NVIDIA announced an enterprise AI agent stack built around Nemotron models, Agent Toolkit software, OpenShell runtime and CUDA-X agent skills, with partners including Cadence, Siemens, Microsoft, Canonical and Red Hat.
NVIDIA announced a set of models and software tools aimed at helping enterprises build autonomous AI agents for coding, research, engineering and other business workflows.
In a press release, NVIDIA said its new enterprise AI agent offering includes Nemotron models, Agent Toolkit software, NemoClaw blueprints, the OpenShell runtime and CUDA-X agent skills. The company positioned the package as infrastructure for building and deploying autonomous enterprise AI agents across software, industrial and engineering use cases.
NVIDIA said the portfolio includes Nemotron 3 Ultra, a 550-billion-parameter mixture-of-experts open model designed for coding, research and enterprise agent workflows. According to NVIDIA’s investor release, the company claims Nemotron 3 Ultra can deliver up to five times faster inference and up to 30% lower cost compared with comparable open frontier models. Those performance and cost figures are NVIDIA’s claims, and the company did not provide independent third-party benchmark validation in the supplied release excerpts.
The NVIDIA Newsroom release also described Agent Toolkit software, NemoClaw blueprints, OpenShell runtime and CUDA-X agent skills as components for creating autonomous enterprise AI agents. NVIDIA said the system is intended to support governed agent behavior, domain-specific skills and deployment on enterprise infrastructure.
NVIDIA named several enterprise software and infrastructure companies as partners, including Cadence, Siemens, Microsoft, Canonical and Red Hat, according to the NVIDIA Newsroom announcement. The company framed the partnerships as part of a broader effort to embed AI agent capabilities into business software and technical workflows.
Cadence separately said its ChipStack AI Super Agent is powered by NVIDIA Nemotron models and secured with NVIDIA OpenShell. In its announcement, Cadence described ChipStack as a fully autonomous virtual engineer for chip design workflows. Cadence said the product is intended to support governed autonomy in semiconductor design, a field where engineering workflows often involve complex verification, simulation and optimization steps.
The Cadence announcement is notable because chip design is a high-value, specialized domain where AI assistants are being tested not only for code generation but also for technical decision support. However, the supplied Cadence excerpt does not include independent performance data for ChipStack, so claims about its impact should be treated as company statements rather than externally verified results.
The announcements show NVIDIA continuing to move beyond selling GPUs and into higher-level enterprise AI software. By packaging open models, runtime controls, blueprints and domain skills, NVIDIA is trying to give enterprises more of the stack needed to build agentic systems.
The emphasis on governance and runtime controls also reflects a practical concern for enterprise AI adoption. Companies experimenting with autonomous agents need tools that can constrain actions, integrate with existing systems and provide operational oversight. NVIDIA’s OpenShell runtime and related software are presented by the company as part of that control layer.
At the same time, the releases leave several questions unanswered. NVIDIA’s claims about faster inference and lower cost for Nemotron 3 Ultra are not accompanied in the provided material by detailed benchmark methodology. The announcements also do not specify how widely the partner integrations are deployed in production or what measurable productivity gains customers have achieved.
For now, the news is best understood as a major platform push: NVIDIA is promoting Nemotron models and supporting software as a foundation for enterprise AI agents, while partners such as Cadence are presenting early domain-specific applications built on that stack.
NVIDIA announced a set of models and software tools aimed at helping enterprises build autonomous AI agents for coding, research, engineering and other business workflows.
The company positioned the package as infrastructure for building and deploying autonomous enterprise AI agents across software, industrial and engineering use cases.
NVIDIA said the portfolio includes Nemotron 3 Ultra, a 550 billion parameter mixture of experts open model designed for coding, research and enterprise agent workflows.
Continue reading
