① At NVIDIA’s annual shareholder meeting, Jensen Huang once again defined AI data centers as “factories that manufacture tokens,” with each token being a profit unit.
② He emphasized that the core advantage of NVIDIA systems lies not in the lowest procurement price, but in the lowest token cost, the highest token throughput, and stronger revenue-generating capability.
③ Huang also mentioned that physical AI is expected to become the next-stage growth engine.
NVIDIA will train models through AI factories, use Omniverse for simulation, and then run models on robots and devices through computing platforms such as Jetson.
Jensen Huang also expressed optimism about long-term growth and ASIC competition, emphasizing that “this round of buildout will be measured in decades” and that NVIDIA’s infrastructure can provide “the best inference economics.” On shareholder returns, Jensen Huang stated that the plan is to return 50% or more of free cash flow to shareholders this year, next year, and in the future. The annual shareholder meeting votes were largely uneventful. Shareholders approved the company’s executive compensation plan in an advisory vote and approved the re-election of all 10 board members. Another external shareholder proposal also passed, requiring amendments to the company’s bylaws so that all shareholder voting items can be approved by simple majority. The following is a transcript of NVIDIA’s shareholder meeting “Jensen Huang Business Update” and Q&A session (AI-assisted translation, some content abridged). Jensen Huang Business Update This has been an extraordinary year for NVIDIA and the entire computing industry. Every 10 to 15 years, the computer industry undergoes a reset: from mainframes to PCs, from PCs to the Internet, from the Internet to the cloud, and from the cloud to the mobile cloud. This reset is even bigger. For the past 60 years, humans have written software and computers have executed instructions. This paradigm has changed. With AI, computers can understand, reason, plan, use tools, and perform useful work. Computers are no longer just a tool. In the AI era, a computer is an assistant that uses tools. By extension, the data center is no longer just a “tool shed,” but an AI factory composed of digital assistants, 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, reasoning AI learned to think through problems. Now, agentic AI has arrived. Agents can use tools, access memory, write code, call other agents, test results, and continue working until the task is complete. Software programming is the first major breakthrough application scenario on the enterprise side. This matters: AI is now useful. When AI can complete useful work, tokens become valuable; when tokens can generate profit, demand for computing power will accelerate.Review the evidence. GitHub developers merged 300 million pull requests in 2023, 400 million in 2024, and 500 million in 2025, showing a clear and steady upward trend. But in the first few months of 2026, that pace nearly tripled. Clearly, this means that the world’s approximately 30 million software developers—who earn about $3 trillion in compensation each year and whose work underpins $100 trillion in global economic activity—are being amplified by AI. With agents, this same workforce is now generating close to $9 trillion in output, an increase of $6 trillion.
Programming is only one demand driver. Useful AI has arrived. The question of AI’s return on investment now has an answer, and industries across the board are racing to adopt agentic AI. NVIDIA revenue grew 65% to $216 billion. Operating income grew 60% to $130 billion. Diluted earnings per share grew 67% to $4.90. We generated $103 billion in operating cash flow and returned $41 billion to shareholders. Data center revenue reached $194 billion, up 68%. Blackwell significantly expanded the reach of NVIDIA infrastructure across customer segments, including hyperscalers, cloud service providers, AI labs, industrial enterprises, and sovereign customers. Model developers and hyperscalers have each deployed hundreds of thousands of Blackwell GPUs cumulatively. AI factory buildouts continue to expand 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, exceeding $30 billion. Nearly 40 countries and regions, representing a combined $50 trillion in GDP, are building AI factories powered by NVIDIA infrastructure. AI infrastructure is no longer experimental; it has entered the production stage. 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. AI factories manufacture tokens. Tokens become code, answers, designs, actions, and services.Useful AI is profitable. Every token is a unit of profit, which is precisely why compute demand is extremely strong. Customers are not buying computers; they are building AI factories capable of generating revenue.
(AI) factory architecture matters greatly. The question is how much revenue this factory can generate, and what the cost is. 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,” delivering the lowest cost per token and achieving 30 times the token throughput of the second-place platform. This is why architecture is so critical. NVIDIA systems may not have the lowest purchase price, but NVIDIA can generate the lowest-cost tokens, the highest token throughput, and the highest revenue. Vera Rubin is the next step. Hopper was built for pre-training. Blackwell brought inference to rack scale. Vera Rubin is built for agents. Agentic AI changes the computing model. Agents think, use tools, access databases, retrieve memory, execute code, and repeatedly call applications until tasks are completed. Large language models running on GPUs handle the thinking, while CPUs must keep pace. If the CPU becomes a bottleneck, GPUs sit idle. And in an AI factory, idle GPUs mean lost revenue. This is why Vera matters. Vera is the CPU built for agents, and Rubin is the GPU responsible for thinking. NVLink, Spectrum-X, storage and security capabilities on BlueField, and software together connect these systems. In fact, NVIDIA is the only company with three networking businesses. NVLink connects GPUs within a rack into a single giant computer. Spectrum-X is Ethernet built for AI, capable of scaling horizontally both inside and outside AI factories, and its scale now exceeds the combined total of all other Ethernet networking peers. InfiniBand provides the lowest-latency networking for the world’s largest AI and scientific computing systems. Combined, these technologies enable AI factories to be optimized holistically from GPU to rack, and from within a data center to across data centers.Vera Rubin is not a chip, but an AI factory platform, and the ecosystem has already begun to act. Vera Rubin is now in full production. Every major model developer, public cloud, AI cloud and hyperscaler is preparing to build on it.
