Why Enterprises Are Moving AI On-Prem | Built to Scale: Secure AI Factory Ep. 1 (The Stack)

Cloud AI bills are getting out of hand, and most enterprises can’t even tell you where the money went. A single bad pull request merged into production can burn through an entire month’s token budget overnight. That’s not a hypothetical. It’s happening.

In this episode, Rajeev Khanolkar and Matt break down Cisco’s Secure AI Factory: what it is, why on-prem AI is quickly becoming a strategic priority, and how the full stack (UCS + NVIDIA GPUs, HyperFabric AI, AI Defense, and smart networking) gives enterprises the control, security, and cost predictability that cloud alone can’t deliver.

What we get into:

• Why cloud AI costs are unsustainable (IBM’s 5-year TCO data puts on-prem at roughly 30% of cloud cost)
• Token governance and how unmanaged spend quietly destroys ROI
• GPU utilization reality check: enterprise GPUs average just 2% utilization and how to actually fix that
• How Cisco’s Secure AI Factory wraps security guardrails around the entire AI workload stack
• Agentic AI and identity: how do you authenticate an agent with no face in a Kubernetes environment?
• Cisco x NVIDIA GTC expansion: Nexus 9100 switches, Hypershield, Cilium network mesh, and eBPF-level security
• How Cisco AMs can help enterprise customers move from cloud pilots to on-prem production

Gruve is a Cisco partner specializing in AI infrastructure deployment through its CAISS practice, helping enterprises move from pilot to production using Cisco Validated Designs and AI PODs.