In the early hours of Thursday morning, NVIDIA, the world's most valuable listed company, held its annual shareholder meeting. Renxun Huang, a leader in the AI industry, said at the shareholders' meeting that the question about the return on investment in AI "already has an answer." In the business update, Huang Renxun repeatedly emphasized,The AI data center is a factory that “makes tokens”, tokens can become codes, answers, designs, actions and services,Therefore each token is a profit unit.
This also contains his most sentimental sentence in the whole game:Useful AI has arrived and is already making money(Useful AI has arrived, and it is profitable).
Huang Renxun also emphasized that the NVIDIA system may not be the lowest purchase price, but it can produce the lowest cost words and the highest word throughput - and the highest revenue.
In other words, Nvidia customers are not buying a bunch of servers, but building AI factories that can generate revenue.
This logic also supports Nvidia's amazing financial performance. The company's full-year revenue increased by 65% to US$216 billion, operating income increased by 60% to US$130 billion, operating cash flow reached US$103 billion, and US$41 billion was returned to shareholders. Among them, data center revenue increased by 68% to US$194 billion.
As NVIDIA's next-stage growth engine, the Vera Rubin architecture has now entered full mass production.
Huang Renxun said,Hopper is built for pre-training, Blackwell brings inference to rack scale, and Vera Rubin is built for agents. Agents need to constantly think, access databases, call tools, and execute code. If the CPU cannot keep up, the GPU will be idle; and in an AI factory, idle GPU means loss of revenue.
In addition to the AI factory, Huang Renxun also called "physical AI" the next wave of growth. He thinks,Robots, cars and factories will become real-world agents capable of sensing, reasoning, planning and acting autonomously. NVIDIA will train the model through the AI factory, use Omniverse for simulation, and then run the model on robots and equipment through computing platforms such as Jetson.
In terms of shareholder returns, Huang Renxun said that he plans toReturn 50% or more of free cash flow to shareholders this year, next year and beyond.
The vote at the annual general meeting was overall uneventful. Shareholders approved the company's executive compensation package in an advisory vote and approved the re-election of all 10 board members. A separate outside shareholder proposal was passed calling for amending the company's articles of incorporation so that all shareholder matters can be passed by a simple majority vote.
The following is a transcript of the NVIDIA shareholder meeting "Huang Renxun Business Update" and the Q&A session (AI-assisted translation, some content has been deleted)
Jen-Hsun Huang Business Update
It's been an extraordinary year for Nvidia and the entire computing industry. Every 10 to 15 years, the computer industry will undergo a reset: from mainframe to PC, from PC to Internet, from Internet to cloud, and from cloud to mobile cloud. And this time the reset is bigger.
For the past 60 years, humans wrote software and computers executed instructions. This paradigm has changed. With AI, computers can understand, reason, plan, use tools, and complete useful work. The computer is no longer just a tool. In the age of AI, it is an assistant who knows how to use tools.By extension, the data center is no longer just a "tool shed", but an AI factory composed of digital assistants, which is the infrastructure for producing digital intelligence.
NVIDIA is building computing infrastructure for this new era.
Two years ago, generative AI captured the world's attention. ChatGPT can write, draw, summarize and answer questions. Subsequently, the reasoning AI learned to think about the problem. Now, agent AI has arrived. Agents can use tools, access memory, write code, call other agents, test results, and continue working until the task is completed.
Software programming is the first major breakthrough application scenario on the enterprise side. This is important:AI is now useful. When AI can complete useful work, tokens become valuable; when tokens can bring profits, the demand for computing power will accelerate.
Look at the evidence. GitHub developers merged 300 million pull requests in 2023 (Note: literally translated as a pull request, which refers to a programmer submitting a request to the project "please merge my code into it" after writing a piece of code), 400 million in 2024, and 500 million in 2025, showing a clear and stable upward trend. But in the first few months of 2026, this rate has almost tripled. Obviously, this means that the world’s approximately 30 million software developers—who receive about $3 trillion in compensation every year, and whose work supports $100 trillion in global economic activity—are being amplified by AI. With the help of agents, this same workforce is now producing nearly $9 trillion in output, an increase of $6 trillion.
Programming is just one of the demand drivers.Useful AI has arrived. The question of AI’s return on investment already has an answer, and all walks of life are rushing to adopt intelligent AI.
Nvidia's revenue increased 65% to $216 billion. Operating income increased 60% to $130 billion. Diluted earnings per share increased 67% to $4.90. We generated $103 billion in operating cash flow and returned $41 billion to shareholders.
