Astera Labs Extends Leadership in Open, AI Scale-Up Networking with New 320 Lane Scorpio X-Series Smart Fabric Switch

Astera Labs, Inc., a player in semiconductor-based connectivity solutions for rack-scale AI infrastructure, today announced the Scorpio X-Series 320 Lane Smart Fabric Switch, the industry’s largest open, memory-semantic fabric switch engineered to improve token economics and support large scale-up clusters with minimal latency.

Astera Labs also announced an expanded Scorpio P-Series PCIe fabric switch family—now spanning 32 to 320 lane configurations—designed to give data center architects at AI labs, hyperscalers, and neo-clouds the flexibility to rapidly scale compute capacity across diverse accelerators for training and serving frontier AI models.

 

As AI deployments move into large-scale production and achieve operational efficiency, infrastructure teams face a new set of constraints: multi-trillion-parameter models, agentic workflows, and multi-step reasoning distributed across heterogenous compute infrastructure. The industry needs connectivity solutions purpose-built for these workloads—higher radix to simplify topologies, intelligent fabric capabilities to reduce communication overhead, open and platform-specific optimizations, and datacenter-grade diagnostics to maintain uptime when a single fault can cost millions of dollars in idle compute. Scorpio Smart Fabric Switches are purpose-built to address each of these constraints and meet the demands of production AI at scale.

 

Scorpio’s software-defined architecture is designed to integrate seamlessly with leading merchant and custom silicon, enabling AI labs and hyperscalers to integrate and deploy new accelerator platforms for both training and inference. Its memory-semantic connectivity enables accelerators to access fabric resources through native load/store operations, eliminating software overhead and improving fabric efficiency at scale. Scorpio X-Series delivers simplified high-radix scale-up topologies, cutting hops and reducing end-to-end latency across the cluster.  Newly introduced Hypercast and In-Network Compute engines accelerate collective operations by up to 2x to maximize GPU utilization and tokens-per-watt performance. Rounding out the family, the Scorpio P-Series complements the X-Series in front-end network and AI compute system deployments, to deliver resiliency, dynamic configurability, and broad interoperability.

 

Cosmos software unifies the platform with purpose-built resiliency and serviceability, delivering non-disruptive firmware updates, OpenBMC management, and real-time telemetry. Hypercast and In-Network Compute are configured through Cosmos, extending its capabilities to now include platform performance optimizations. These capabilities are designed to maximize uptime, accelerator utilization, and ensure operational reliability for continuous production workloads while accelerating rack validation and compressing time-to-deployment. Cosmos extends across Astera Labs’ complete rack-scale portfolio of fabric switches, copper connectivity, and optical solutions, enabling composable AI infrastructure from a single unified management software stack.

 

“The frontier models driving today’s most demanding AI applications require connectivity infrastructure that keeps pace with the accelerators powering them,” said Jitendra Mohan, CEO, Astera Labs. “The Scorpio X-Series 320 Lane high-radix AI fabric switch replaces multiple legacy switches to enable larger scale-up cluster sizes in a single hop and reduce overall latency. Hardware-accelerated Hypercast and In-Network Compute engines further boost collective operations by up to 2x to improve tokens-per-watt performance. Together, with Astera Labs’ complete rack-scale connectivity portfolio, including the expanded 32 to 320 lane Scorpio P-Series PCIe fabric switches, give infrastructure teams a platform to build and scale AI entirely on their terms.”

 

The Scorpio X-Series 320 Lane Smart Fabric Switch is shipping into a merchant scale-up switch silicon market projected to reach $20 billion by 2030, with production ramp in 2H 2026.