Cisco builds Silicon One G300 for massive AI workloads

A networking titan just dropped a 102.4-terabits-per-second switching chip built to handle AI cluster workloads at scale and slash energy waste in data centers.

Silicon One G300 chip targets AI infrastructure

• Cisco's new silicon will power upcoming Nexus 9000 and 8000 systems designed for training and inference.
• The chip hits 102.4 Tbps throughput to support massive distributed AI deployments.
• Hardware-level security is baked in, and the design allows future upgrades without infrastructure replacement.
• Intelligent Collective Networking features include shared buffers, path-based balancing, and proactive telemetry.

Performance and efficiency gains stack up

• Cisco claims 33% higher network utilization and 28% faster job completion versus non-optimized setups.
• Liquid-cooled and air-cooled 102.4T systems deliver nearly 70% energy efficiency improvement over prior generations.
• New 1.6T optics and 800G Linear Pluggable Optics cut optical power consumption by up to 50%.
• Overall switch power use drops by 30% with the new optics tech.

Nexus One software platform gets AI observability upgrades

• The unified AI networking platform now includes AI job observability and Splunk integration.
• API-driven automation and AgenticOps use guided troubleshooting to simplify network management.
• Features aim to reduce operational complexity for sovereign and compliance-sensitive deployments.

Shipping and partnerships locked in

• The G300 chip, systems, and optics are set to ship this year.
• Cisco is collaborating with NVIDIA, AMD, Intel, NetApp, DDN, and VAST on integrated AI stacks.