--- 格式版本: 2 标题: "NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI | NVIDIA Blog" 原文链接: "https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news" 发布日期: "2026-05-21" 发布时间校准状态: "found" 发布时间需复核: "否" 发布时间来源: "rule:local:strict_original_metadata" 发布时间证据: "article:published_time: 2026-05-21T16:00:17+00:00" 发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_metadata" 发布时间校准置信度: "high" 发布时间候选数量: 12 发布时间严格候选数量: 4 发布时间原页读取状态: "原页面已读取" 发布时间未找到原因: "" 发布时间校准时间: "2026-07-22T17:31:40+08:00" 发布时间仲裁状态: "skipped" 发布时间仲裁尝试次数: 0 发布时间仲裁耗时毫秒: 0 发现时间: "2026-06-16T05:48:44+08:00" 入库时间: "2026-07-22T09:31:41.115Z" 来源平台: "NVIDIA Blog 搜索" 搜索渠道: "source_template" 搜索词: "https://blogs.nvidia.com/?s=Rack%20level%20server" 匹配关键词: - "Rack level server" 相关厂家: - "NVIDIA" 相关专家: [] 内容类型: "网页" 抓取工具: "XCrawl Scrape" 清洗工具: "XCrawl Markdown + LLM 正文裁剪（重试成功）" 原始附件: [] AI优质: "否" AI打分: 25 AI分档: "非优质" AI质检状态: "不通过" AI打分理由: "正文为NVIDIA GTC大会滚动报道，聚焦Nemotron模型、智能体、机器人等AI软件层面，未涉及超节点/AI Rack/机柜级系统、硬件架构、供电散热、互连或量产落地等硬件基础设施核心主题，与项目关注范围无关。" AI质检模型: "deepseek-v4-flash" AI质检时间: "2026-07-24T17:54:55+08:00" AI主题相关性: 0 AI来源权威性: 15 AI新颖性: 5 AI技术细节: 0 AI商业部署信号: 0 AI完整性: 5 采集批次: "2026年6月16日5点48分29秒" 采集批次ID: "20260616-054829-217" 去重键: "https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news" --- # NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI Rolling coverage on the conference, including NVIDIA founder and CEO Jensen Huang’s keynote, event highlights, live demos and on‑the‑ground color. May 21, 2026 by [NVIDIA Writers](https://blogs.nvidia.com/blog/author/nvidiawriters/ "Posts by NVIDIA Writers") ![图片](https://blogs.nvidia.com/wp-content/uploads/2026/05/26gtc-tpe-keynote-GM_05466-1-1280x720.jpg) At NVIDIA GTC Taipei at COMPUTEX, the world’s developers, researchers and industry leaders are converging to dive into the latest breakthroughs shaping every industry, covering topics spanning AI factories and scaling infrastructure to agentic and physical AI and more. This is the place to find all the latest — stay tuned to the blog for live updates. **[Jump to the keynote recap](https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news/#keynote).** * * * _Thursday, June 4, 6:00 a.m. PT **[![🔗](https://s.w.org/images/core/emoji/17.0.2/svg/1f517.svg)](https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news/#nemotron-3-ultra)**_ ## **NVIDIA Nemotron 3 Ultra Is Live, With Early Adopters Perplexity, Palantir and ServiceNow Powering Long-Running AI Agents** ![图片](https://blogs.nvidia.com/wp-content/uploads/2026/06/Copy-of-press-agentic-ai-nemotron-3-ultra-launch-corp-blog-1920x1080-1.jpg) NVIDIA today released Nemotron 3 Ultra, an open model built for long-running agents with contributions from the Nemotron Coalition. Models powering long-running agents do more than generate text. They interpret information, plan next steps, call tools, evaluate results and iterate across turns to complete complex coding, research and enterprise tasks. This requires efficient models that can explore more of the search space in less time to deliver higher-accuracy results faster. Nemotron 3 Ultra is built for that new workload. It’s a frontier smart model that delivers up to 5x faster inference and lowers the cost of complex agentic tasks by up to 30%. This enables agents to finish the same job in less time or complete more jobs in the same time. ![图片](https://blogs.nvidia.com/wp-content/uploads/2026/06/Nemotron-3-Ultra-AA-Quadrant-Chart-scaled.png) Nemotron 3 Ultra, a 550-billion-parameter [mixture-of-experts](https://www.nvidia.com/en-us/glossary/mixture-of-experts/) model, handles the orchestration and hardest reasoning calls in an autonomous workflow: architectural decisions in long-running coding sessions, synthesis across hundreds of research sources and verification across thousands of interdependent constraints. Enterprise software leaders are building agents with the new model, including for workflows spanning software development, deep research, customer service and enterprise automations. * [Aible](https://tinyurl.com/AibleClaw-NemotronUltra) is integrating Nemtoron 3 Ultra into the AIbleClaw platform, allowing its customers to build secure long-running agents at scale for various domains. * [Glean](https://www.glean.com/press/glean-adds-support-for-nvidia-nemotron-3-ultra-expanding-model-choice-for-cost-effective-enterprise-ai) is making Nemotron 3 Ultra available in its model-agnostic agent harness, alongside an