Nvidia is widening its ambitions beyond graphics chips with a new desktop system that can run large AI models locally on Windows machines. The company introduced DGX Station for Windows at Computex in Taiwan, positioning it as a tool for engineers, researchers, developers and enterprises that want access to substantial AI compute without relying entirely on the cloud.
The new desktop system is designed to run AI models with up to 1 trillion parameters locally. Nvidia says that approach can improve performance while also offering security and cost benefits. The announcement adds to a larger set of product launches that show the chipmaker pushing deeper into the full AI stack, from processors and software to infrastructure and robotics.
DGX Station was one of several announcements Nvidia made at Computex. The company also unveiled Vera, a CPU it says is purpose-built for agentic workloads. Nvidia described those workloads as tasks involving tool use, code writing and data processing. The company said Vera can handle those jobs 1.8 times faster than conventional x86 CPUs, according to its own claims.
Jensen Huang, Nvidia’s chief executive, framed the move as a bet on the next major class of AI users. In a statement, he said AI agents will become the largest consumers of computing and that Vera was built for that future. Nvidia said the chip has already drawn interest from companies including OpenAI, Anthropic, SpaceXAI, CoreWeave, Lambda, Dell, HPE and Lenovo.
The CPU announcement is notable because it places Nvidia in territory long dominated by firms such as Intel, Arm, Apple and Qualcomm. It also underscores how the company is expanding beyond its core graphics processing business as AI demand reshapes the market.
Nvidia also introduced DSX, a software platform aimed at planning and operating AI factories. The platform includes DSX MaxLPS, which Nvidia says is designed to optimize power and cooling so operators can fit 40% more GPUs within the same power budget. Another component, DSX OS, is being pitched as an open-source operating system for AI factories.
The company’s broader message is that AI infrastructure now needs to be treated like a specialized industrial system. Rather than simply selling chips, Nvidia is offering tools for designing, simulating and running the facilities that house those chips.
Beyond compute hardware, Nvidia announced new models and reference designs for robotics and autonomous systems. These include Cosmos 3, a world model that can work across text, images, video, sound and actions, as well as Alpamayo 2 Super, an open reasoning model for robotaxis. Nvidia also introduced H2 Plus, a reference design for humanoid robots that combines hardware, software and onboard compute.
The company said its Vera Rubin platform, which combines GPUs, CPUs, networking, storage and security, is moving into full production across 350 factories in 30 countries. Nvidia claims the system delivers 10 times higher agent throughput than its Grace Blackwell generation.
Taken together, the announcements show Nvidia extending its reach across almost every layer of AI development. The company, once known primarily for GPUs, is now staking a claim as a supplier of CPUs, desktop systems, factory software and robotics infrastructure as demand for AI agents and other advanced models continues to grow.