In this interview from VMware Explore 2026, Krish Prasad, senior vice president and general manager of the VMware Cloud Foundation Division at Broadcom, joins theCUBE's John Furrier to discuss how three years into VCF's journey, private cloud and private AI are converging with the rise of agentic AI workloads. Prasad details how VCF 9.1 is seeing broad momentum across customers of all sizes as enterprises bring more AI workloads on-premises. He explains why unifying VMs, containers and AI workloads on the ESX compute platform eliminates infrastructure silos and undercuts public cloud costs. Prasad also breaks down how early access to Anthropic's Glasswing program and its Mythos frontier models has hardened VCF's software development lifecycle, backed by a commitment to continuous patching and live-patching capabilities that keep environments secure without disrupting running workloads.

The conversation also explores how VCF's memory tiering innovation helps customers cut DRAM costs by roughly half amid the current supply-driven price spike, without affecting application performance. Prasad unpacks the VCF AI Factory, which packages drivers, open-source components and more than 150 supported models — from NVIDIA's Nemotron to Google's Gemma — into a push-button stack now expanding beyond NVIDIA to include AMD accelerators and running across OEM servers from Dell to Supermicro. He touches on the resurgence of VMware's practitioner community and its growing partner ecosystem, increasingly anchored around native Kubernetes and container workloads. Prasad also reflects on how sovereignty has evolved beyond a legal or geographic concept into an enterprise-wide infrastructure control issue, with VCF giving organizations full control over where and how their data and models run. From memory innovation to AI Factory deployments, Prasad outlines how Broadcom is positioning VCF as the foundation for the next phase of enterprise AI.

In this interview from VMware Explore 2026, Chris Wolf, global head of AI and advanced services for the VMware Cloud Foundation Division at Broadcom Inc., joins theCUBE's John Furrier to discuss how enterprise AI is maturing from experimentation into production-grade private AI cloud deployments. Wolf explains why the convergence of container-based workloads, agentic loops and GPU-driven inference is forcing IT teams to rethink infrastructure around security, governance and efficient resource use. He underscores that software, not hardware procurement, must lead architectural decisions, giving organizations the flexibility to route workloads between frontier models and local specialized models as needs evolve.

Wolf breaks down Broadcom's AI Factory, an end-to-end solution spanning compute, networking, storage and model runtimes that eliminates the "buyer's remorse" many customers face when piecing together AI infrastructure themselves. He details how the offering solves for change management, lifecycle patching and memory optimization across GPU pools, problems that otherwise inflate server and licensing costs. The conversation also explores Broadcom's growing ecosystem, including support for more than 150 on-prem models and 40 cloud-based providers, plus a deepening partnership with Google around Gemma models and hybrid agent architectures. Wolf highlights sovereignty as a major growth area, pointing to Shared Services Canada's CANChat deployment as proof that customers can retain full control of data, identity and encryption keys. From architecting for constant change to keeping software at the center of every AI decision, Wolf offers a clear framework for enterprises navigating the next phase of private AI.

In this interview from VMware Explore 2026, Umesh Mahajan, vice president and general manager of the Application Networking and Security Division at Broadcom Inc., joins theCUBE's John Furrier to discuss how agentic AI is reshaping the enterprise attack surface. Mahajan explains why perimeter-based defenses no longer suffice as LLMs, MCP servers and autonomous agents proliferate across infrastructure. He details how Broadcom's vDefend platform layers zero trust, zero day protection and advanced threat prevention into an integrated stack, citing throughput of 75 terabits per vCenter cluster for firewalling and 17 terabits for IDS/IPS to keep pace with AI-scale traffic without adding latency.

The conversation also explores how Broadcom's Avi Load Balancer is evolving into a full AI gateway, combining web application firewall, bot management and API protection to secure Kubernetes-based agentic workloads. Mahajan breaks down the need for real-time discovery of agents and MCP tools, arguing that lateral segmentation could have contained incidents like the Hugging Face breach. He touches on how AI is also transforming security operations themselves, automating policy enforcement while keeping humans focused on setting guardrails rather than manual configuration. From rising hardware costs to the shift toward scale-up throughput per core, Mahajan outlines why security must now be built into AI infrastructure from day one rather than bolted on afterward.

In this interview from VMware Explore 2026, Purnima Padmanabhan, general manager of the Tanzu Division at Broadcom Inc., joins theCUBE's John Furrier to discuss how Tanzu is evolving into the secure runtime enterprises need for agentic AI. Padmanabhan explains how AI has moved past experimental pilots into broad-based, everyday use across engineering teams, creating demand for a standardized way to build and govern agents at speed. She details the newly announced Tanzu Platform for Agents, now integrated into VMware Private AI Cloud, which operates on a "deny by default" model — giving agents access only to the specific models, skills, tools, data and services they are explicitly granted.

The conversation also explores Broadcom's expanding investment in trusted open source, including its Mythos-powered scanning and patching programs for Spring, Python, Node.js and the broader Java ecosystem, which Padmanabhan says represents the largest patch release in Spring's 23-year history. She unpacks the platform's new AI-ready data foundations, which automatically chunk, vectorize and curate structured and unstructured data into governed data products that agents can safely consume, reducing token usage and operational risk. Padmanabhan also touches on how MCP is emerging as the connective tissue between agents, models and services across the ecosystem, and why coding and workflow automation remain the leading production use cases. From sandboxed runtimes to curated data access, she outlines how enterprises can move agentic AI from pilot projects into secure, production-grade deployments.

In this interview from VMware Explore 2026, Clayton Donley, vice president and general manager of the Identity Management Security Division at Broadcom, joins theCUBE's John Furrier to discuss why the rise of agentic AI is forcing enterprises to rethink identity, trust and governance for machine-driven workloads. Donley explains how AI has moved from simple, user-directed queries to autonomous agents making high-stakes decisions with far less oversight. He details the need to assign unique identities to agents, define the scope of their intent and enforce policy at the point where agents call large language models and APIs. Donley also unpacks emerging risks, including agents manipulated through falsified context, and why traditional attack-surface thinking must expand to account for an organization's own agents acting unpredictably.

The conversation also explores how Broadcom's AgentMinder platform builds on decades of application-monitoring and distributed-tracing expertise, adapted for agentic environments rather than built from scratch. Donley breaks down its three pillars: identity for agents, a control point between agents and LLMs, and observability into prompts and tool calls. He notes how AI itself is now being used to assign more granular, auditable permissions than human-managed systems ever could, while emphasizing that adoption should layer onto existing identity infrastructure rather than require disruptive overhauls. From early-adopter customers piloting agent registration and traffic monitoring to the broader push toward private AI and on-premises control, Donley outlines a practical path for enterprises to scale agentic AI without losing sight of governance.