At GTC Taipei, NVIDIA unveiled a sweeping lineup built for the age of AI agents, including Vera CPUs for AI factories, RTX Spark Windows PCs, DGX Station for enterprise desks, Vera Rubin data-center systems and an open humanoid robot platformDennis Bihler|NVIDIA used GTC Taipei to lay out one of its broadest visions yet for the next phase of artificial intelligence: not just faster chips, but an entire computing stack designed for AI agents, from hyperscale data centers and enterprise workstations to personal Windows PCs and humanoid robots.The announcements point to a clear strategic shift. NVIDIA is no longer positioning AI mainly as a cloud workload powered by graphics processors. Instead, the company is preparing for a world in which AI agents write code, use tools, run tasks, reason through workflows, control software and eventually operate physical machines. To make that possible, NVIDIA is trying to own more of the infrastructure around those agents, including CPUs, networking, security, personal computers and robotics platforms.4 View gallery NVIDIA and Microsoft reinvent Windows PCs for the age of personal AI (Photo: Nvidia)At the center of the data-center push is NVIDIA Vera, which the company described as its first CPU built specifically for AI agents. Now in full production, Vera is designed to handle the CPU-heavy parts of agentic AI workloads, including Python runtimes, sandboxed code execution, orchestration logic, analytics pipelines and data processing. NVIDIA said Vera enables 1.8 times faster task completion compared with x86 CPUs across workloads such as agentic AI, reinforcement learning and data processing.“AI agents will be the largest users of computing,” said NVIDIA founder and CEO Jensen Huang. “Vera is the first CPU designed for that future, built to run agentic AI at hyperscale with extraordinary performance, efficiency and programmability.”The company said customers exploring Vera include NYSE, Anthropic, OpenAI, SpaceXAI, ByteDance, CoreWeave, Lambda, Nebius, Nscale and Oracle Cloud Infrastructure. System makers including Dell Technologies, HPE, Lenovo and Supermicro are also integrating it into AI infrastructure.For NVIDIA, Vera is meant to solve a bottleneck that has become more visible as AI moves from simple chatbot responses to multi-step agents. When an agent responds to a prompt by searching, calling tools, executing code, retrieving data, evaluating results and generating a final response, the GPU is only part of the story. The surrounding CPU work can slow the entire system. Vera, built around NVIDIA’s custom Olympus CPU core, includes 88 cores and a memory subsystem delivering up to 1.2TB per second of bandwidth, helping keep accelerators fed and reducing time spent waiting on CPU-bound steps.That CPU is also part of NVIDIA’s larger Vera Rubin platform, which the company said is ramping into full production for what it calls agentic AI factories. Vera Rubin is a rack-scale system built for large AI labs, cloud providers and hyperscalers. It combines Vera CPUs, Rubin GPUs, BlueField-4 infrastructure processors, storage and Spectrum networking into a five-rack platform that operates as one large AI supercomputer.4 View gallery Vera, the CPU for agents (Photo: Nvidia)NVIDIA said Vera Rubin delivers 10 times the agent throughput at scale compared with the previous-generation Grace Blackwell platform. It is being manufactured through a global supply chain that includes more than 350 factories in 30 countries, with 150 partners in Taiwan alone.“Agentic AI is a new kind of workload. One prompt can launch a thousand-step journey of reasoning, retrieval, tool use and response generation,” Huang said. “Vera Rubin was built for this moment, an AI factory engine that delivers intelligence at scale, with the performance, efficiency and security needed to power the next industrial revolution.”The scale of these systems also requires new networking. NVIDIA said Vera Rubin introduces Spectrum-X Ethernet Photonics, a co-packaged optics switching technology now in production and designed to support million-GPU AI factories. The company said the technology offers better power efficiency, longer AI uptime and faster deployment than networks using traditional transceivers.Security is another major theme. As AI factories process proprietary data, regulated content and mission-critical models, NVIDIA is emphasizing confidential computing and hardware-level protection. Vera Rubin is designed with full-stack NVIDIA Confidential Computing at rack scale, encrypting data across high-speed interconnects and using BlueField-4 and DOCA software for multi-tenant isolation, zero-trust policy enforcement, runtime threat detection and end-to-end encryption.The same security logic carries into NVIDIA’s Windows push. Together with Microsoft, the company unveiled RTX Spark, a new superchip for Windows PCs designed for personal AI agents. NVIDIA said RTX Spark-powered laptops and compact desktops will be available this fall from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI, with Acer and GIGABYTE models to follow.RTX Spark is intended to turn the PC from a machine that runs apps into a device that can host private, local agents. The chip combines a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores, connected to a 20-core Grace CPU. NVIDIA said it offers up to 1 petaflop of AI performance and up to 128GB of unified memory.