Marvell to Show AI Memory Tech at Supercomputing 2026
Marvell will exhibit memory and optical interconnect products at booth 4003 during the Chicago event, November 15 to 20.
Updated on Nov. 13, 2026 in Semiconductors
Marvell Technology will showcase AI memory and interconnect products at Supercomputing 2026 in Chicago, running November 15 to 20, 2026, at booth 4003. Its headline offerings are Photonic Fabric memory modules and Structera CXL memory solutions.
Why it matters
The products target rising memory demands from agentic AI inference, where short-term model memory, known as KV cache, stays in use for longer. Marvell is positioning shared memory over optical links and CXL-attached memory as ways to meet those demands.
Marvell reports Photonic Fabric memory pools of up to 32 TB spanning multiple racks and claims 2-3x inference throughput gains, while its second-generation Structera X 3406/3506 CXL 3.2 expanders use six-channel designs compatible with PCIe Gen 6 and support up to 9 TB. Independent verification of the throughput claim has not been reported.
The players
Marvell Technology
Marvell Technology is a publicly traded semiconductor company that designs custom silicon, networking, and storage chips for data centers and cloud infrastructure.
Preet Virk
Preet Virk is an executive at Marvell Technology who is scheduled to speak at the SC26 conference.
The details
At the Chicago conference, Marvell will display SSD controllers, DPUs, 1.6T and 800G active electrical cables, and advanced switching silicon alongside its memory products. The company's Teralynx switches are already deployed in hyperscale AI clouds.
Timeline
November 15 to 20, 2026: Supercomputing 2026 exhibition in Chicago
November 18, 2026: Preet Virk scheduled to speak at SC26
Q4 2026: Bravera SC6 SSD controllers expected to sample
First quarter 2027: Structera X expanders expected to sample
The Tech Race
Marvell's Structera X expanders build on the Compute Express Link standard for attaching extra memory to processors over high-speed connections, now moving to CXL 3.2 and PCIe Gen 6. The move positions Marvell against other memory and interconnect suppliers competing to serve AI inference workloads in data centers.
For data center operators and AI developers, the announced memory pooling and CXL expanders could change how much memory a single inference workload can access. Adoption timing depends on the first-quarter 2027 sampling schedule, so near-term infrastructure decisions will not be affected.
The takeaway
Memory capacity is becoming a central constraint for AI inference, and vendors are responding with shared and expandable memory architectures. Readers tracking data center hardware should watch whether the sampling dates hold in 2027.
What happens next
Marvell is expected to sample its Bravera SC6 SSD controllers in Q4 2026 and its Structera X CXL memory expanders in the first quarter of 2027.
Further reading
For broader context on chip design and memory hardware, see Semiconductors coverage on Wisevoter.
