At NVIDIA GTC 2026, Jensen Huang outlined how NVIDIA is building the foundation for the next phase of artificial intelligence, where AI factories, autonomous agents, and physical robots will become core components of the global computing infrastructure.
At this year’s NVIDIA GTC, Jensen Huang delivered a sweeping vision of how artificial intelligence is evolving beyond software models into the foundation of a new computing era. According to NVIDIA, the industry is entering a phase where AI infrastructure, autonomous agents, and robotics will reshape how computing systems are built and deployed worldwide.
For much of the past decade, progress in artificial intelligence has been driven by training – the process of teaching AI models using massive datasets so they can recognize patterns in language, images, or video. But Huang argued that the industry is now shifting its focus toward inference, the stage where AI models actually perform tasks and generate outputs when used by people or businesses.
If training is comparable to educating a specialist, inference is when that specialist starts working. And at global scale, inference demand can far exceed training workloads. Every prompt to a chatbot, every AI-generated image, and every automated decision triggered by enterprise software requires inference processing in real time.
What makes this shift even more significant is the emergence of reasoning AI. Unlike earlier AI systems that simply predicted the next word or image, newer models can break complex problems into multiple steps, analyze intermediate results, and refine their conclusions before producing an answer. This multi-step reasoning dramatically increases the computational load behind each query.
To support this surge in AI workloads, NVIDIA continues to push the boundaries of accelerated computing with new GPU architectures such as NVIDIA Blackwell and the next-generation roadmap platform NVIDIA Rubin. These chips are designed not only for training massive models but also for running inference at unprecedented scale.
Huang described this shift as the birth of AI factories – a new type of data center optimized specifically to generate intelligence. In traditional data centers, servers primarily store and process enterprise data. In an AI factory, however, the output is not just processed information but tokens, the fundamental units of AI-generated language, images, and digital content.
In this model, data centers effectively become production facilities for intelligence, transforming raw data and compute power into AI-driven services and applications.
Another major theme of the keynote was the rise of AI agents – software systems capable of autonomously performing tasks on behalf of users. Unlike conventional chatbots, AI agents can plan actions, use tools, access databases, and coordinate with other systems to complete complex workflows.
To accelerate this shift, NVIDIA introduced the open-source platform NemoClaw reference OpenClaw, designed to help developers build agent-based AI systems that can operate securely within enterprise environments. These agents can interact with internal data, execute software commands, and communicate with external systems while remaining governed by corporate security policies.
Huang suggested that in the near future, companies may operate thousands – or even millions – of AI agents working alongside human employees. Entire business processes could be automated through networks of intelligent software entities capable of reasoning and decision-making.
Beyond digital intelligence, NVIDIA also emphasized the growing importance of physical AI, where artificial intelligence interacts directly with the physical world through robots and autonomous machines.
Training robots presents a unique challenge: the real world is chaotic and unpredictable, filled with edge cases that are difficult to capture through real-world data alone. Collecting enough physical-world data to train robots for every possible scenario is nearly impossible.
To address this challenge, NVIDIA is investing heavily in simulation technologies that can generate vast amounts of synthetic data. Platforms such as NVIDIA Isaac Lab enable developers to train robots in highly detailed virtual environments before deploying them in the real world. Meanwhile, NVIDIA Omniverse allows engineers to build digital twins of factories, warehouses, and even cities where robots can learn to operate safely and efficiently.
Using these simulation systems, developers can expose robots to millions of virtual scenarios – from navigating cluttered environments to manipulating delicate objects – dramatically accelerating the training process.
One of the more surprising moments in the keynote came from NVIDIA’s collaboration with Disney Research. The partnership is exploring how AI-driven robotics could bring lifelike characters into future theme parks. These robots could move naturally, interact with visitors, and adapt to real-world environments thanks to advanced AI reasoning and physics-based simulation.
Taken together, the announcements at GTC 2026 highlight NVIDIA’s ambition to expand far beyond GPUs. The company is positioning itself as the architect of the infrastructure that will power the AI economy – from hyperscale data centers and enterprise software platforms to autonomous robots operating in the physical world.
If Huang’s vision materializes, the next decade of computing may be defined not simply by faster chips, but by an interconnected ecosystem of AI factories, intelligent agents, and physical machines – all powered by accelerated computing./.