Data center revenue reached $194 billion, an increase of 68%. Blackwell has significantly expanded the coverage of NVIDIA infrastructure across different customer groups, including hyperscale cloud vendors, cloud service providers, AI labs, industrial enterprises, and sovereign customers. Model developers and hyperscale cloud vendors have deployed hundreds of thousands of Blackwell GPUs. AI factory construction is still expanding rapidly across major industries. Companies such as Capital One, Hyundai Motor Group, Jane Street and Eli Lilly are expanding Nvidia infrastructure to deploy AI.
International revenue more than tripled to more than $30 billion. Nearly 40 countries/regions, representing a combined GDP of US$50 trillion, are building AI factories driven by NVIDIA infrastructure.
AI infrastructure is no longer experimental, it has entered production. AI is not just a model, it is a new industry. You can think of it as a five-layer cake: energy, chips and systems, infrastructure, models, and applications.
Traditional data centers store and serve files. The AI factory manufactures tokens. Tokens become codes, answers, designs, actions and services.
Useful AI is profitable.Each token is a profit unit, which is why the demand for computing power is extremely strong.Customers are not buying computers, they are building AI factories that can generate revenue.
The architecture of the (AI) factory is very important. The question is how much revenue the plant will generate, and at what cost. Inference is the process of generating tokens, and NVIDIA Blackwell sets the standard. In SemiAnalysis's InferenceX benchmark, Blackwell was recognized as the "King of Inference," able to provide the lowest cost per token and achieve 30 times higher token throughput than the second-place platform.
This is why architecture is so critical. NVIDIA systems may not be the cheapest to purchase, but NVIDIA can generate the lowest cost tokens, the highest token throughput, and the highest revenue.
Vera Rubin is the next step. Hopper is built for pre-training. Blackwell brings inference to rack scale. Vera Rubin is built for intelligent agents.
Agent AI changes the paradigm of computing. The agent thinks, uses tools, accesses databases, retrieves memories, executes code, and calls applications repeatedly until the task is completed. The big language model running on the GPU does the thinking, and the CPU has to keep up. If the CPU becomes the bottleneck, the GPU will sit idle. In an AI factory, idle GPUs mean lost revenue.
This is why Vera is important. Vera is the CPU for intelligent agents, and Rubin is the GPU for thinking. NVLink, Spectrum-X, storage and security capabilities on BlueField, and software connect these systems together.
In fact,Nvidia is the only company with three networking businesses. NVLink connects GPUs in a rack into a giant computer. Spectrum-X is Ethernet built for AI that scales out inside and outside the AI factory and is now larger than all other Ethernet networking peers combined. InfiniBand provides the lowest latency network for the world's largest AI and scientific computing systems.
Combined, these technologies enable AI factories to optimize holistically from GPU to rack, from within the data center to across the data center.
Vera Rubin is not a chip, but an AI factory platform, and the ecosystem has already started to act.Vera Rubin is now in full production. Every major model developer, public cloud, AI cloud, and hyperscale cloud vendor is preparing to build on it.
Vera opens up a whole new market. So far, every CPU has been built for humans. We live in a world where seconds are measured. Agents live in a world measured in nanoseconds. Every moment the CPU makes the agent wait is a moment when the most expensive thing in the entire building—the GPU—is idle.
Therefore, we built a brand-new, agent-oriented CPU from scratch. This is a new market. CPUs that serve humans are divided and rented by core. Instead of renting cores, agents require ultra-fast responses. There will be billions of intelligent agents in the future, and they will need CPUs built for themselves.
We believe Vera will be one of the most important product launches in the company's history, and orders are already starting to come in.
CUDA is one of the most important investments we have ever made. We've been working on the same accelerated computing architecture for 20 years. The installed base attracts developers, developers create breakthrough applications, applications create new markets, and new markets expand the installed base. This flywheel is accelerating.
CUDA-X is a library stack built on top of CUDA. These libraries are NVIDIA's crown jewel. They solve some of the most difficult problems in science and industry, including computational lithography, optimization, genomics, physics, data processing, robotics, AI, wireless networks, and more.
Now these libraries are becoming tools for agents. This week we announced BioNeMo, a suite of digital biology and drug discovery tools for agents.
NVIDIA is vertically integrated and horizontally open. We build a complete technology stack that enables end-to-end optimization of the system. Then we open it up for the entire industry to build on top of it.
Our business is broad and becoming more diverse. To make it easier for everyone to understand our business, we now use two market platforms to describe NVIDIA: data center and edge computing.
在数据中心领域，我们服务两个市场：超大规模数据中心，以及AI云、工业和企业市场。我们的客户基础多元且不断增长。
Edge computing includes PCs, workstations, games, AI base stations, robots and cars. We are able to serve all of these markets because we have a unified architecture, a software stack, and a rich ecosystem.
Physics AI is Nvidia’s next wave of growth. Physical AI is intelligent agent AI in the real world. Robots, cars, and factories will be able to sense, reason, plan, and operate in dynamic environments.
