HBM stacks today sit 1.5 millimeters from a GPU. What if they didn’t have to? In this clip from a recent conversation with @theCUBE, Marvell Executive Vice President and General Manager of Data Center Networking Dave Lazovsky walks through the Photonic Fabric Memory Module, the first optically interconnected high-capacity, high-bandwidth memory solution, capable of delivering up to 32 terabytes of disaggregated memory capacity up to 50 meters away from compute. For the first time, memory capacity can scale independently from compute. Watch the full interview:

Today in New York, Marvell's leadership team shared why connectivity is the next defining unlock for AI infrastructure, and how a decade of investment has positioned the company to lead across every distance in the data center, from inside the package to between data centers. Watch the full event:

Marvell is demonstrating its co-packaged optics platform at @ECOC_Exhibition, including a 102.4 Tbps capable CPO implementation built on class-leading silicon photonics. The physics driving CPO adoption are straightforward. As per-lane speeds in scale-up fabrics increase, the distance over which signals can travel over passive copper falls. AI pods spanning multiple racks require longer interconnect distances than copper can support. CPO brings optical elements directly into the package alongside the IC, reducing distance, improving performance, and lowering power consumption. The demonstration at booth 2002 showcases 200G per lane with aggregate bandwidth in multiples of 6.4 Tbps, alongside detachable ELSFPs for rapid laser replacement and simplified manufacturability. #ECOC26# #ECOCExhibition#

The choice of optical interconnect architecture has significant implications for bandwidth, power, cost, and deployment flexibility as GPU clusters grow in size and complexity. Pluggable optics offer ecosystem maturity and operational simplicity. Co-packaged optics offer bandwidth density and power efficiency. Near-packaged optics sits between the two. At @ECOC_Exhibition, Marvell Vice President of Connectivity Lian Qin will present a comparative analysis of all three approaches, examining performance, power consumption, serviceability, ecosystem readiness, and total cost of ownership, and exploring how NPO addresses the evolving requirements of AI scale-up infrastructure. Learn more: