KubeCon 2026 Day 1: AI Economics, Kubernetes, and Neoclouds | Sarbjeet Johal on theCUBE
At KubeCon + CloudNativeCon North America 2026 in Salt Lake City, Stackpane’s Sarbjeet Johal joined theCUBE’s John Furrier and Rebecca Knight to discuss what it takes to run AI in production, from Kubernetes and distributed inference to token budgets and the economics of neoclouds.
Kubernetes and the AI factory
The panel looks at how teams get more work out of existing GPUs by coordinating inference across Kubernetes, virtual machines, and containers. Johal describes short-lived virtual machines that start on demand and disappear when the job is done, while context can carry across sessions.
AI agents change what teams need to budget
As developers run multiple agents across models with different costs, model choice becomes an operating decision. The panel discusses caching and token budgets as early ways to control consumption, while developers focus more of their time on architecture and creative work.
What neoclouds need beyond compute
Johal asks how specialized cloud providers can serve enterprises that have to run new AI workloads alongside existing systems. The panel considers software expertise and systems integration as factors that could distinguish providers as customers move from experiments into production.
Jump to a chapter
- 00:07 — Kubernetes as the foundation for AI infrastructure
- 05:42 — Short-lived virtual machines and open source agents
- 08:00 — How autonomous coding changes development
- 13:25 — AI FinOps and token budgets
- 16:00 — Neoclouds and traditional hyperscalers
- 18:49 — Semiconductors and systems integration
- 21:06 — Neocloud risks and KubeCon takeaways
Published by SiliconANGLE theCUBE on November 10, 2026. Watch the original discussion on YouTube.