Nutanix expands cloud platform with controls for agentic AI

INFRA

Nutanix Inc. today introduced new capabilities intended to help enterprises run agentic artificial intelligence applications alongside existing virtual machines and containerized workloads without splitting their infrastructure into separate management silos.

The updates include the general availability of Nutanix Enterprise AI 2.8 and a forthcoming release of Nutanix Kubernetes Platform 2.19. Nutanix also made its Service Provider Central program for cloud partners generally available and detailed a partner program for building cloud, Kubernetes, AI and virtual machine migration services.

The company calls its approach “dual-native” because its platform treats VMs and containers as first-class infrastructure. That means customers can run Kubernetes on Nutanix’s Acropolis Hypervisor virtualization platform when isolation and operational consistency are priorities or deploy Kubernetes directly on bare-metal systems for workloads that require different performance or resource profiles.

The distinction is not simply the ability to support both architectures, said Thomas Cornely, executive vice president of product management at Nutanix.. “It’s not about getting containers working; it’s about how you operate and manage those containers,” he said.

The customer selects the environment for each workload, and Nutanix optimizes deployment based on that decision. “They decide,” Cornely said. “What we can do is provide optimizations of how the workload is actually getting deployed once you choose the location and the substrate. But the choice is theirs.”

NAI 2.8 adds a generally available Model Context Protocol gateway to Nutanix Agent Gateway. It provides a central point for governing the tools and data that AI agents can access through MCP. A separate MCP Server for Nutanix Cloud Platform gives agents controlled access to infrastructure managed by the company’s software.

The gateway also tracks token use and allows organizations to impose quotas at the team, user or agent level. That addresses a growing customer concern as autonomous agents make repeated model calls that users may not see or anticipate.

“The first things that customers ask for are visibility, control and governance,” Cornely said. “They don’t control the cost per token.” He said this can leave customers with bills they can’t control or predict.

Cornely said Nutanix has encountered customers whose monthly AI allocations disappeared much faster than planned due to rampant token usage. “We’ve seen scenarios where customers had a budget for a month that was spent in a week,” he said.

NAI can steer appropriate work toward privately deployed open-weight models, where customers pay for infrastructure rather than individual tokens. “If I’m doing basic scripting of Python coding, I don’t need a frontier model,” Cornely said. “I could do that with an open-weight model.”

Private Inference in NAI 2.8 adds Low-Rank Adaptation fine-tuning for models containing fewer than 8 billion parameters, multi-graphics processing unit inference using tensor parallelism, batch inference and speculative decoding. Nutanix claims that speculative decoding can increase token-generation speed by up to 2.5 times, though Cornely acknowledged that results depend on the model and its infrastructure configuration.

The release also expands auditing. Consumption controls can be applied to individual agents, then aggregated across users, agents and teams. MCP activity logs can record which data sources and applications an agent has accessed, as well as what actions it has taken. However, inspecting prompts is not the product’s default purpose, Cornely said.

“Your MCP servers are only as secure as the backend infrastructure and the back-end set of API keys and role access controls,” he said.

NKP 2.19, scheduled to be available soon, will extend Kubernetes management across virtualized and bare-metal environments. NKP Metal automates operating system, firmware and container deployment, while NKP on AHV integrates with Nutanix Flow for network isolation. An application catalog will offer curated deployments of the open-source Kubeflow, Milvus and Slurm tools favored for AI development. The platform has also received Cloud Native Computing Foundation Kubernetes AI Conformance certification, Cornely said.

Nutanix also said its Unified Storage product has received enterprise-level Nvidia certification.  It’s used to feed data to GPUs with low latency and high throughput. Use of Nutanix storage is optional.

For service providers, SP Central provides a multitenant control plane for providing infrastructure, application, cloud-native and AI services. The Powered by Nutanix: Verified Services program provides onboarding, delivery materials and badges intended to help partners build recurring services businesses.

Although the announcements strengthen Nutanix’s pitch to customers of Broadcom Inc.’s VMware virtualization software, Cornely said displacing VMware needs to be part of a broader modernization effort. “You don’t just replace the VMware with the same old thing,” he said.

Photo: Nutanix