Odaily Planet Daily reports: Marvell Technology has announced a new generation of memory solutions for AI infrastructure, covering server-level AI storage, rack-level CXL memory expansion and pooling, and multi-rack optical interconnect shared memory, designed to address the growing memory capacity and bandwidth bottlenecks in Agentic AI inference processes.
Marvell stated that as AI model sizes grow, context windows expand, and KV Cache demands increase, traditional tightly coupled compute and memory architectures are limiting AI inference efficiency. Through memory disaggregation, memory resources can scale independently of compute resources, improving GPU utilization and reducing data movement latency. The newly released products include:
Bravera SC6 PCIe 6.0 SSD Controller: Designed for AI inference storage scenarios, it enables cloud service providers to migrate more KV Cache to high-performance SSDs, enhancing infrastructure efficiency. Featuring an architecture compatible with multi-vendor NAND, sampling is expected to begin in Q4 2026.
Marvell Structera X Memory Expansion Solution: Based on CXL technology, it supports rack-level memory expansion and resource pooling, enabling data centers to more flexibly share and allocate memory resources, reducing AI infrastructure costs.
Marvell Photonic Fabric Optical Interconnect Memory Solution: Builds a cross-rack shared memory architecture using optical interconnect technology, supporting up to 32TB of warm KV cache offloading and enhancing the throughput of AI inference clusters.
Marvell states that the Photonic Fabric solution can achieve up to 2 to 3 times higher token throughput within existing data center space and power constraints, enabling larger models and AI applications with longer contexts.
Marvell executive Will Chu said that AI infrastructure is shifting from a single-server architecture to systems where compute, memory, and connectivity work together协同运行, and in the future, memory will need to scale more independently to improve resource utilization and token efficiency.
As demand for AI agents and large model inference continues to grow, memory capacity, bandwidth, and data transfer efficiency have become the new focal points of competition in AI infrastructure, following computational power.