NVIDIA delivered a comprehensive roadmap for the next phase of the AI industry at GTC Taipei 2026. Jensen Huang outlined NVIDIA's full-stack strategy centered around Vera Rubin, Vera CPU, DSX AI Factory, RTX Spark, and the Physical AI platform, extending from cloud data centers to desktop terminals, and from digital agents to real-world robots.

【Taipei, June 1, 2026】During the keynote at GTC Taipei 2026, NVIDIA founder and CEO Jensen Huang systematically showcased the company's next phase of technology layout, announcing key highlights covering the next-generation computing architecture Vera Rubin, the DSX reference design for AI factories, the Vera CPU serving Agentic AI, the RTX Spark/DGX Spark aimed at developers and edge, and the Physical AI platform built around Omniverse, Cosmos, and Isaac GR00T.

During the speech, Jensen Huang emphasized that AI has shifted from a training-driven phase to a reasoning-driven phase, and the industry's demand is rapidly changing from "generating content" to "generating tokens, executing tasks, calling tools, and driving agents." This change means that data centers, enterprise software, endpoint devices, and robotic systems all need to be redesigned around AI-native computing architectures.

AI Factory Becomes NVIDIA's New Core

In this speech, "AI Factory" was one of the keywords that resonated throughout. Jensen Huang defined AI infrastructure as a new type of industrial system, suggesting that future enterprises will not only purchase servers or GPUs, but rather "factories" for producing tokens, training models, running agents, and deploying physical AI.

Building on this positioning, NVIDIA further advanced the DSX AI Factory reference design and incorporated Omniverse digital twin capabilities into the deployment process, enabling enterprises to complete simulation, planning, and optimization of data center-level AI factories before cabinet installation. According to the speech, NVIDIA aims to integrate hardware, network, system software, runtime, and simulation platforms into a unified delivery paradigm, lowering the threshold for building large-scale AI infrastructure.

Vera Rubin Takes Over from Blackwell

In terms of hardware roadmap, Vera Rubin was the absolute focus of this keynote. Jensen Huang positioned it as the core platform for the next generation of AI and Agentic AI workloads following Grace Blackwell and repeatedly emphasized its leaps in system-level interconnect, memory, CPU/GPU collaboration, and inference throughput.

According to NVIDIA's public statements in GTC 2026 news materials, Vera Rubin is defined as the platform that unlocks new frontiers in Agentic AI, while the Vera CPU is a new type of processor "born for Agentic AI." From the speech content, it is clear that NVIDIA is attempting to expand the competition that previously focused on GPUs to an overall competition among CPUs, GPUs, DPUs, switches, software stacks, and model toolchains.

Agentic AI Raises CPU Importance

Differing from previous emphasis on GPUs, Jensen Huang repeatedly mentioned the importance of CPUs in the Agentic AI era during this speech. His core logic is that agent systems require not only model inference but also higher-frequency context management, tool invocation, state orchestration, memory management, and system scheduling. Thus, CPUs are no longer just complementary components but are key components of AI systems.

Around this judgment, NVIDIA has deeply coupled the Vera CPU with the Rubin platform and placed it at the center of the Agentic AI infrastructure. This change also means that future AI data center competition will no longer be simply about "who has the fastest GPU," but rather who can provide a more complete, lower latency, and better-suited system architecture for multi-agent cooperation.

From Cloud to Desktop and Edge

Besides data centers, Jensen Huang also dedicated significant time to local AI devices. In the speech, RTX Spark and DGX Spark were positioned as key products that allow developers, creators, and enterprise users to run the latest open models and AI agents on the desktop side, emphasizing local deployment, continuous operation, and low-threshold development capabilities.

Jensen Huang highlighted RTX Spark's capabilities including the Blackwell RTX GPU, CUDA cores, Grace CPU, NVLink interconnect, and large-capacity unified memory, describing it as one of the representative forms of AI-native PCs. NVIDIA's official GTC news materials also clearly stated that GTC 2026 focused on showcasing scenarios of locally running open models and AI agents on RTX PCs and DGX Spark.

Physical AI: Interconnection of Omniverse, Cosmos, and Robotic Platforms

Another main theme of this speech was Physical AI. Jensen Huang showcased the collaborative relationship between Omniverse, Cosmos, Isaac, and the GR00T platform, attempting to build a complete closed loop from world models, simulation training to robotic deployment.

From the speech, it can be seen that Cosmos is used to support modeling capabilities for the physical world, Omniverse continues to serve as the foundation for simulation and digital twins, while Isaac GR00T advances robotic development and training scenarios. NVIDIA's GTC 2026 news materials also list robots, autonomous driving, visual AI agents, and open physical AI data factories as key directions, indicating that the company is elevating "Physical AI" to a long-term strategy on par with Agentic AI.

Software Stack and Open Ecosystem Continue to Expand

In addition to hardware, NVIDIA also reinforced the layout of CUDA-X, NeMo, Nemotron, and development toolchains for Agentic AI in the speech. Public information shows that NVIDIA concurrently promotes the open model family, Nemotron alliance, agent development platform, as well as reference designs for the OpenClaw community during GTC 2026, intending to string models, tools, inference systems, and enterprise implementation capabilities into a unified ecosystem.

This means that NVIDIA's competitive boundary has expanded from accelerated computing platforms to model supply, development frameworks, runtime systems, enterprise integration, and even industry solutions. For the tech industry, the strongest signal released by this Taipei speech is not a parameter of a specific chip, but that NVIDIA is attempting to become an "operating system-level company" in the era of AI infrastructure.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。