WS Investor
17 Mar 2026, 19:57
Marvell Technology Inc. Unveils CXL Switch to Address AI “Memory Wall”
Marvell Technology Inc. announced the launch of its next-generation Structera S 30260 CXL switch, designed to enable rack-level memory pooling and tackle growing memory constraints in AI data centers.
The new 260-lane CXL device allows hyperscalers to access disaggregated memory across CPUs, GPUs, and accelerators, significantly improving memory utilization, bandwidth, and data flow efficiency. As AI workloads—particularly large language models—drive exponential memory demand, the solution provides a scalable alternative to traditional architectures and reduces reliance on costly high-bandwidth memory.
Supporting CXL 3.0 with up to 4TB/s bandwidth, the switch forms part of Marvell’s broader end-to-end CXL portfolio, following its acquisition of XConn Technologies. The product is expected to begin sampling in the third quarter of 2026, positioning Marvell to capitalize on rising demand for AI infrastructure optimization.
Business Wire
Marvell Technology Inc. announced the launch of its next-generation Structera S 30260 CXL switch, designed to enable rack-level memory pooling and tackle growing memory constraints in AI data centers.
The new 260-lane CXL device allows hyperscalers to access disaggregated memory across CPUs, GPUs, and accelerators, significantly improving memory utilization, bandwidth, and data flow efficiency. As AI workloads—particularly large language models—drive exponential memory demand, the solution provides a scalable alternative to traditional architectures and reduces reliance on costly high-bandwidth memory.
Supporting CXL 3.0 with up to 4TB/s bandwidth, the switch forms part of Marvell’s broader end-to-end CXL portfolio, following its acquisition of XConn Technologies. The product is expected to begin sampling in the third quarter of 2026, positioning Marvell to capitalize on rising demand for AI infrastructure optimization.
Business Wire