Marvell brings AI memory and interconnect tech to SC26 in Chicago

The company will showcase at booth 4003 from November 15 to 20, 2026, focusing on memory capacity and optical connectivity where AI infrastructure faces limitations. Marvell suggests that AI's biggest challenge may not be computation but rather getting data to compute resources quickly enough.

Memory Technologies on Display

The headliners include Photonic Fabric memory modules supporting up to 32 TB of shared memory across multiple racks via optical links, and the second-generation Structera X 3406/3506 CXL 3.2 memory expanders offering six-channel designs, PCIe Gen 6 compatibility, and support for up to 9 TB of memory.

Marvell claims Photonic Fabric delivers 2-3x throughput gains for inference workloads, though this is a company claim without independent verification. The CXL expanders are expected to sample in Q1 2027, making the Chicago showcase a preview rather than immediately available products.

Supporting Technologies

Marvell's booth will also feature Bravera SC6 SSD controllers (sampling Q4 2026), DPUs, 1.6T/800G active electrical cables, and Teralynx switches already deployed in hyperscale AI clouds. Preet Virk, Marvell's SVP and GM, is scheduled to speak at SC26 on November 18.

The Memory Bottleneck

These products address rising KV cache demands in agentic AI inference, where models maintain short-term memory during multi-step tasks. The longer and more complex the task, the greater the memory consumption becomes, making technologies supporting 9 TB and 32 TB capacity increasingly relevant.

Market Implications

Marvell aims to establish its CXL-based products and optical architectures as core components for hyperscale cloud and enterprise environments. While the Teralynx switches provide existing footholds with target customers, the 2-3x inference throughput claim for Photonic Fabric requires validation beyond Marvell's own materials.

Key developments to watch include whether Structera X samples arrive on schedule in early 2027, whether Photonic Fabric's performance claims hold in customer deployments, and if any named partners emerge for the shared-memory approach. Sampling represents only an early step, with volume orders still uncertain.