agentic search model fine-tuned with Nemotron 3 Nano, expanding enterprises’ access to cost-effective, enterprise agentic AI. * Greptile is integrating Nemotron 3 Ultra into its code review platform for codebase indexing, enabling code reviews with leading accuracy at lower cost. * Harvey is enabling support for Nemotron 3 Ultra and post-trained versions of the model through its platform, helping customers build and deploy AI-powered legal workflows with greater control over their data. * Perplexity is using Nemotron 3 Ultra for search and Perplexity Computer, and using its agent router to direct workloads to fine-tuned open models or proprietary models based on the task, helping its AI assistants operate with speed, efficiency and scale. Announced [earlier this week](https://nvidianews.nvidia.com/news/enterprise-software-leaders-build-ai-agents-with-nvidia), CrowdStrike and Palantir are adopting Nemotron 3 Ultra to enable a new class of long-running AI agents to help teams analyze complex data, coordinate tasks and streamline operations across cybersecurity and enterprise environments. Additional companies adopting the model include [Applied Compute](https://www.appliedcompute.com/research/agentic-model-router), [CodeRabbit](https://www.coderabbit.ai/blog/coderabbit-supports-nvidia-nemotron-3-ultra), Dataiku and ServiceNow. The model is trained on agent traces and optimized for agent harnesses, enabling developers to choose their preferred frameworks while maintaining accuracy. Agent platforms and harnesses including [BlackBox AI](https://www.blackbox.ai/blog/nemotron-on-blackbox), Cline, [Factory AI](https://docs.factory.ai/cli/user-guides/choosing-your-model), [Hermes Agent](https://developer.nvidia.com/blog/deploy-self-evolving-agents-for-faster-more-secure-research-with-a-hermes-agent-and-nvidia-nemoclaw), [Kilo Code](https://blog.kilo.ai/nvidia-nemotron-3-ultra), LangChain Deep Agents, OpenClaw, OpenCode, OpenHands and Pi support the new Nemotron models. Nemotron 3 Ultra works with the [NVIDIA NemoClaw](https://www.nvidia.com/en-us/ai/nemoclaw/?ncid=pa-srch-goog-984177&_bt=804567865336&_bk=nvidia%20nemoclaw&_bm=p&_bn=g&_bg=197993095849&gad_source=1&gad_campaignid=23744621431&gbraid=0AAAAAD4XAoGNA_zzTwhZS-06KDdmP-e2x&gclid=Cj0KCQjwlLDQBhDjARIsAPlIefEhntGjCrWijw9hXN1Y-L5LsxzxUJPFwENjXhSQ8IT6uRggd-160JgaAud9EALw_wcB) blueprint, which provides enterprises with a secure runtime, open models and domain-specific skills to put autonomous agents to work at scale. ### **H Company, Naver, Nous and Prime Intellect Join Nemotron Coalition** [H Company](https://hcompany.ai/h-joins-nemotron-coalition), NAVER Cloud, Nous Research and [Prime Intellect](http://primeintellect.ai/blog/nemotron-3) are joining the Nemotron Coalition. These members will contribute unique strengths spanning data, training environments, evaluation frameworks and domain expertise to support the collaborative development of an open frontier model trained on NVIDIA DGX Cloud, which will serve as the foundation for the upcoming Nemotron 4 family. By combining forces, the coalition is bringing together leading global AI labs and infrastructure providers to accelerate the development of open frontier models. This collaborative approach aims to broaden access to cutting-edge AI innovation while enabling developers and enterprises worldwide to build and customize models for their industries, regions and use cases. **New Nemotron Speech and Safety Models** Also live today, a new [Nemotron ](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-ASR-Streaming-Multilingual-0.6b)[speech recognition model](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-ASR-Streaming-Multilingual-0.6b) brings real-time streaming ASR to 40 language locales for voice agent workflows across global enterprise deployments. The [Nemotron 3.5 Content Safety model](https://huggingface.co/nvidia/Nemotron-3.5-Content-Safety) — a 4-billion-parameter open multimodal model — classifies content across 23 safety categories and a dozen languages, with support for custom enterprise policies. **Open and Customizable, Deployable Anywhere** Nemotron models are released with open weights, datasets and recipes, giving organizations transparency and control to customize models for domain-specific workflows and deploy them where their applications and data reside. Developers can use tools like [NVIDIA NeMo](https://www.nvidia.com/en-us/ai-data-science/products/nemo/) for customization, evaluation and optimization for their use cases. Because the Nemotron family of models is open, organizations can deploy them in environments that meet regulatory, sovereignty or data localization requirements. The models are available on [Hugging Face](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4), [ModelScope](https://modelscope.ai/collections/nv-community/Nemotron-3x), [OpenRouter](https://openrouter.ai/nvidia/nemotron-3-ultra-550b-a55b) and [build.nvidia.com](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b) as NVIDIA NIM microservices and through a broad ecosystem of [NVIDIA Cloud Partners](https://www.nvidia.com/en-us/data-center/gpu-cloud-computing/partners/), inference platforms and cloud service providers. * * * _Tuesday, June 2, 10:30 p.m. PT **[![🔗](https://s.w.org/images/core/emoji/17.0.2/svg/1f517.svg)](https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news/#build-a-claw)**_ ## **Build-a-Claw Comes to Taipei, Bringing Long-Running AI Agents** ![图片](https://blogs.nvidia.com/wp-content/uploads/2026/05/KEV_5040-1680x945.jpg) The Build-a-Claw experience has come to GTC Taipei — putting secure, long-running agent development directly into the hands of the APAC developer community. Build-a-Claw signals how rapidly the developer community and ecosystem are scaling agents. Starting with OpenClaw and Hermes Agent, attendees configured their claw’s persona, added agent skills and set its schedule. Then, they used [NVIDIA NemoClaw](https://www.nvidia.com/en-us/ai/nemoclaw/) blueprints and the [NVIDIA OpenShell](https://build.nvidia.com/openshell) runtime to deploy their agent safely and securely for their environment. Video Player Media error: Format(s) not supported or source(s) not found [Download File: https://blogs.nvidia.com/wp-content/uploads/2026/05/build-a-claw-13mb.mp4?_=1](https://blogs.nvidia.com/wp-content/uploads/2026/05/build-a-claw-13mb.mp4?_=1) 00:00 00:00 00:00 [Use Up/Down Arrow keys to increase or decrease volume.](javascript:void\(0\);) Claws, aka long-running agents, are a class of AI systems that go beyond mere prompt-answering. Unlike agents that complete a single task and vanish, claws persist: They work toward a goal, adapt when they hit obstacles, surface status updates and keep executing in the background even after a developer steps away. They’re the engine behind intelligent enterprise automation, agentic commerce and autonomous infrastructure — and building them right demands more than clever architecture. It demands a secure runtime. NVIDIA NemoClaw combines flexible support for agent harnesses — aka agent orchestration frameworks — with NVIDIA OpenShell as the secure runtime. It works with harnesses such as OpenClaw and Hermes Agent, giving developers a hardened, sandboxed foundation for claw development. OpenShell provides the runtime security boundary: isolating agent workloads, enforcing policy and keeping autonomous execution within guardrails that developers and their organizations can trust. NemoClaw blueprints further lower the barrier to entry by giving builders ready-to-adapt patterns for creating secure, enterprise-ready agents. * * * _Tuesday, June 2, 5:00 p.m. PT **[![🔗](https://s.w.org/images/core/emoji/17.0.2/svg/1f517.svg)](https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news/#isaac-gr00t)**_ ## **NVIDIA Isaac GR00T Accelerates Humanoid Robot Development From Data to Deployment** ![图片](https://blogs.nvidia.com/wp-content/uploads/2026/05/isaac-gr00t-cptx-rolling.png) Building [humanoids](https://www.nvidia.com/en-us/use-cases/humanoid-robots/) is complex, and progress often depends on how quickly teams can move through the full development loop: collect demonstrations, generate and refine data, train policies, test in simulation, validate the full software stack and deploy on real hardware. Developers today must handle many disconnected tools and handoffs across that workflow. Major updates to [NVIDIA Isaac GR00T](https://developer.nvidia.com/isaac/gr00t), an open, end-to-end development platform for humanoid robots, are accelerating that cycle. The platform unifies technologies including [Isaac Teleop](https://nvidia.github.io/IsaacTeleop/main/index.html), [Isaac Lab](https://developer.nvidia.com/isaac/lab), [Isaac Sim](https://developer.nvidia.com/isaac/sim), [Isaac ROS](https://developer.nvidia.com/isaac/ros), [GR00T open models](https://github.com/Nvidia/Isaac-GR00T) and [NVIDIA Jetson Thor](https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/) for real-time inference and control, giving developers a prescriptive way to move from data to deployment. Agility, Boston Dynamics, Dyna Robotics, Figure, FieldAI, Noble Machines, Richtech Robotics and Skild AI are using core components of NVIDIA’s humanoid robotics stack to accelerate robot development. The development flywheel is already gaining momentum. GR00T models have reached 274,000 downloads, while the [GR00T X Embodiment Sim dataset](https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim) has crossed more than 10 million downloads on Hugging Face. ### **Isaac GR00T Updates Speed Robot Development** [Isaac Teleop](https://github.com/NVIDIA/IsaacTeleop), now generally available, is an open source framework for real-time robot teleoperation and data capture across simulated and physical robots. It connects extended-reality headsets, gloves, motion trackers and other teleoperation devices to workflows that integrate with Isaac Lab, Isaac Sim, ROS 2 and Isaac ROS, reducing duplicated work. Leading teleoperation device makers such as PICO support Isaac Teleop natively, and robotics developers including Foxconn and Lightwheel are incorporating it into their training pipelines. The latest GR00T 1.7 model — pretrained on 20,000 hours of human egocentric data and built on Cosmos Reason 2 as its backbone — enables more complex bimanual and dexterous manipulation tasks, such as selecting a card from a stack and inserting it into a holder. Currently in early access, GR00T 1.7 is integrated with HuggingFace’s LeRobot and available under a commercial license so developers can build and deploy derived models beyond research settings. [Techman](https://www.tm-robot.com/en/company/news/Techman-Robot-Launches-End-to-End-Physical-AI-Development-Package) Robot uses the GR00T development platform and [GR00T 1.7 model](https://huggingface.co/nvidia/GR00T-N1.7-3B) to accelerate its development pipeline, bringing AI into real industrial use faster. Now part of the [OpenMDW-1.1 license from the Linux Foundation](https://vmblog.com/news/linux-foundation-releases-openmdw-1-1-nvidia-adopts-openmdw-for-cosmos-isaac-gr00t-ising-and-nemotron-ai-model-families/), future Isaac GR00T open model releases will be available under a single, model-centric license, making it easier for developers to build, customize and deploy GR00T model materials across robotics workflows. Enactic and Nexuni are integrating GR00T 1.7 to help robots reason, adapt and operate in unpredictable environments like nursing homes and laundry facilities. In addition, the Isaac Lab 3.0 Developer Preview expands robot learning with richer physics through Newton physics engine integration and multi-GPU scaling for large physical AI experiments. Developers can train policies against more realistic scenarios, including complex mechanisms, materials and environments. Unified actuator models across Isaac Lab and Isaac Sim help reduce mismatches between policy learning and software-in-the-loop testing, exposing issues before full-stack validation or hardware deployment. Flexion AG has achieved up to a 5x speedup on perceptive workload training using Isaac Lab for humanoid locomotion and manipulation policies. [Isaac Sim 6.0](https://github.com/isaac-sim/IsaacSim), now generally available, gives developers a simulation environment to validate robot behavior and test the full software stack before deployment. New agent skills help teams automate simulation workflows, while Newton authoring and software-in-the-loop testing let policies trained in Isaac Lab be evaluated against more realistic robot software and physics. The release also adds more than 1,000 simulation-ready graspable assets to accelerate manipulation testing. RLWRLD developed its RLDX-1 dexterity foundation model with Isaac Sim, while [Robotiq](https://blog.robotiq.com/robotiq-releases-tsf-85-digital-twin-on-nvidia-isaac-sim) has integrated Isaac Sim into its open workflows for tactile sensing to improve contact-rich manipulation. [Lyte ](https://lyte.ai/News/ScanToSimReady)is working with NVIDIA to connect LyteVision’s real-world multimodal capture with Isaac Sim, OpenUSD, SimReady and NVIDIA Warp workflows, turning captured scenes into SimReady assets and environments for training robot perception and manipulation policies The final piece of the workflow is deployment. Isaac ROS 4.4 connects learned robot skills from Isaac Sim and Isaac Lab to the ROS 2 software stack, sensors and accelerated compute needed for real-world testing, with new support for extended reality teleoperation, manipulation workflows and Jetson Thor-class hardware. Developers can explore the [NVIDIA Isaac open robotics development platform](https://developer.nvidia.com/isaac) for the tools, models and compute needed to accelerate humanoid development across the full workflow. The end-to-end validated reference workflow from data to deployment will be available in the second half of this year. _Watch the_[ _GTC Taipei keynote_](https://www.nvidia.com/en-tw/gtc/taipei/keynote/) _from NVIDIA founder and CEO Jensen Huang and explore these_[ _physical AI sessions_](https://www.nvidia.com/en-tw/gtc/taipei/session-catalog/?tab.catalogallsessionstab=16566177511100015Kus&search=STW61026%2C%20STW61028%2C%20STW61011%2C%20STW61066%2C%20STW61024%2C%20STW61062%2C%20STW61036%23/) _._ _See [_notice_](https://www.nvidia.com/en-eu/about-nvidia/terms-of-service/) _regarding software product information._ * * * _Tuesday, June 2, 3:00 p.m. PT **[![🔗](https://s.w.org/images/core/emoji/17.0.2/svg/1f517.svg)](https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news/#secure-agents)**_ ## NVIDIA Brings Secure Agent Workspaces and Confidential Computing to AI Factories ![图片](https://blogs.nvidia.com/wp-content/uploads/2026/05/confidential-computing-rolling-blog-1680x945.jpg) Enterprise AI is rapidly evolving from conversational chatbots to persistent, autonomous agents capable of reasoning, writing their own software tools, executing complex cross-system workflows and driving tangible business outcomes. Moving these agentic capabilities from pilot to production triggers tremendous growth in token generation and fundamentally rewrites the enterprise model for security, compliance, infrastructure and cost. To scale safely, organizations need [AI factories](https://www.nvidia.com/en-us/solutions/ai-factories/) — built for secure, trusted and efficient AI production. To help enterprises realize these capabilities, NVIDIA is publishing new, comprehensive reference architectures: [Secure Agent Workspaces](https://docs.nvidia.com/enterprise-reference-architectures/secure-agent-workspace-reference-design/latest/index.html), and Confidential Computing as [Confidential VMs](https://docs.nvidia.com/enterprise-reference-architectures/deploying-proprietary-models-confidential-compute-self-hosted-vms/latest/index.html) and [Confidential Containers](https://docs.nvidia.com/enterprise-reference-architectures/deploying-proprietary-models-confidential-compute-self-hosted-kubernetes/latest/index.html). At GTC Taipei this week, NVIDIA also demonstrated confidential computing for AI with Protopia AI — featuring NVIDIA Confidential Computing protecting model IP and Protopia’s Stained Glass Transform model protecting sensitive data across the entire inference data path. ### **Evolving Governance Requirements for Autonomous AI Workers** Legacy IT security controls — static credentials, network allowlists and standard role-based access — were not designed for autonomous agents. Securing the AI factory requires a paradigm shift to runtime-enforced, policy-driven guardrails. Enterprises need secure agent workspaces — persistent, single-user environments accessed via enterprise single sign-on. Secure agent workspaces are governed by these core principles: The agent runs inside the managed workspace rather than the endpoint, persists beyond the user session for long-running autonomous work and never receives raw credentials. Instead, all access is mediated through trusted brokers, and any consequential actions require human approval. ### **The Growing Need for Zero-Trust Architectures** Agentic AI is reshaping enterprise economics. As agents run continuously and orchestrate complex workflows, efficiency is increasingly measured by cost per token, throughput and GPU utilization. At the same time, sensitive and regulated enterprise data cannot always move to centralized clouds, driving organizations to bring AI to their data and creating a major opportunity for on-premises AI factories. To do this safely, enterprises must protect both data in use and model weights. [NVIDIA Confidential Computing](https://www.nvidia.com/en-us/data-center/solutions/confidential-computing/) provides the hardware-based foundation for zero-trust AI, enabling secure deployme…