AI networking is quickly becoming a critical part of the enterprise. You need make the right choice about which hardware you want to use but you aren’t necessarily locked into a single choice. In this episode of the Tech Field Day podcast, brought to you by Cisco, Tom Hollingsworth is joined by Scott Robohn and Richard Licon of Cisco as they discuss rapidly evolving landscape of AI networking, emphasizing that organizations no longer have to settle for a “one-size-fits-all” approach. They explore the shift from purely software-driven innovation back to high-performance hardware, where custom silicon like Cisco Silicon One and merchant silicon are becoming central to infrastructure. The conversation highlights how the transition from AI model training to real-time inferencing in 2026 demands more flexible and programmable hardware pipelines. By leveraging tools like P4 programming, network engineers can now adjust their infrastructure mid-cycle to optimize traffic patterns and ensure consistent job completion times as workload needs shift.

The discussion also highlights Cisco’s strategic partnership with NVIDIA, offering customers a choice between distinct networking ecosystems based on their specific goals. For those requiring strict adherence to NVIDIA’s validated designs and the Secure AI Factory, the Cisco Nexus 9100 with Spectrum-X silicon provides a certified, high-performance path. Conversely, organizations seeking maximum flexibility and custom tuning for diverse workloads can opt for Cisco Silicon One. Ultimately, the experts agree that while the technical requirements of AI networking are increasingly complex, the availability of diverse silicon options and consistent operating systems like NX-OS or SONiC allows enterprises to build future-proof, composable infrastructure that can adapt to the unpredictable pace of AI development.