Something significant happened on June 1, 2026: NVIDIA announced a coordinated push into enterprise AI agents that involves 16 of the largest software platform providers in the world. This is not a single partnership or a narrow product update. It is a deliberate attempt to establish NVIDIA’s infrastructure as the foundation layer for autonomous agents running inside real businesses.

What NVIDIA Actually Released

The core product is the NVIDIA Agent Toolkit, software that equips enterprises to build agents capable of working alongside employees at scale. It includes:

• NemoClaw blueprints: Pre-built agent frameworks designed for specific enterprise workflows
• Nemotron open models: Foundation models optimised for agentic reasoning and long-running tasks
• OpenShell secure runtime: A secure execution environment that ensures agents run safely and under user control
• CUDA-X libraries with agent skills: Specialised capabilities agents can call on for complex tasks

The headline new model is Nemotron 3 Ultra, a smaller and faster version built specifically for long-running agents. NVIDIA claims it delivers 5x faster inference and up to 30% lower cost per task compared to prior approaches. For businesses thinking about the operational economics of running agents at scale, that gap matters.

Who Is Building With It

The list of companies adopting the NVIDIA Agent Toolkit reads like a who’s-who of enterprise software: Adobe, Atlassian, Amdocs, Box, Cadence, Cisco, Cohesity, CrowdStrike, Dassault Systemes, IQVIA, Red Hat, SAP, Salesforce, Siemens, ServiceNow, and Synopsys.

These are not pilot programs or press-release partnerships. Each company is integrating NVIDIA’s infrastructure into its own agent offerings:

Engineering automation: Cadence, Dassault Systemes, Siemens and Synopsys are building autonomous AI engineers using NemoClaw that execute simulation and verification workflows. The claim is that weeks of engineering work can be compressed into hours. For industries like aerospace, automotive and chip design, that represents a meaningful shift in how technical work gets done.

Cybersecurity and operations: CrowdStrike and Palantir are building long-running agents powered by Nemotron open models for cybersecurity threat detection and operational decision-making. Agents that can monitor, reason, and act continuously are a different category of security tool than anything available a year ago.

Windows integration: Microsoft and NVIDIA are partnering on new Windows security primitives and the NVIDIA OpenShell runtime, ensuring agents running on Windows machines do so safely and under full user control. This is significant for enterprise IT teams worried about rogue agents accessing systems they should not.

The Governance Question

There is a tension running through all of this. An OutSystems survey published the same week found that 96% of organisations are already using AI agents in some capacity, but 94% report concern that AI sprawl is increasing complexity, technical debt, and security risk. More agents means more surface area to manage.

NVIDIA’s OpenShell runtime and the Microsoft security primitives partnership are direct responses to that concern. The pitch is that NVIDIA is not just building faster agents, it is building a governed platform where agents run within known boundaries.

Whether enterprise IT teams will trust that framing is a separate question, but the market signal is clear: the companies selling agents now understand that governance is part of the product.

What This Means for Business

If you are evaluating AI agents for your business, this announcement reshapes the procurement landscape in a few ways.

Vendor consolidation is coming. When Adobe, SAP, Salesforce and ServiceNow all build on the same underlying infrastructure, it becomes easier to connect agents across systems. That interoperability has been one of the biggest practical barriers to multi-agent deployments.

Cost economics are shifting. Nemotron 3 Ultra’s claimed 30% cost reduction for agentic tasks means the economic case for running agents continuously (rather than on-demand) becomes more viable. Businesses that have been waiting for the unit economics to make sense have fewer reasons to wait.

Security is now table stakes. The emphasis on governed runtimes and security primitives signals that the market has moved past “build fast and fix later.” Enterprises that deploy agents without a governance framework in place will find themselves at odds with their own software vendors’ standards.

The infrastructure layer matters. Just as cloud computing consolidated around AWS, Azure and Google Cloud, enterprise AI infrastructure is beginning to consolidate. NVIDIA’s position in this stack is becoming as relevant to AI agent deployments as GPU availability was to model training.

For business leaders thinking about AI workforce strategy, the message is straightforward: the tools are maturing, the major platforms are committing, and the window for early-mover advantage in agent-led operations is narrowing. The question is no longer whether agents will run your workflows but how well governed and how cost-efficient they will be when they do.

---

If you want help thinking through what an AI agent workforce could look like for your business, the team at Enterprise DNA’s Omni practice works with companies on exactly this. Book a discovery call to explore what’s possible.