NVIDIA pioneered this field and built a complete closed loop. AI Factory trains models; Omniverse simulates them in the virtual world; NVIDIA Jetson computers run in robots; Cosmos is the basic model of the world that drives it all.
Robots and robotic systems are being built in a variety of industries, from transportation, manufacturing, and surgical robots to hospitality and service industries.
There has been a huge acceleration in AI over the past few months. Finally, I would like to emphasize a few points:Useful AI is already here, and it's profitable. Therefore, computing power is income. Vera Rubin is now in full production. Vera is opening a new CPU market for intelligent agents. Every business is becoming an intelligent company, and they're all running on NVIDIA.
The AI era is advancing at full speed, and NVIDIA is building the infrastructure to drive this era.
As we grow, we will continue to expand our R&D investments, invest in our ecosystem, and return capital to shareholders. We recently announced a significant dividend increase and expanded our share repurchase program.We plan to return 50% or more of our free cash flow to shareholders this year, next year and beyond.
Thanks.
Part of the Q&A record
Question: How sustainable is the current AI infrastructure construction? When will the key drivers of your business mature, and when will growth slow?
Jensen Huang: As I mentioned in my speech, AI is more than just a model. It represents a fundamental shift in computing. For 60 years, computing was mainly about retrieving, storing and sending information; now, computing is being reinvented by AI and turned into generative intelligence.
Token is the basic unit of intelligence. They are manufactured in a new type of data center—the AI factory—and generated revenue through commercialization. The more computing power, the more tokens, and the more income.
This round of construction will be measured on a multi-decade scale, similar to other critical infrastructure construction such as power grids, transportation systems and the Internet. We believe this will be the largest infrastructure development in human history.
Agent AI is accelerating infrastructure investment because it is the first time that AI is actually starting to do real work and create real economic value. As organizations of all types seek to manufacture intelligence at scale, the need for AI factories will extend well beyond today’s cloud computing to enterprise, sovereign nation, and regional AI clouds.
NVIDIA is uniquely positioned to advance the construction of AI factories with its full-stack, end-to-end co-designed infrastructure and large partner ecosystem that supports these projects.
The next major growth phase is physics AI, and it's just getting started. Over time, AI will move from the digital world into self-driving taxis, humanoid robots, and industrial systems, enabling these machines to sense, reason, and act with autonomy in the physical world.
Driving physical AI requires a new round of infrastructure investment. In this area, NVIDIA provides AI infrastructure, Omniverse simulation and digital twin platform, open models, Jetson Thor embedded computing platform, and software stack for developing and deploying physical AI at scale.
We still have a long growth runway ahead of us.
Question: NVIDIA GPUs currently power most of the world's AI training infrastructure, but as inference workloads exceed training, how confident is NVIDIA that the GPU architecture will remain the platform of choice for large-scale inference?
Jen-Hsun Huang: Nvidia is really good at training. With Blackwell, we have also established our leading position in the field of reasoning.
Reasoning is enabling AI to realize commercial monetization. Since data centers are power-constrained, their token throughput and revenue potential depend on the performance per watt of the AI infrastructure.
In SemiAnalysis's Inference-X benchmark, Blackwell was dubbed the "King of Inference," achieving the best performance per watt, the lowest token cost, and 30 times higher token throughput.
We achieved our seventh consecutive victory in the latest round of the MLPerf inference benchmark. For AI agents, Artificial Analysis' New Agent Perf results show that compared to Hopper, the Grace Blackwell 300 NVLink 72 system can run up to 20 times more agents per megawatt of power consumption.
This superior performance is the result of NVIDIA's extreme collaborative design at the chip, system, algorithm and software levels. NVIDIA AI infrastructure delivers the best performance and therefore the best inference economics.
But our advantage goes beyond leading performance. Our large installed base allows developers to reach the largest number of AI users. NVIDIA's programmable GPU architecture runs more than 7,000 applications, providing cloud customers with the greatest revenue opportunities, as well as financing and offtake potential.
For enterprise and sovereign customers, NVIDIA offers unparalleled deployment flexibility, the broadest range of leading open source models for custom AI, and the deepest IT ecosystem support in the industry.
Today, the vast majority of Nvidia's computing footprint is in inference. We are well positioned to further gain share, as evidenced by our recent announcements with Anthropic and Apple.
Question: Can you update the capital return policy? What should investors expect in terms of dividends and share buybacks?
Jensen Huang: Nvidia offers investors a unique combination of superior growth, strong margins, free cash flow execution, and rising returns on capital.
On our most recent earnings call, we announced a 25-fold increase in our quarterly dividend and authorized an additional $80 billion in share repurchases.
Based on our confidence in sustainable market growth and free cash flow generation, we plan to return 50% or more of free cash flow to shareholders this year, next year and beyond, and continue to increase share repurchases and dividends over time.