Marvell accelerated its push into AI-scale connectivity today by announcing broad industry deployment of its Alaska P PCIe 6 retimers across leading server, accelerator, and interconnect platforms. The company positioned its 5 nm PAM4-based devices as essential components for next-generation compute fabrics, where longer reaches, higher bandwidths, and stricter latency budgets are redefining how GPUs, XPUs, CPUs, CXL devices, and storage subsystems connect inside hyperscale data centers. Server OEMs have now integrated Alaska P retimers directly into GPU and XPU platforms, while general-purpose servers are adopting dedicated retimer cards to stabilize PCIe 6 links across increasingly complex system topologies.
Cable and optical module vendors have also taken up the technology, introducing PCIe AEC and AOC offerings that embed Marvell retimers to support rack-scale AI cluster designs. Several storage system providers are evaluating Alaska P retimers to maintain signal quality on CPU-to-SSD connections as PCIe 6’s 64 GT/s PAM4 signaling introduces significant channel loss. With modern accelerator-rich systems often pushing PCIe traces and cabling to physical limits, Marvell emphasized that retimers are no longer optional—they are now foundational to scaling AI infrastructure.
The Alaska P family compensates for roughly 40 dB of PCIe 6 channel loss while adding telemetry, diagnostics, and fleet-management features aimed at hyperscalers building large-scale PCIe fabrics for AI, storage, and CXL. The devices can be deployed on server boards, integrated into copper assemblies, or paired with electrical-to-optical components for emerging optical PCIe architectures. Marvell said this flexibility enables cloud operators to design disaggregated, accelerator-centric compute fabrics capable of meeting rapidly growing AI workload demands.
• Alaska P PCIe 6 retimers now deployed in GPU/XPU platforms across multiple OEMs
• Adopted in retimer cards for general-purpose servers and evaluated for CPU–SSD signal integrity
• Embedded in PCIe AEC and AOC products from cable and optical partners
• PCIe 6 at 64 GT/s PAM4 requires retimers to overcome ~40 dB channel loss
• Built on Marvell 5 nm PAM4 SerDes with fleet-level telemetry and diagnostics
• Supports disaggregated AI and CXL fabrics across servers, racks, and clusters
• Available now through server vendors and cable partners
“The adoption of the Alaska P PCIe retimers underscores our leadership in enabling the AI infrastructure transition from traditional server architectures to disaggregated, accelerator-centric compute fabrics,” said Xi Wang, senior vice president and general manager of the Connectivity Business Unit at Marvell.