Cisco CEO Chuck Robbins told CNBC Thursday that the company is riding a wave of AI-driven demand that he termed a “networking supercycle.” The stock popped 14%, its best single-day gain in over two decades, after the company raised its fiscal year AI infrastructure and hyperscaler guidance from $5 billion to $9 billion.

From an architecture standpoint, this shift reflects a fundamental rebalancing in data center topology. As AI workloads—particularly large-scale training and inference—move from GPU-heavy clusters to distributed systems, the network fabric becomes the bottleneck. Cisco’s bet is that its silicon and optics portfolio can address latency and bandwidth constraints that traditional Ethernet struggles with. The 5% workforce reduction, focused on legacy routing and switching, is a deliberate reallocation toward these high-margin, high-complexity components.

Robbins acknowledged the challenge of forward booking in a volatile AI market, noting that Cisco has walked away from some hyperscaler projects where design wins didn’t align with capital commitments. This is a pragmatic engineering trade-off: locking in long-term contracts for custom silicon or optical modules requires precise demand signals, which are still immature.

On the security front, Cisco’s involvement in Anthropic’s Project Glasswing—testing AI model vulnerabilities—signals a growing intersection between networking and cybersecurity. For data engineers, this means the network layer is no longer just a pipe; it’s a vector for model attacks and data exfiltration. Robbins’ advice to be “agile and ready to move” translates operationally to building infrastructure that can rapidly re-route traffic, isolate compromised nodes, and log anomalies without adding latency.

Cisco may trail Nvidia in GPU dominance, but in the race to scale AI infrastructure, the network is becoming the critical path. The question is whether Cisco’s reallocation is fast enough to keep pace with the hyperscalers building their own custom silicon.

Source: CNBC