Introduction

On May 14, 2025, Cisco announced a new family of AI-native networking hardware and software, marking its most aggressive move yet into the high-performance infrastructure powering large language models and edge inference. The launch includes the Silicon One G200 ASIC, a switch/router platform optimized for low-latency AI clusters and distributed workloads across hybrid environments.

“We’re not just connecting packets—we’re connecting reasoning systems,” said Jonathan Davidson, EVP and GM of Cisco Networking.¹ The G200 chip boasts 51.2 Tbps of switching capacity and introduces telemetry hooks built for LLM flow classification and token-throughput monitoring.

Cisco’s portfolio now includes adaptive optical interconnects, energy-aware load balancing, and integrated support for NVIDIA and AMD’s AI platforms. With zero-trust policy enforcement baked into every flow, Cisco positions its stack as the networking glue for both hyperscale clouds and secure enterprise edge AI.

Why it matters now

• LLM training and inference create unpredictable east-west traffic patterns.
• Enterprise networks must evolve to support AI-native observability and routing.
• Cisco faces pressure to stay relevant as data center designs shift from CPU-centric to AI-accelerator-first.

Call-out: Cisco repositions itself from network provider to AI infrastructure backbone

The G200 family can deliver microburst mitigation and adaptive flow control for token-based inference traffic, slashing tail latencies by over 40%.

Business implications

Enterprise CIOs must revisit network design assumptions—AI-native workloads demand flow introspection, fine-grained telemetry, and protocol support far beyond legacy IP switching.

Security leaders benefit from the built-in segmentation and per-token enforcement, potentially reducing attack surfaces in inference-driven workflows. Meanwhile, edge AI deployments (retail, manufacturing, logistics) can gain from simplified orchestration and embedded zero-trust at the packet level.

Looking ahead

Cisco plans to integrate G200 into Meraki-managed stacks and announced early pilots with Salesforce, Siemens, and the U.S. Department of Energy. It’s also expected to release an open SDK for AI network telemetry by Q3 2025.

The upshot: Cisco is betting that the future of AI is as much about transport as it is about compute. As networks become co-processors in their own right, the infrastructure game shifts once again—and disruption moves to the wire.

––––––––––––––––––––––––––––
¹ Jonathan Davidson, Cisco Live Keynote, May 14, 2025.