XConn Technologies and MemVerge Unveil Transformative Scalable CXL Memory Solution for AI
In a groundbreaking demonstration at Supercomputing 2025 (SC25) in St. Louis, XConn Technologies and MemVerge have introduced an innovative Compute Express Link (CXL) memory pool aimed at resolving critical memory challenges associated with artificial intelligence (AI) inference workloads. The event, taking place from November 16 to 21, 2025, highlights a 100 TiB-scale memory solution that targets the emerging “memory wall” problem as enterprises increasingly adopt large language models (LLMs) and real-time AI applications.
The rapid progress in server FLOPS has not been matched by memory bandwidth expansions, with many AI tasks now facing significant limitations due to insufficient memory capabilities. This imbalance has prompted experts to voice concerns about an impending memory bottleneck that necessitates advancements in memory and interconnect architecture. AI’s reliance on memory-intensive processes — including retrieval-augmented generation, vector searches, and LLM inference — is outpacing the capabilities of traditional server architectures, leading to performance and total cost of ownership (TCO) challenges.
Gerry Fan, CEO of XConn Technologies, emphasized the collaboration with MemVerge: “As AI workloads and model sizes explode, the limiting factor is no longer just GPU count, but also memory access speed and cost efficiency.” He noted that the joint effort demonstrates the feasibility of CXL memory pooling at substantial scales, asserting the solution’s readiness for production environments.
The demonstration showcases how the integrated technology — featuring XConn’s Apollo hybrid CXL/PCIe switch and MemVerge’s GISMO — facilitates dynamic sharing of key-value (KV) cache across GPUs and CPUs. This enhanced memory pooling enables over 5× performance improvements compared to traditional SSD caching methods, highlighting a significant advancement for AI inference workloads. The innovative architecture addresses the prefill and decode disaggregation challenges effectively, contributing to a more efficient data processing pipeline.
Charles Fan, CEO and co-founder of MemVerge, echoed this sentiment: “Memory has become the new frontier of AI infrastructure innovation.” He highlighted how the dynamic, low-latency nature of CXL memory pooling revolutionizes memory provisioning in AI-centric data centers.
SC25 attendees will have the opportunity to witness the capabilities of this state-of-the-art CXL memory pool, with live benchmarks demonstrating substantial performance and efficiency gains for caching and AI model execution. For more information about SC25 and registration details, visit the official website at https://sc25.supercomputing.org.
About XConn Technologies
XConn Technologies is at the forefront of next-generation interconnect technology, revolutionizing high-performance computing and AI applications with its hybrid switch that integrates CXL and PCIe capabilities. Learn more at xconn-tech.com.
About MemVerge
MemVerge specializes in AI memory software, empowering enterprises to optimize performance in AI data centers through efficient memory sharing between GPUs. For more information, visit memverge.ai.
