Nvidia used its Computex appearance in Taiwan to unveil a wide-ranging set of products that pushed the company further beyond graphics chips and deeper into CPUs, infrastructure, and robotics. The announcements included a new CPU tailored for AI agents, a robotics-focused world model, updated software for AI factories, and new reference designs for humanoid systems.
The biggest attention may have gone to Vera, Nvidia’s new CPU. The company says the chip is built for agentic workloads such as tool use, coding, and data processing. Nvidia also said the processor is intended to handle those tasks about 1.8 times faster than traditional x86 CPUs. Chief executive Jensen Huang described AI agents as the largest future users of computing and said Vera was designed for that shift.
Nvidia said several major companies have already lined up to use the chip, including OpenAI, Anthropic, SpaceXAI, CoreWeave, Lambda, Dell, HPE, and Lenovo. The move signals a more direct challenge to established CPU vendors, even as Nvidia remains best known for its dominant position in AI accelerators.
Alongside the CPU announcement, Nvidia introduced Cosmos 3, a new world model aimed at physical AI and robotics. The company said the model can work across text, images, video, sound, and actions, which Nvidia sees as important for systems that interact with the physical world rather than only digital data.
Nvidia also announced Alpamayo 2 Super, an open reasoning model intended for robotaxis. In addition, it introduced H2 Plus, an open reference design for humanoid robots that combines hardware, software, and onboard compute. The releases suggest Nvidia is building a broader toolset for developers working on autonomous vehicles and robots that need to perceive, reason, and act in complex environments.
The robotics push comes as Nvidia continues to position its platform as a foundation for what it calls physical AI, a category that links model development with sensors, movement, and real-world decision-making.
Nvidia also expanded its infrastructure lineup with DSX, a software platform for designing, simulating, building, and running AI factories. One component, DSX MaxLPS, is designed to improve power and cooling efficiency so customers can run more GPUs within the same power budget. Another, DSX OS, is being promoted as an open-source operating system for AI factories.
For local development and enterprise use, Nvidia unveiled DGX Station for Windows. The system is designed to run models of up to 1 trillion parameters on a desktop machine, which Nvidia says can improve performance, security, and cost efficiency for engineers, researchers, developers, and businesses.
The company also said its Vera Rubin infrastructure platform has moved into full production across 350 factories in 30 countries. Nvidia claims the platform, which combines GPUs, CPUs, networking, storage, and security, delivers 10 times the agent throughput of the prior Grace Blackwell generation.
Taken together, the announcements show Nvidia continuing to expand into multiple layers of the AI market. Rather than relying only on GPU demand, the company is now targeting the compute, software, and systems expected to support agentic AI and robotics in the years ahead.