“The PC is being reinvented,” Huang said. “For forty years, you launched apps. Click. Type. With RTX Spark and Microsoft Windows, you ask, and the PC does the work. This is the new PC. The personal AI computer.”The key challenge for local AI agents is not only performance, but trust. Agents that can open applications, search files, generate code, send information to models or act across workflows need stronger controls than a standard app. NVIDIA and Microsoft said RTX Spark will use new Windows security primitives and NVIDIA OpenShell, a runtime designed to let users define what agents can and cannot do, route sensitive queries to local models and mask personal information when cloud models are used.NVIDIA is pitching RTX Spark to creators, AI developers and gamers as well. The company said users will be able to render 90GB 3D scenes, edit 12K video, generate 4K AI video, run 120-billion-parameter language models with up to 1 million tokens of context and play AAA games at 1440p and more than 100 frames per second. Adobe is also rearchitecting Photoshop and Premiere for RTX Spark, with NVIDIA saying the changes will deliver up to twice the AI and graphics performance across creative workflows.For enterprises, NVIDIA and Microsoft are also bringing a much larger system to Windows: DGX Station for Windows, a deskside AI supercomputer built on the GB300 Grace Blackwell Ultra Desktop Superchip. NVIDIA said the system, expected in the fourth quarter from ASUS, Dell Technologies, GIGABYTE, HP, MSI and Supermicro, will be capable of running AI models of up to 1 trillion parameters locally.4 View gallery NVIDIA DGX Station (Photo: Nvidia)The system is designed to bridge a gap between data-center AI infrastructure, which has traditionally run on Linux, and the Windows environment where many companies already run design, engineering, research and productivity workflows. DGX Station for Windows includes a 72-core Grace CPU, Blackwell Ultra GPU, up to 748GB of coherent memory and up to 20 petaflops of FP4 performance. It can also be paired with an RTX PRO 6000 Blackwell Workstation GPU for visualization and simulation.NVIDIA said DGX Station can run hundreds of agents simultaneously and connect them directly to enterprise applications and workflows. It also supports Windows Subsystem for Linux, giving organizations access to Linux AI toolchains while keeping Windows security, management and compliance controls.The final piece of NVIDIA’s GTC Taipei push moves AI from screens and data centers into the physical world. The company announced the Isaac GR00T Reference Humanoid Robot, an open humanoid robot reference design for academic research. The platform combines a Unitree H2 Plus humanoid robot, Sharpa five-fingered hands, Jetson Thor onboard compute and NVIDIA’s Isaac GR00T software and models.The goal is to give researchers a more unified platform for humanoid development, bringing together hardware, data capture, simulation, training, evaluation and deployment. The robot stands nearly 6 feet tall, weighs 150 pounds, has 31 degrees of freedom across the body and reaches 75 degrees of freedom when including the hands. Its Jetson AGX Thor T5000 onboard computer includes a Blackwell GPU, 14-core Arm CPU and 128GB of unified memory for real-time inference and control.4 View gallery The NVIDIA Isaac GR00T Reference Humanoid Robot (Photo: Nvidia)“Humanoid robots will bring physical AI to the world’s largest industries, opening a multitrillion-dollar economic opportunity,” Huang said. “The NVIDIA Isaac GR00T Reference Humanoid Robot gives researchers a single, open platform to make breakthrough discoveries toward general-purpose physical intelligence.”Leading research institutions including Ai2, ETH Zurich, Stanford Robotics Center and UC San Diego’s Advanced Robotics and Controls Laboratory are expected to use the platform. The reference humanoid robot will be available from Unitree in late 2026, while a reference workflow for the Unitree G1 is expected soon on GitHub and Hugging Face.Taken together, the announcements show NVIDIA trying to define the next AI platform before the market fully takes shape. Its bet is that agents will not live in one place. They will run in massive AI factories, on enterprise desks, inside consumer PCs and eventually in robots that act in the real world.That vision is still early, and much depends on whether developers, companies and users adopt agents deeply enough to justify the new hardware. But NVIDIA’s message from Taipei was unmistakable: the AI boom is moving beyond model training, and the next race is to build the machines that let AI act.