NVIDIA founder Jensen Huang: Accelerating computing and AI factories define the future.

Author: Techub News Compilation

Introduction

On October 28, 2025, NVIDIA Founder and CEO Jensen Huang delivered a nearly 100-minute keynote speech at the GTC (GPU Technology Conference) held in Washington, D.C. As an annual barometer in the fields of AI and accelerated computing, this year’s GTC moved to the political center of the United States, carrying significance beyond a technical launch event. In his speech, Huang not only reviewed the 30-year journey of NVIDIA pushing the paradigm shift in computing but also announced a series of heavyweight collaborations and products involving communications, quantum computing, national research, next-generation AI infrastructure, and robotics technology, outlining a grand blueprint centered around “accelerated computing” and “AI factories” that drives industrial transformation in the U.S. and globally.

Summary

Announced a strategic cooperation with Nokia to launch the NVIDIA Arc 6G wireless network computing platform, aimed at integrating AI into the core of wireless communications and restoring American technology's leadership in global telecommunications infrastructure.
Released the NV-Q Link quantum interconnect architecture and the CUDA-Q platform to connect quantum processors with GPU supercomputers, achieving quantum error correction and hybrid computing, and announced support from seven national laboratories under the U.S. Department of Energy and 17 quantum companies.
Revealed details about the next-generation AI supercomputer architecture Vera Rubin, emphasizing performance improvement through “extreme collaborative design” and significant cost reduction in token generation, and announced securing $500 billion worth of Blackwell and early Rubin orders.
Proposed the concept of “AI factory,” believing AI is not a tool but a “worker” driving a dual platform shift from general computing to accelerated computing, from handwritten software to AI, impacting an economy on the scale of trillions of dollars.
Demonstrated the fusion of Physical AI and robotics by training and deploying robots through the Omniverse digital twin platform, and announced progress in smart factories, humanoid robots, and autonomous driving (Hyperion platform) with partners like Foxconn, Figure, and Disney.

From GPU to CUDA: Thirty Years of Paradigm Shift

At the beginning of his speech, Jensen Huang reviewed the fundamental transformation NVIDIA has driven over the past thirty years: accelerated computing. He pointed out that as the performance improvements of transistors slow down due to physical laws (the "Dennard scaling" has stopped), the limitations of general computing (CPU) are becoming increasingly apparent. NVIDIA's insight was that by introducing GPUs designed specifically for parallel processing along with the accompanying programming model CUDA, computing power could be greatly expanded. However, this was not merely a hardware replacement; it was a comprehensive reconstruction of algorithms, libraries, and applications, taking nearly thirty years to reach a turning point.

Huang emphasized that NVIDIA's core wealth is not just the GPU hardware, but the complete ecosystem consisting of CUDA and over 350 accelerated libraries built upon it. From CUDA Litho for computational lithography, scientific computing to AI training frameworks (Megatron Core), and medical imaging (Monai), each library represents a field reshaped by accelerated computing. He showcased a fully simulated generated video to demonstrate how the CUDA-X ecosystem covers nearly all industries from healthcare, manufacturing, and robotics to autonomous driving and computer graphics, emphasizing that the driving force behind this is the "beauty of mathematics" and deep computer science.

Reshaping Communications, Quantum, and National AI Infrastructure

Before announcing specific products, Huang first pointed out a critical issue in current global communication infrastructure: heavy reliance on foreign technology. In response, NVIDIA announced a collaboration with Nokia, the second largest telecommunications equipment manufacturer in the world, to introduce a brand-new product line NVIDIA Arc (Aerial Radio Network Computer). The Arc platform, based on Grace CPU, Blackwell GPU, and ConnectX networking technology, runs a CUDA-X library named Aerial, aiming to create the first software-defined, programmable wireless computing platform that handles AI tasks. Huang stated that this is not only about applying AI to wireless access networks (AI for RAN) to improve spectrum efficiency but also about building an edge industrial robot cloud on wireless networks (AI on RAN), similar to how the internet gave rise to cloud computing giants like AWS.

In the field of quantum computing, NVIDIA introduced the NV-Q Link interconnect architecture and the CUDA-Q open platform. NV-Q Link can transfer terabytes of data between quantum hardware and GPUs at thousands of times per second, which is crucial for achieving quantum error correction, device calibration, and hybrid simulation. Huang announced that 17 quantum computing companies, along with seven national laboratories under the U.S. Department of Energy (DOE)—including Berkeley, Los Alamos, and Oak Ridge—support this architecture. Following this, he announced another major news: the U.S. Department of Energy will collaborate with NVIDIA to build seven new AI supercomputers to promote national scientific development. He praised Secretary of Energy Chris Wright's vision and emphasized that computing is the foundational tool for science, while accelerated computing, AI, quantum computing, and robotics are converging to reshape the research paradigm.

AI is not a Tool, but a “Worker”: Defining the Era of AI Factories

Huang spent considerable time explaining his understanding of the nature of AI. He believes that the public equating AI to chatbots is one-dimensional. The fundamental significance of AI is that it radically reconstructs the computation stack: shifting from handwritten code (CPU) to data-driven machine learning (GPU). He referred to the output of AI as “tokens,” which are the language units of AI. Any structured information—text, images, video, 3D structures, chemical molecules, proteins—can be tokenized and learned, understood, and generated by AI.

He further proposed a key distinction: past software (such as Excel, Word) are “tools,” whereas AI is a “worker.” The market size for tools is limited (about a trillion dollars), but AI, as a “worker” that can use tools, will directly participate in and enhance global economic activities on the scale of trillions, solving labor shortages. This fundamental shift creates a demand for new computational infrastructure: AI factories.

An “AI factory” differs from traditional general-purpose data centers; it is designed specifically for the efficient production of “valuable tokens,” requiring extreme response speed, throughput, and cost-effectiveness. Huang pointed out that two “exponential” drivers underpin the demand for AI factories: one is the exponential demand for computing power across three phases of AI model—pre-training, fine-tuning, and “thinking” (inference); the other is that the smarter the model, the more it is used, generating a positive feedback loop for more computing power. However, the end of Moore's Law poses challenges to meet these exponential demands. NVIDIA's answer is: extreme collaborative design.

Extreme Collaborative Design: Exponential Leap from Blackwell to Vera Rubin

Huang emphasized that NVIDIA is the only company today that can start with a blank page and simultaneously think about new computing architectures, chips, systems, software, model architectures, and applications. The result of this “extreme collaborative design” is rack-level supercomputers like the GB200 NVL72. He compared the entire rack to a “giant GPU,” achieving performance far beyond traditional interconnect architectures by connecting 72 GPUs (144 chiplets) through NVLink 72 technology.

He cited third-party benchmarks stating that the Grace Blackwell platform achieved 10 times the performance improvement per GPU and a 10 times reduction in token generation costs compared to the previous generation H200 under the same workload. This is key to maintaining the “flywheel” of AI growth—sustaining reductions in unit costs through architectural innovation rather than merely relying on an increase in transistor count as power demand grows exponentially.

He then showcased the next-generation AI supercomputer platform Vera Rubin. This is the third-generation NVLink 72 rack-level system, featuring a fully cable-free, 100% liquid cooling design, and introducing a new “context processor” to handle the increasing model context lengths. Huang revealed that the Blackwell platform has delivered 6 million GPUs (by package), and the order visibility for the next five quarters has reached an astonishing $500 billion, five times the total revenue of the previous Hopper architecture over its entire lifecycle. He particularly mentioned that this manufacturing is returning to the U.S., with factories in places like Arizona starting full-speed production of Blackwell, resonating with the theme of “reindustrialization of America.”

To scale these AI factories, NVIDIA launched the Omniverse DSX digital twin blueprint. Through integration with partners like Jacobs, Siemens, and Schneider Electric, DSX allows for the complete digitalization of processes from design, thermal simulation to operational optimization before physical construction, greatly shortening construction time and optimizing energy efficiency.

Open Source Models, Enterprise Empowerment, and the Rise of Physical AI

Huang reiterated the importance of open-source AI models, calling them a “lifeline” for startups and researchers. NVIDIA itself has invested heavily in the open-source field, having leading models in various domains (language, physical AI, biology, etc.). He also announced deep collaborations with two major enterprise software giants: partnering with CrowdStrike to create the next-generation AI-based cybersecurity defense system; partnering with Palantir to accelerate its data platform, providing real-time insights for governments and enterprises' vast data.

The climax of the speech turned to “Physical AI.” Huang pointed out that Physical AI (like robotics) requires coordination among three computers: the GB200 NVL72 for training models, the “Omniverse computer” for simulating in the Omniverse digital twin, and the “robot computer” (like Jetson Thor) ultimately deployed on the robot body. He showcased a digital twin of a smart factory in Texas built in collaboration with Foxconn, where robots learn collaboration in a virtual environment through Isaac Sim. He also mentioned collaborations with humanoid robot company Figure (valued at nearly $40 billion), Agility Robotics, Johnson Surgical Robots, and Disney Research. Disney's robot “Blue” learns to interact with the real world in the Omniverse using the Newton physics simulator, heralding a huge consumer-grade robotics market.

In the field of autonomous driving, NVIDIA announced that its Drive Hyperion reference architecture has become the "robotic taxi-ready" platform for automakers like Lucid, Mercedes-Benz, and Stellantis. More importantly, a major collaboration was established with Uber for global network integration, enabling users to call autonomous vehicles based on the Hyperion platform through Uber in the future.

Conclusion: Dual Platform Shift and Reindustrialization of America

In the summary of his speech, Huang defined the current era as a convergence point of two platform shifts: from general computing to accelerated computing, and from handwritten software to artificial intelligence. CUDA and its ecosystem are the engine of the first shift, which has now formed a self-reinforcing “flywheel.” AI is the second shift, and its “flywheel” has also begun to spin.

He concluded by emphasizing that through new platforms like ARC (6G), Hyperion (autonomous driving), DSX (AI factory), and MEGA (AI chemical factory), NVIDIA is, together with partners, injecting AI capabilities into every corner from communications, transportation to manufacturing. The GTC held in Washington is not just a showcase of technology, but a declaration about how America can drive reindustrialization through AI and reclaim leadership in technology and the economy. Huang ended with “Make America Great Again,” eliciting prolonged applause from the audience.

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。