true memory scale-up solution is essential. CXL memory pooling, now commercially viable and rapidly expanding, stands as the only proven path forward. By enabling dynamic, low-latency, and high-bandwidth sharing of massive memory resources across CPUs and accelerators, it breaks through traditional architectural limits. 100 TiB commercial CXL memory pools are available in 2025 and even larger deployments are on the horizon for 2026 and beyond.

Also Read: AiThority Interview With Dmitry Zakharchenko, Chief Software Officer at Blaize

The demo will highlight a CXL memory pool, powered by the XConn Apollo switch and MemVerge Gismo technology, integrated into NVIDIA’s Dynamo architecture and NIXL software, to handle the KV cache exchange and offloading. It will show the CXL memory pool not only a suitable solution to the memory wall issue, but also a significant performance boost (> 5x) for AI inference workloads, in comparison with SSD. By combining the XConn Apollo switch, the industry’s first hybrid CXL/PCIe switch, with MemVerge’s Memory Machine X software, the companies will showcase how enterprises can achieve breakthrough scalability, performance, and efficiency for large AI inference and training models.

Demonstrations will be available in the OCP Innovation Village Booth 504, providing attendees with multiple opportunities to explore the joint solution in action. During the event, XConn’s Jianping Jiang, Senior Vice President of Business and Product, will also detail the benefits of scale up memory solution for AI workload powered by XConn’s Ultra IO Transformer technology during the session, “Co-Designing for Scale: CXL-Based Memory Solution for Data-Centric Workloads,” to be presented during OCP on Wednesday, October 15 at 11:05 a.m.

“As AI workloads hit the memory wall issues, CXL memory pool is the only viable memory scale up solution for today and the near future. It not only dramatically boosts AI workload performance but also provides significant TCO benefits,” said Gerry Fan, CEO of XConn Technologies. “Our collaboration with MemVerge at OCP demonstrates how CXL memory pool is a ready for deployment solution to be applied to even the most demanding AI applications.”

“AI is fueling a revolution in infrastructure design, and memory is at the heart of it,” said Charles Fan, CEO and co-founder of MemVerge. “By pairing GISMO with the XConn Apollo switch, we are showcasing how software-defined CXL memory can deliver the elasticity and efficiency needed for AI and HPC. This collaboration extends the possibilities of CXL 3.1 to help organizations run larger models faster and with greater resource utilization.”

The joint demo will illustrate how MemVerge’s Global IO-free Shared Memory Objects (GISMO) technology enables NVIDIA’s Dynamo and NIXL to tap into huge CXL memory pool (up to 100TiB in 2025) and serve as the KV Cache store for AI inference workloads, where prefill GPUs and Decode GPUs work in synchrony to take advantage of the low latency and high bandwidth memory access to complete the computing. When combined with XConn’s low-latency and high lane count switch fabric, the result is a new class of memory infrastructure capable of supporting large and scalable memory pool size with lower TCO, ready to tackle the increasing challenging work for AI inference, generative AI, real-time analytics, and in-memory databases.

Also Read: Neuro-Symbolic AI Cities – Designing “Thinking Cities”