Specifications
Capacity
512 GB per module
Interface
CXL 2.0 carried over an EDSFF connector
Device class
CXL Type 3 memory expansion device
Form factor
EDSFF E3.S
Media
Kioxia XL-FLASH Generation 2
Media characteristics
Low-latency NAND with higher write endurance than TLC or QLC
Demonstrated latency
755 ns
Demonstrated throughput
Just under 1 GB/s
Random read
Close to 2 million IOPS at 100% 512B loads
Memory tiering
DRAM for hot pages, XL1 NAND for cold memory pages
Cost profile
Capacity added below DRAM cost per gigabyte
Bay flexibility
The same E3.S bay can host NVMe SSDs or CXL devices
CXL topology
Memory expansion behind a CXL controller, similar to OMI approaches
Target workload
AI and memory-intensive data centre server workloads
Announced
August 3, 2026
Shown at
FMS 2026 — the Future of Memory and Storage
Overview
The Kioxia XL1 is a NAND-based CXL memory expansion module that attacks the capacity side of the memory wall rather than the bandwidth side. It puts 512 GB of Kioxia XL-FLASH Generation 2 behind a CXL 2.0 interface in a standard EDSFF E3.S form factor, and presents itself to the host as a Type 3 memory expansion device — capacity, not cache.
The point of XL-FLASH here is its latency and endurance profile: unlike TLC or QLC NAND it was designed for low-latency, write-intensive use, which is what makes a NAND tier viable behind CXL. In live demonstrations at FMS 2026 the 512 GB device ran at just under 1 GB/s with 755 ns latency and came within a fraction of 2 million IOPS on a pure 512B load test.
Because the module uses an EDSFF connection running CXL instead of PCIe, system designers keep flexibility at the bay level: the same slot accepts conventional NVMe SSDs or CXL devices. The intended architecture is tiered memory — DRAM holding hot pages while the XL1 holds colder regions — which lowers the blended cost per gigabyte for AI and memory-hungry server workloads as DRAM capacity costs continue to climb.
Key Benefits
Capacity without DRAM pricing — 512 GB of addressable memory expansion at a cost per gigabyte below DRAM. Built for tiering — 755 ns latency and near-2M IOPS on 512B reads is fast enough to act as the cold tier under DRAM. Standards-based integration — CXL 2.0 Type 3 in E3.S, so existing EDSFF bays and CXL-capable platforms can host it. Durable media — XL-FLASH Generation 2 offers markedly higher write endurance than commodity TLC or QLC NAND.
Applications
AI inference and training servers short on memory capacity, CXL memory tiering and pooling designs, in-memory analytics and large key-value caches with cold-page spill, vector database and RAG pipelines, and any platform where DRAM cost per gigabyte is the binding constraint.
Request a Quote — KIOXIA XL1 — CXL 2.0 XL-FLASH MEMORY EXPANSION MODULE
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →