In this two-hour and 20-minute speech, Jensen Huang looked ahead to the evolution of AI technology and the prospects for computing demands, while also unveiling the latest generation of NVIDIA's Blackwell architecture products.
Author: Liu Rui, Financial Union
On Tuesday, March 18, local time, NVIDIA CEO Jensen Huang delivered a keynote speech at the NVIDIA AI conference GTC 2025 held in San Jose, California.
In this two-hour and 20-minute speech, Huang looked ahead to the evolution of AI technology and the prospects for computing demands, while also announcing the latest generation of NVIDIA's Blackwell architecture products, the planned shipping times for subsequent generations, and revealing the progress of NVIDIA's collaborative R&D with other tech giants in the fields of autonomous driving, AI networks, and robotics.
Despite the wealth of information, Wall Street's reaction to the speech seemed relatively muted. By the close of trading on Tuesday, NVIDIA's stock price had fallen by 3.43%, with an additional drop of 0.56% in after-hours trading.
Envisioning the Future: Huge Potential for Increased Computing Demand
At the beginning of his keynote, Huang presented his vision for artificial intelligence based on the current timeline of AI development. He described four waves of artificial intelligence:
Perception AI: Launched about 10 years ago, focusing on speech recognition and other simple tasks.
Generative AI: The focus of the past five years, involving text and image creation through predictive patterns.
Agentic AI: The current stage where AI interacts digitally and autonomously performs tasks, characterized by reasoning models.
Physical AI: The future of AI, powering humanoid robots and real-world applications.
Huang pointed out that the AI industry faces "huge challenges" in computing, stating that in the current stage of generative AI, the tokens and resources required for computation are 100 times more than initially expected. He explained that this is due to the numerous steps in the reasoning process that require tokens.
However, Huang insisted that the industry feedback is positive, and the demand for more computing is being met, emphasizing that in just one year, parts of the AI infrastructure market have shown remarkable growth.
He revealed that in 2024, the top four cloud service providers (CSPs) in the U.S., known as hyperscalers, have purchased 1.3 million NVIDIA Hopper architecture chips, and in 2025, they will purchase 3.6 million Blackwell architecture chips.
He emphasized that data center infrastructure is expected to expand rapidly, predicting that driven by AI and accelerated computing demands, capital expenditures on data center infrastructure will exceed $1 trillion by the end of 2028.
Showcasing the Product Roadmap for the Coming Years
Following this, as widely anticipated, Huang confirmed in his speech that NVIDIA will launch the successor to the current generation Blackwell GPU, called Blackwell Ultra, in the second half of 2025.
Huang stated, "Blackwell is fully in production, and the production growth is incredible. Customer demand is unbelievable… We will easily transition to the upgraded version (Blackwell Ultra)."
In addition to the Blackwell Ultra chip, NVIDIA also introduced the GB300 super chip, which combines two Blackwell Ultra chips and one Grace CPU.
Huang also mentioned that NVIDIA will launch the next-generation AI super chip Vera Rubin in the second half of 2026, and the next-generation super chip Vera Rubin Ultra in the second half of 2027—aligning with previous expectations.
Huang revealed that the next generation of chips after the Rubin chip will be named after physicist Richard Feynman, continuing the tradition of naming chips after scientists. According to slides presented by Huang, the Feynman chip is expected to be released in 2028.
New AI Computers Debut
In addition to chips, Huang announced the launch of new laptops and desktops powered by its chips, including two AI-focused computers named DGX Spark and DGX Station, which will be capable of running large AI models such as Llama or DeepSeek.
Among them, DGX Spark is the Project Digits that first appeared at CES, while DGX Station is a larger workstation-level desktop.
Huang claimed that DGX Spark is "the world's smallest supercomputer," packing the GB10 Grace Blackwell super chip into a body not much larger than a Mac mini, with up to 1000 TOPS of AI computing power, making it suitable for "AI developers, research experts, data scientists, and students to develop and fine-tune large AI models in offline environments." Spark is expected to be priced around $3,000 and will be available for pre-order today, with shipments in the summer. Dell, Lenovo, HP, and others are expected to launch corresponding Spark products.
DGX Spark
As for the more powerful DGX Station, it uses the GB300 Grace Blackwell Ultra, offering 20,000 TOPS of AI computing power and up to 784GB of memory. The price of DGX Station has not yet been announced, but it is expected to be released later this year.
Dynamo: The Core Operating System of the AI Factory
To further accelerate large-scale inference, Huang also released an open-source software called NVIDIA Dynamo, designed to accelerate and scale AI inference models in AI factories.
Huang pointed out, "It is essentially the operating system of the AI factory." It is named after the first machine that launched the last industrial revolution, hinting that this technology will play a key role in the new round of AI revolution.
With Dynamo, inference models like DeepSeek can achieve a 30-fold performance increase under the same architecture and using the same number of GPUs.
Launching the World's First Open-Source Customizable General Robotic Model
Huang noted that the labor shortage is an urgent issue facing humanity, and robots are a solution, with this industry holding enormous potential. We have now entered the era of agentic AI, and in the future, we will further advance to physical AI.
To this end, NVIDIA introduced a general foundational model designed specifically for robots, named GR00T N1. This is the world's first open and fully customizable general humanoid inference and skill foundational model.
NVIDIA is also collaborating with Google DeepMind and Disney to develop a robotic platform called Newton. Huang specifically invited a robot named "Blue" to the stage, which is one of the products developed on the Newton platform.
Robots created by NVIDIA in collaboration with Disney Research and Google DeepMind also appeared on stage.
Partnering with General Motors to Develop AI Self-Driving and Smart Factories
Huang also announced that General Motors will expand its partnership with NVIDIA, promoting innovation through accelerated computing and simulation.
General Motors will use NVIDIA's computing platform (including Omniverse and Cosmos) to build customized artificial intelligence (AI) systems to optimize General Motors' factory planning and robotics technology.
Additionally, General Motors will use NVIDIA DRIVE AGX as onboard hardware to enable future advanced driver-assistance systems and enhanced in-car safety driving experiences. DRIVE AGX is a scalable open platform that serves as the AI brain for autonomous vehicles.
Collaborating to Develop AI-Native 6G Networks
Huang stated that NVIDIA will collaborate with T-Mobile, Mitre, Cisco, ODC, and Booz Allen Hamilton to develop hardware, software, and architecture for AI-native 6G wireless networks.
For more details, please click to read "NVIDIA Announces Major Collaboration with Telecom Giants to Develop AI 6G Wireless Technology."
Establishing a Quantum Computing Research Center
In addition to the above, NVIDIA also announced on Tuesday that it will establish a research center in Boston to advance quantum computing with cutting-edge technology.
According to NVIDIA's official website, the NVIDIA Accelerated Quantum Research Center (NVAQC) will integrate leading quantum hardware with AI supercomputers to achieve what is known as accelerated quantum supercomputing. NVAQC will help address some of the most challenging problems in quantum computing, from qubit noise to transforming experimental quantum processors into practical devices.
Leading quantum computing innovators, including Quantum, Quantum Machines, and QuEra Computing, will leverage NVAQC to drive progress in collaboration with researchers from top universities such as Harvard's Quantum Science and Engineering Initiative (HQI) and MIT's Engineering Quantum Systems (EQuS) group.
Huang stated, "Quantum computing will enhance AI supercomputers to solve some of the world's most important problems, from drug discovery to materials development." "By collaborating with the broader quantum research community to advance CUDA - quantum hybrid computing, NVIDIA's Accelerated Quantum Research Center will achieve breakthroughs in creating large-scale, useful accelerated quantum supercomputers."
Launching the World Foundation Model
NVIDIA also announced on Tuesday the launch of the new NVIDIA Cosmos™ World Foundation Models (WFMs), introducing an open and fully customizable inference model for physical AI development, providing developers with unprecedented control over world generation.
NVIDIA also introduced two new blueprints, supported by the NVIDIA Omniverse™ and Cosmos platforms, providing developers with large-scale, controllable synthetic data generation engines for post-training robots and autonomous vehicles.
Industry leaders such as 1X, Agility Robotics, Figure AI, Foretellix, skillai, and Uber are among the first to adopt Cosmos to generate richer training data for physical AI faster and at a larger scale.
Huang stated, "Just as large language models have fundamentally changed generative and agentic AI, the Cosmos World Foundation Model is a breakthrough for physical AI… Cosmos introduces an open, fully customizable inference model for physical AI and creates opportunities for leapfrog advancements in robotics and the physical industry."
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。
