Compute Express Link, or CXL, is a switch heterogeneous memory and accelerator network that was initially introduced by Intel to encourage the use of their Optane memory. CXL enables increasing local memory, but even more interesting, enables the creation of a storage pool that could include DRAM and other memory technologies as well as various processor accelerators. CXL has been discussed at the FMS since it’s start. This article looks into recent CXL developments discussed at the conference.

The chart below is from a talk by my colleague, Jim Handy from the conference, showing his projections for CXL revenue growth. He projects significant growth from 2027 through 2031.

SK hynix’s G2.5 addition to their storage and memory hierarchy shown during their FMS keynote, is CXL, shown below.

CXL can be used to expand local storage and especially to create pooled memory that can be allocated to applications using software. SK hynix had several CXL products on display at the 2026 FMS Conference, including accelerators, memory expansion modules and their MARS memory-centric AI rack system.

An interesting observation is that Meta reported that it had used a CXL memory pool to repurpose and extend the life of older DRAM DDR4 memory rather than paying the price to buy DDR5 memory to support some of their data center operations .

Cameron Brett from Kioxia spoke about their latest NAND flash solutions. These include BiCS10 with 332-layers (jointly manufactured for Kioxia and Sandisk) as well as their XL-Flash. XL-Flash is built with Kioxia’s BiCS Flash 3D memory technology. It is an extremely low-latency, high-performance flash memory. XL-FLASH™ is designed to address the performance gap that currently exists between volatile memories such as DRAM and current flash memory.

XL-FLASH™ features a 128 gigabit (Gb) die for SLC / 256 gigabit (Gb) die for MLC (in a 2-die, 4-die, 8-die package), a 4kB page size for more efficient operating system reads and writes, fast page read and program times, and a read latency of less than 5 microseconds. The slide below shows advantages of an XL-Flash CXL memory module off-loading data from DRAM.

Kioxia is also using its XL-Flash products to support Nvidia’s Storage-Next initiative to make flash storage behave more like high-speed memory for AI data centers.

Jeremy Werner, Senior Vice President and General Manager of Micron's Core Data Center Business Unit, delivered the Micron keynote and discussed disaggregated memory pools using CXL. As shown below, the pool using DDR5 DRAM RDIMM memory, was able to achieve 40TB of shared memory in a 4U form factor.

This memory pool was able to provide up to 20 times the DRAM capacity as the direct attached memory and was able to generate 6.7X more tokens/second with 85% less GPU resources.

CXL and its applications was on display in many of the exhibits at the 2026 FMS, including at system maker exhibits. For instance, AIC was showing its CXL-ready enterprise server lineup, including their SB201-SU all-flash NVMe server equipped with a CXL-ready drive cage for AI data workloads.

Anil Godbole, Xeon CXL Strategy & Marketing Manager at Intel gave a talk about the current state and the future of CXL. He showed this slide of Tech Insights data on projected adoption of memory expansion and memory pooling, showing that memory pooling would be used in 49% of servers by 2031.

The 2026 FMS featured growing adoption of CXL for memory expansion and memory pooling applications with projections of significant growth in adoption over the next few years.