As artificial intelligence systems handle larger context windows and expanding datasets, standard server architecture increasingly runs into memory capacity and bandwidth constraints. To address these infrastructure hurdles, Marvell Technology, Inc. has expanded its memory infrastructure portfolio to help hyperscalers and cloud operators separate memory expansion from processing power. The strategy aims to optimize resource usage and improve token efficiency across agentic AI workloads.
Traditional server-attached storage often restricts hardware efficiency by forcing compute units to wait for data, which lowers overall throughput. By introducing memory disaggregation through expansion, pooling, and sharing, the updated portfolio reduces data movement and latency. This structural shift allows data centers to maximize model processing performance while operating within strict power and space constraints.
“AI infrastructure is moving beyond isolated servers to systems where compute, memory and connectivity operate seamlessly together,” said Will Chu, executive vice president and general manager, Custom Cloud Solutions, at Marvell. “As AI scales, memory must scale more independently of compute so resources can be deployed where they deliver the greatest value. With the industry’s broadest AI memory infrastructure portfolio, Marvell is helping customers improve utilization, boost token efficiency and scale AI without compromising performance, power or cost.”
Industry analysts highlight data transmission limits as a core challenge for modern processing environments. “As AI workloads grow larger and more complex, memory capacity, bandwidth, latency and data movement are becoming primary constraints on AI performance,” said Alan Weckel, co-founder and technology analyst at 650 Group. “Marvell’s memory and storage portfolio gives hyperscalers and cloud providers a strong foundation for building scalable, efficient AI systems capable of supporting increasingly advanced workloads.”
The updated product lineup addresses data management across three distinct operational levels. For server-level storage, the Marvell Bravera SC6 PCIe 6.0 SSD controller delivers double the performance of the earlier SC5 generation. By offloading key-value caches to SSD storage, the component lowers write amplification and supports various NAND suppliers for flexible deployment.
For rack-level setups, the Structera X solution enables memory pooling across multiple servers to lower total cost of ownership. At the pod level, Marvell introduced Photonic Fabric optics, including modules, network interface cards, and chiplets. This optical architecture creates a shared-memory tier spanning up to 50 meters, enabling low-latency offloading for up to 32TB of warm key-value cache to significantly increase token output. Sampling for the Bravera SC6 controller is scheduled for the fourth quarter of 2026.