Vera opens an entirely new market. Until now, every CPU has been built for humans. We live in a world measured in seconds. Agents live in a world measured in nanoseconds. Every moment a CPU makes an agent wait is a moment when the most expensive thing in the building, the GPU, sits idle. Therefore, we built a new CPU for agents from the ground up. This is a new market. CPUs serving humans are partitioned and rented by core. Agents do not rent cores; they demand ultra-fast response. There will be billions of agents in the future, and they need CPUs built for them. We believe Vera will be one of the most important product launches in the company’s history, and orders are already coming in. CUDA is one of the most important investments we have ever made. For 20 years, we have committed to a single accelerated computing architecture. 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 the library stack built on top of CUDA. These libraries are NVIDIA’s crown jewels. They solve some of the hardest 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 the full technology stack so that we can optimize systems end to end. We then open it up so that the entire industry can build on top of it. Our business is broad and becoming more diversified. To make our business easier to understand, we now describe NVIDIA through two market platforms: data center and edge computing. In data center, we serve two markets: hyperscale data centers, and AI cloud, industrial and enterprise markets. Our customer base is diverse and growing. Edge computing includes PCs, workstations, gaming, AI base stations, robotics and automobiles.We are able to serve all of these markets because we have a unified architecture, a single software stack, and a rich ecosystem.
Physical AI is NVIDIA’s next wave of growth. Physical AI is agentic AI in the physical world. Robots, cars, and factories will be able to perceive, reason, plan, and operate in dynamic environments. NVIDIA pioneered this field and built a complete closed loop. AI factories train models; Omniverse simulates them in virtual worlds; NVIDIA Jetson computers run inside robots; and Cosmos is the world foundation model that drives it all. Robots and robotic systems are being built across industries, from transportation, manufacturing, and surgical robotics to hospitality and services. Over the past few months, AI has accelerated enormously. In closing, I want to emphasize a few points: useful AI has arrived, and it is profitable. Therefore, compute is revenue. Vera Rubin is in full production. Vera is opening an entirely new CPU market built for agents. Every company is becoming an agentic company, and they all run on NVIDIA. The AI era is moving forward at full speed, and NVIDIA is building the infrastructure that powers this era. As we grow, we will continue to expand R&D investment, invest in our ecosystem, and return capital to shareholders. We recently announced a significant dividend increase and an expanded share repurchase program. We plan to return 50% or more of free cash flow to shareholders this year, next year, and beyond. Thank you. Excerpts from the Q&A Session Question: How sustainable is the current AI infrastructure buildout? When will the main drivers of your business mature, and when will growth slow? Jensen Huang: As I mentioned in my presentation, AI is not just a model. It represents a fundamental shift in computing. For 60 years, computing has mainly been about retrieving, storing, and transmitting information; now, computing is being reinvented by AI to generate intelligence. Tokens are the basic unit of intelligence. They are manufactured in a new type of data center—AI factories—and generate revenue through commercialization. The more compute, the more tokens, and the more revenue. This buildout will be measured over decades, similar to other critical infrastructure such as power grids, transportation systems, and the internet. We believe this will become the largest infrastructure buildout in human history.Agentic AI is accelerating infrastructure investment, as this is the first time AI is truly beginning to do real work and create real economic value. As organizations seek to manufacture intelligence at scale, demand for AI factories will extend far beyond today’s cloud computing to enterprises, sovereign nations, and regional AI clouds.
With full-stack, end-to-end co-designed infrastructure and a large partner ecosystem supporting these projects, NVIDIA is uniquely positioned to drive AI factory construction. The next major growth phase is physical AI, and it is only just beginning. Over time, AI will move from the digital world into robotaxis, humanoid robots, and industrial systems, enabling these machines to perceive, reason, and act autonomously in the physical world. Driving physical AI requires a new wave of infrastructure investment. In this area, NVIDIA provides AI infrastructure, the Omniverse simulation and digital twin platform, open models, the Jetson Thor embedded computing platform, and a software stack for developing and deploying physical AI at scale. A long growth runway lies ahead. Question: NVIDIA GPUs currently support most of the world’s AI training infrastructure, but as inference workloads surpass training, how confident is NVIDIA that GPU architecture will remain the preferred platform for large-scale inference? Jensen Huang: NVIDIA is indeed very strong in training. And with Blackwell, we have also established leadership in inference. Inference is enabling AI to be monetized. Because data centers are power-constrained, their token throughput and revenue potential depend on the per-watt performance of AI infrastructure. In SemiAnalysis’s Inference-X benchmark, Blackwell was named the ‘King of Inference,’ achieving the best per-watt performance, the lowest token cost, and 30 times higher token throughput. In the latest MLPerf inference benchmark round, we achieved our seventh consecutive victory. For AI agents, Artificial Analysis’s New Agent Perf results show that the Grace Blackwell 300 NVLink 72 system can run up to 20 times more agents per megawatt of power compared with Hopper. This superior performance is the result of NVIDIA’s extreme co-design across chips, systems, algorithms, and software.NVIDIA AI infrastructure delivers the best performance, and therefore also delivers the best inference economics.
But our advantage is not only in leading performance. Our large installed base enables developers to reach the largest number of AI users. NVIDIA’s programmable GPU architecture runs more than 7,000 applications, bringing cloud customers the largest revenue opportunities, as well as potential in financing and offtake. For enterprise and sovereign customers, NVIDIA offers unparalleled deployment flexibility, the industry’s most widely used 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 used for inference. We are well positioned to further expand our share, as demonstrated by our recent announcements with Anthropic and Apple. Question: Can you update us on the capital return policy? What should investors expect regarding dividends and share repurchases? Jensen Huang: NVIDIA offers investors a unique combination: exceptional growth, strong margins, free cash flow execution capability, and rising capital returns. On our most recent earnings call, we announced a 25-fold increase in our quarterly dividend and an additional $80 billion share repurchase authorization. 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 over the longer term, and to continue increasing share repurchases and dividends over time.
