Astera Labs Starts Shipping Scorpio X-Series 320-Lane Smart Fabric Switch

Astera Labs, Inc. reported that it started shipping the Scorpio™ X-Series 320 Lane Smart Fabric Switch and expanded its Scorpio P-Series PCIe fabric switch family to configurations spanning 32 to 320 lanes. The company tied the updates to supporting larger scale-up cluster sizes and reducing latency for large-model deployments. It also described COSMOS software as part of the platform used to manage updates and telemetry.

Astera Labs said the products target constraints it associated with large-scale production AI, including multi-step reasoning and distributed workloads across heterogeneous compute infrastructure. The company cited requirements for higher radix to simplify topologies, intelligent fabric capabilities to reduce communication overhead, open and platform-specific optimizations, and datacenter-grade diagnostics for uptime.

The Scorpio Smart Fabric Switches were described as software-defined and designed to integrate with merchant and custom silicon. The memory-semantic connectivity was described as enabling accelerators to access fabric resources through native load/store operations to remove software overhead. The Scorpio X-Series was also described as providing simplified high-radix scale-up topologies, and introducing Hypercast™ and In-Network Compute engines configured through COSMOS.

Astera Labs said COSMOS software unifies the platform with resiliency and serviceability features, including non-disruptive firmware updates, OpenBMC management, and real-time telemetry. It said COSMOS extends across Astera Labs’ complete rack-scale portfolio of fabric switches, copper connectivity, and optical solutions under a unified management software stack. Jitendra Mohan, CEO of Astera Labs, said, “The frontier models driving today's most demanding AI applications require connectivity infrastructure that keeps pace with the accelerators powering them,” adding that the Scorpio X-Series “replaces multiple legacy switches to enable larger scale-up cluster sizes in a single hop and reduce overall latency.” He also said, “Hardware-accelerated Hypercast and In-Network Compute engines further boost collective operations by up to 2x to improve tokens-per-watt performance.”

Forward-looking statements in the release covered expectations for Scorpio product features, capabilities, benefits, a production ramp in 2H 2026, and related risks and uncertainties.