Nvidia used its presence at COMPUTEX 2026 to outline a broader push into agentic AI, unveiling a stack that ties together hardware, robotics and model development. The company said the effort is aimed at helping businesses move AI agents from experimental projects into production systems.
The announcements reflect a wider shift in the market as companies increasingly look beyond chatbots and toward AI tools that can carry out multi-step tasks. Nvidia positioned its latest updates as infrastructure for that transition, covering the chips and systems that run AI, the software layers that support development, and the robotics platforms that bring AI into the physical world.
Nvidia framed the new stack as part of the next generation of AI factories, a term it uses for the infrastructure needed to build and deploy advanced AI systems at scale. The company has been emphasizing agentic AI in recent weeks, including new benchmarking and enterprise-focused partnerships that suggest it sees agents as a major commercial opportunity.
The hardware side remains central to that strategy. Nvidia continues to build around its Blackwell platform, which the company has recently promoted in training and infrastructure benchmarks. By linking agentic software to its latest compute systems, Nvidia is making a case that the performance demands of autonomous or semi-autonomous AI will require dedicated infrastructure rather than general-purpose cloud resources alone.
Nvidia also highlighted robotics as a key part of the stack. That aligns with the company’s broader messaging around physical AI, a category that includes systems capable of interacting with the real world through robots, vehicles and industrial machinery.
The company has already been expanding partnerships in this area. In recent days, it announced collaborations with LG Group and Doosan Group to advance physical AI and AI factory infrastructure. Those deals point to Nvidia’s effort to connect its AI platform with industrial automation, mobility and manufacturing use cases.
By bundling robotics into an agent-focused strategy, Nvidia is signaling that it sees AI agents as useful not only in software workflows but also in environments where machines must perceive, plan and act in the physical world.
The COMPUTEX announcements also fit into Nvidia’s broader work with model developers and infrastructure providers. On the model side, the company has been optimizing systems for local and cloud deployment, including recent work around Google DeepMind’s DiffusionGemma model. On the infrastructure side, Nvidia recently said its Blackwell platform performed strongly on MLPerf Training 6.0 and led the first agentic AI infrastructure benchmark from Artificial Analysis.
Taken together, the updates show Nvidia trying to shape both sides of the AI stack. The company is not only supplying the chips that power models, but also the systems, software and reference architectures intended to make agentic applications practical for enterprise customers.
The timing of the announcements suggests Nvidia wants to define the early market for AI agents before it becomes crowded. As companies search for ways to automate more complex work, Nvidia is pitching itself as the provider of the underlying platform needed to train, deploy and run those systems. Its message at COMPUTEX was clear: the next phase of AI will require more than models alone, and Nvidia intends to supply the full stack.