During the latter half of his GTC 2026 keynote, Nvidia CEO Jensen Huang shifted the focus from AI factories to the next frontier of artificial intelligence: enterprise software, agentic systems, and physical robotics.
The central theme of Huang's address marked a pivot in how the industry evaluates AI—moving away from asking “how intelligent is the model” and toward “can it actually act on your behalf.”
Huang devoted considerable time to OpenClaw, describing it as a next-generation computing interface. The significance of open-source agentic software, he explained, is that users will increasingly delegate tasks to AI agents rather than issuing direct commands to traditional software.
From file management and calendar scheduling to cross-application task execution, AI agents are moving beyond simple chat interfaces into genuinely operational workflows.
Over the past decade, enterprises have competed on cloud adoption and software-as-a-service (SaaS) integration. Going forward, competition will center on who can most quickly translate internal knowledge, permissions, and governance rules into systems that AI agents can execute. In this framework, software is no longer just a tool operated by humans; it is becoming a unit of digital labor capable of completing tasks autonomously.
Security, governance, and the enterprise reality
Huang acknowledged the central challenge of this transition: security and governance. He highlighted capabilities that enterprises prioritize, including private deployment, policy engines, network guardrails, and privacy routers.
NVIDIA's ambition is to combine open agentic capabilities with enterprise-grade toolchains. For businesses—particularly Taiwanese enterprises closely watching this shift—the question is no longer whether a model can answer a query, but whether an agent can be deployed safely, reliably, and with proper oversight.
The software logic of the SaaS era is being rewritten. Instead of purchasing traditional customer relationship management (CRM) or enterprise resource planning (ERP) software, companies may soon purchase AI agents equipped with defined roles, memory, permissions, and workflows. A software's value will be measured not by its feature count, but by how effectively it completes real work within enterprise constraints.
（Related：
GTC 2026: Nvidia CEO Reframes Company's Future From Chips to AI Factories
｜
Latest
）
Bridging the digital and physical worlds
Moving beyond enterprise workflows, Huang connected OpenClaw and digital agents to open models and physical AI, specifically robotics and autonomous vehicles. NVIDIA anticipates that agents will not remain confined to screens. As models learn to perceive, reason, plan, and execute, they will naturally integrate into robotic arms, vehicle systems, and factory floors.
Huang identified the greatest bottleneck for physical AI: the high cost, danger, and limited coverage of real-world data, particularly for rare, long-tail scenarios. The solution lies in simulation, world models, and synthetic data pipelines that recreate the physical world within controllable training environments. Moving AI into the physical world requires an entire ecosystem for data generation, simulation-based validation, and structured deployment.
A unified technological backbone
Looking at the keynote as a whole, Huang is advancing a unified technological backbone. By using open models to develop capabilities and scaling them through simulation and inference systems, AI can evolve to reason and act in real-world environments.
You've read it. Now let's talk. Follow us on X. Editor: Chase Bodiford
