HPE and NVIDIA announced a major expansion of their AI Factory platform designed to help enterprises deploy autonomous AI agents in real-world production environments with greater security, governance, and cost efficiency. The new capabilities address a critical gap: while many organizations have experimented with AI agents, few have figured out how to run them safely and economically at scale. The announcement, made at HPE Discover 2026 in Las Vegas, introduces new software, hardware, and security features that work together to give companies the oversight and controls they need to move beyond pilot projects.

What's Actually Changing in How Companies Deploy AI Agents?

The core of the announcement centers on HPE Private Cloud AI, a turnkey platform developed jointly with NVIDIA. The platform now includes NVIDIA Agent Toolkit software, which bundles together NVIDIA Nemotron open models, NVIDIA NemoClaw, and the NVIDIA OpenShell secure runtime. These tools work as an operating system for AI agents, giving companies the ability to monitor what agents are doing, enforce policies, and reduce the risk of unexpected behavior.

One of the most pressing concerns enterprises face is rogue agent actions, where autonomous systems take steps their operators didn't anticipate. To address this, HPE is integrating new capabilities from HPE Zerto Software that can detect when agents behave unexpectedly and use continuous data protection to restore systems to a clean state if something goes wrong. The platform also adds secure local agent registration, allowing companies to approve which AI models, skills, and tools can be deployed while maintaining centralized governance and security policies.

"As AI becomes more autonomous, organizations need a new architecture to run it securely, govern it responsibly, and scale it economically. Across networking, servers, storage and software, HPE is delivering full-stack AI solutions with NVIDIA that build the foundation for agentic enterprises, helping customers move from experimentation to production with control and confidence," said Antonio Neri, President and Chief Executive Officer at HPE.

Antonio Neri, President and Chief Executive Officer, HPE

How to Optimize AI Agent Deployments for Cost and Performance?

• Data Pipeline Efficiency: HPE Alletra Storage MP X10000 automatically applies metadata and governance policies to unstructured data, preparing it for AI applications while cutting token response times by up to 20 times, according to benchmark testing. This means companies can process data faster and reduce the computational overhead of running agents.
• Token Throughput Optimization: New prompt processing efficiency improvements can boost token throughput by up to 20 percent, helping organizations maximize GPU utilization and control operating costs. Token throughput refers to how many words an AI model can process per second, so higher throughput means faster responses and lower per-query costs.
• Multi-Node Scaling: The platform now supports multi-node inferencing across up to 256 graphics processing units (GPUs), a unified model gateway for governed access to frontier AI models, and fine-tuning capabilities through NVIDIA NeMo. This allows companies to scale agent deployments from small pilots to enterprise-wide operations without redesigning their infrastructure.

HPE is also targeting one of the biggest cost drivers in AI deployment: inference, the process of running a trained model to generate predictions or responses. By improving how data is prepared and how tokens are processed, the company says organizations can significantly reduce the computing power and time required to run agents at scale.

How Does Security and Sovereignty Fit Into Enterprise Agent Deployment?

For organizations handling sensitive data or operating in regulated industries, security is non-negotiable. HPE is integrating NVIDIA Confidential Computing into the HPE AI Factory, a technology that uses cryptographic attestation and encryption to protect AI models and private data while they're being used. This approach establishes a chain of trust across hardware, software, and datasets, helping companies comply with regional and industry standards such as Cybersecurity Maturity Model Certification (CMMC), National Institute of Standards and Technology (NIST) 800, and Federal Information Processing Standards (FIPS).

Beyond confidential computing, HPE is deploying NVIDIA BlueField data processing units (DPUs) and NVIDIA DOCA software across the AI Factory to enforce zero-trust security policies, detect threats at runtime, and encrypt all networking traffic tied to AI workloads, agents, and data. Zero-trust means the system assumes no user or device is trustworthy by default and verifies every action before allowing it to proceed.

"Every layer of the computing stack is being reinvented for the age of AI agents. Together with HPE, we are building AI factories for this new era of computing, powered by NVIDIA Vera CPUs, accelerated infrastructure, and secure AI software, to help enterprises transform their data into intelligent action," said Jensen Huang, Founder and Chief Executive Officer at NVIDIA.

Jensen Huang, Founder and Chief Executive Officer, NVIDIA

What Hardware and Software Are Being Added to the Platform?

The announcement includes significant hardware upgrades alongside the software capabilities. HPE Private Cloud AI now supports HPE ProLiant Compute DL394 Gen12 with NVIDIA Vera CPU, a compute-optimized foundation designed specifically for agentic AI workloads and high-performance data processing. The broader HPE AI Factory will also be available with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, NVIDIA Spectrum-X Ethernet, NVIDIA BlueField-3 DPUs, and NVIDIA ConnectX-8 SuperNICs, all based on NVIDIA reference architectures.

HPE Data Fabric Software is being expanded to support agentic workflows by extending support for model context protocol (MCP) to Apache Airflow, an open-source workflow orchestration tool. The company is also introducing an enterprise AI inventory that enriches distributed data with metadata, making it easier for agents to find and use the information they need. A standalone HPE Data Fabric appliance, available on HPE ProLiant Compute servers, simplifies deployment for organizations that want to avoid complex custom installations.

When Will These Capabilities Actually Be Available?

HPE is rolling out the new features in phases. New HPE Private Cloud AI features will be available in July 2026, with HPE Data Fabric Software arriving in October 2026. Additional capabilities, including agentic observability, data intelligence, HPE Alletra Storage MP X10000, NVIDIA Agent Toolkit support, and NVIDIA NemoClaw, will be available in the fourth quarter of 2026. HPE Zerto Software support for agent action monitoring and continuous data protection will also arrive in Q4 2026. The HPE ProLiant Compute DL394 Gen12 with NVIDIA Vera CPU is scheduled for 2027, and NVIDIA Confidential Computing will be available for HPE AI Factory with NVIDIA in Q4 2026. The HPE AI Factory with NVIDIA RTX PRO Blackwell Server Edition GPUs and related networking hardware is available now.

The announcement reflects a broader industry shift toward operationalizing AI agents in enterprise environments. As organizations move beyond experimentation and begin deploying autonomous systems to handle real business processes, they need infrastructure that can provide visibility into agent behavior, enforce policies, protect sensitive data, and scale economically. HPE and NVIDIA's expanded platform is designed to address these requirements, giving enterprises the tools to move from pilot projects to production deployments with confidence.