Nvidia is broadening its AI ambitions beyond graphics chips with a new software platform aimed at the infrastructure behind large-scale AI systems. At Computex in Taiwan, the company introduced DSX, a platform for designing, simulating, building and running what it calls AI factories.
The announcement comes as Nvidia pushes deeper into nearly every layer of the AI stack, from CPUs and servers to robotics and open models. The company framed DSX as a tool for operators that need to plan and manage major AI installations, an area that is becoming increasingly important as demand for compute and power grows.
Within DSX, Nvidia highlighted DSX MaxLPS, a system meant to improve power and cooling efficiency so more GPUs can run within the same power envelope. The company said the approach could allow operators to run 40% more GPUs on the same power budget. Nvidia also introduced DSX OS, which it says is intended to evolve into an open-source operating system for AI factories.
The software launch reflects how AI infrastructure has become a strategic focus for Nvidia, which has dominated the market for AI accelerators and is now extending its reach into the operational layer around them. The company is pitching AI factories as the next major form of industrial computing, where hardware, software and facility design are tightly linked.
One of Nvidia’s biggest announcements was Vera, a new CPU the company says is designed specifically for AI agents. Nvidia said the chip is built for agentic workloads such as using tools, writing code and processing data. It claimed the processor can handle those tasks 1.8 times faster than conventional x86 CPUs.
Chief executive Jensen Huang said in a statement that AI agents will become the largest users of computing and described Vera as the first CPU designed for that future. Nvidia also said a range of companies, including OpenAI, Anthropic, SpaceXAI, CoreWeave, Lambda, Dell, HPE and Lenovo, have signed on to use the chip.
The move is notable because it puts Nvidia into more direct competition with established CPU players such as Intel, ARM-based chip designers and companies like Apple and Qualcomm. For years, Nvidia’s strongest position has come from its GPUs and the CUDA software ecosystem. Now the company appears to be building its own CPU story around the next phase of AI growth.
Nvidia also introduced DGX Station for Windows, a desktop AI supercomputer that can run models of up to 1 trillion parameters locally. The company said the setup is intended to improve performance, security and cost efficiency for engineers, researchers, developers and enterprises.
On the robotics side, Nvidia announced Cosmos 3, a world model that can work across text, images, video, sound and actions. It also unveiled Alpamayo 2 Super, an open reasoning model for robotaxis, and H2 Plus, an open reference design for humanoid robots combining hardware, software and onboard compute.
The company said its Vera Rubin AI infrastructure platform, first shown at CES, is now moving into full production across 350 factories in 30 countries. Nvidia claimed the system delivers 10 times higher agent throughput than the previous Grace Blackwell generation.
Taken together, the announcements show Nvidia moving well beyond its roots as a chip supplier. The company is positioning itself as a full-stack AI provider, with products aimed at the CPUs, infrastructure and software that could power the next wave of agentic systems.