Nutanix has launched Nutanix Enterprise AI (NAI) 2.8, adding a generally available Model Context Protocol (MCP) Gateway designed to govern how artificial intelligence agents connect with enterprise applications and data.
The company said NAI 2.8 is available now as part of a broader expansion of its Nutanix Cloud Platform for production AI. Nutanix Kubernetes Platform (NKP) 2.19 is also planned for release soon, with features aimed at running AI applications across virtualized and bare-metal environments.
MCP Gateway targets agent access to tools and data
The main addition in NAI 2.8 is the MCP Gateway within Nutanix Agent Gateway. MCP has emerged as a standard way for AI agents and assistants to connect with external tools, applications and data sources. As enterprises expose more internal systems through MCP, access control and visibility over those connections become part of the deployment challenge.
Nutanix said its gateway provides a centralized layer for governing agent access to applications and data. The company is also offering an MCP Server for Nutanix Cloud Platform so customers can build agentic applications that interact with infrastructure managed by Nutanix.
The release also expands Nutanix Private Inference, including support for parameter-efficient fine-tuning using Low-Rank Adaptation for smaller models, multi-GPU inference through tensor parallelism, batch inference and speculative decoding. Nutanix said speculative decoding can accelerate token generation by up to 2.5 times, although actual performance will depend on model, hardware and workload configuration.
Security features highlighted by Nutanix include fine-grained identity and access management, custom roles, model sharing controls and support for air-gapped NVIDIA NIM deployments. The company is positioning these controls around the need to limit the scope of what AI agents can access as they move from experimentation into production workflows.
Nutanix links AI adoption to mixed VM and container environments
Nutanix is also emphasizing what it calls a dual-native architecture, where virtual machines and containers can be managed through a common platform. The company argues that this matters because enterprise applications and data often remain distributed across traditional virtualized systems while newer AI workloads increasingly run in containerized environments.
Rather than requiring customers to rearchitect existing applications before deploying AI, Nutanix said its approach is intended to let organizations place AI workloads closer to the systems and data they already operate.
Jay Tuseth, vice president and general manager for Asia Pacific and Japan at Nutanix, said organizations in the region are under pressure to move AI from experimentation into business use while continuing to support existing infrastructure. He said customers are looking for ways to scale AI with governance, security and operational consistency without undertaking large rearchitecture projects.
The approach also builds on Nutanix’s earlier positioning around infrastructure fragmentation in Asia Pacific, where enterprises often operate a mix of legacy applications, virtualized environments, containers and cloud services.
NKP 2.19 and partner programs are next
NKP 2.19, which Nutanix says will be available soon, is expected to add simplified container management across bare metal and virtualized environments. Planned features include NKP Metal for bare-metal Kubernetes, an AI applications catalog for deploying software such as Kubeflow, Milvus and Slurm, and broader hardware and GPU support.
Nutanix said NKP has also received Cloud Native Computing Foundation Kubernetes AI Conformance certification. As with other upcoming NKP 2.19 features, availability and final functionality remain subject to the product’s release.
Alongside the product updates, Nutanix has made Service Provider Central generally available and introduced a Powered by Nutanix Verified Services program for partners building infrastructure, cloud-native and AI services.
The company is pitching the combined changes as a way to support production AI without separating AI infrastructure from the rest of the enterprise application estate. For organizations evaluating agentic AI platforms, the more immediate question will be how consistently the new controls work across existing identity, application and data environments, particularly as MCP connections increase the number of systems an AI agent can reach.
Asia Pacific and Japan’s biggest IT challenge isn’t AI, it’s fragmentation