NVIDIA announced a sweeping set of enterprise AI agent partnerships and software at GTC Taipei during COMPUTEX 2026, positioning the company's Agent Toolkit as the foundational infrastructure for autonomous digital coworkers across the world's largest software platforms. The initiative represents the commercial maturation of agentic AI, moving from experimental demos to production deployments that are being integrated directly into the enterprise software systems where work gets done.
The NVIDIA Agent Toolkit comprises three core components working in concert. NVIDIA NemoClaw provides orchestration blueprints connecting popular agent harnesses for deployment and is available now on GitHub. NVIDIA OpenShell serves as a secure runtime establishing policy and privacy controls for agents across on-premises, hybrid, and cloud environments, creating isolated sandboxes with system-level policy enforcement. And CUDA-X libraries expose NVIDIA's accelerated computing capabilities as agent-callable skills, including cuDF for data processing, cuOpt for routing and scheduling optimization, AI-Q for intelligent routing and persistent context, NeMo for agent evaluation and model customization, PhysicsNeMo for high-fidelity physics models, and CUDA-Q for quantum program generation and simulation.
Jensen Huang, NVIDIA founder and CEO, emphasized that the world's software leaders are bringing AI agents into the systems where work gets done, showing how AI coworkers help employees think faster and execute complex tasks to solve bigger problems.
The partnerships span virtually every major enterprise software category. In design and engineering, Cadence deployed OpenShell to secure its ChipStack AI Super Agent for autonomous chip design and verification. Dassault Systemes integrated NemoClaw and OpenShell into its 3DEXPERIENCE platform for design and manufacturing workflows. Siemens incorporated the tools into its Fuse EDA AI Agent for semiconductor and circuit board design. Synopsys partnered to build always-on autonomous engineers for chip design workflows.
In healthcare and operations, Foxconn is piloting NemoClaw for its Nurabot and CoDoctor platforms supporting clinical reasoning and care coordination, while also developing MoMClaw, a factory operations agent connecting sensor data with AI agents for manufacturing intelligence. In cybersecurity and analytics, CrowdStrike uses Nemotron models for agents that identify and remediate vulnerabilities, while Palantir integrated Nemotron into its AI FDE platform for autonomous enterprise task execution.
Platform-level integrations include Microsoft partnering on Windows security primitives and native agent experiences, Canonical integrating OpenShell with Ubuntu through supported containers, Red Hat embedding OpenShell into its full-stack AI platform with policy oversight, SAP embedding OpenShell in its Joule Studio runtime, and ServiceNow securing its Project Arc autonomous desktop agent with OpenShell.
Powering these deployments is NVIDIA Nemotron 3 Ultra, a 550-billion-parameter mixture-of-experts model delivering 5x faster inference and up to 30% lower cost compared to comparable frontier models for complex agentic tasks. Nemotron 3 Ultra was made available starting June 4 via Hugging Face, ModelScope, OpenRouter, and build.nvidia.com. The model is specifically optimized for agent orchestration frameworks and comes with open weights, datasets, and recipes for customization, enabling enterprises to fine-tune the model for their specific domain requirements while maintaining the efficiency gains of the mixture-of-experts architecture.
The scope of NVIDIA's enterprise agent partnerships reveals how quickly agentic AI is moving from research prototypes to production deployments embedded in mission-critical business systems. The three-component Agent Toolkit — NemoClaw for orchestration, OpenShell for security, and CUDA-X for accelerated skills — provides a standardized foundation that enterprise software vendors can build upon without starting from scratch. The partnership breadth is remarkable: chip design at Cadence and Synopsys, manufacturing at Foxconn, cybersecurity at CrowdStrike, analytics at Palantir, and enterprise operations at SAP and ServiceNow. Nemotron 3 Ultra's open-weight release on Hugging Face signals NVIDIA's strategy of building an ecosystem around its models rather than restricting access, differentiating from competitors who keep their frontier models proprietary. The emphasis on OpenShell integration across operating system vendors including Microsoft, Canonical, and Red Hat positions NVIDIA's agent security framework as a potential industry standard for trusted agent execution.