I believe outside all this time. Can't wait for this. I can't wait to see this play out in real life in the front row through the late night. This is how intelligence is made. A new kind of factory, generator of tokens, the building blocks of AI. Tokens have opened a new frontier, turning data into knowledge, reason, action. They reveal patterns in complexity you could never see. Mirror our cities to keep us safe and lift us high above them. Tokens help robots learn from us.
Work alongside us. They go where we cannot. Lending us helping hands and closing the gap between hope and healing so that we breathe easier. And the smallest hearts beat stronger.
Tokens are helping us break new ground on a scale never attempted so we can reach star cloud one. Separation confirmed to infinity and beyond. Together we take the next great leap into a bright new future built for all mankind.
And here in Taipei is where it all begins.
Welcome to the stage Nvidia founder and CEO Jensen Huang.
Welcome to GTC Taiwan. So great to see all of you. Very good to be home. I brought my parents home. Where are my parents? Everybody give round of applause to my mom and dad.
And a round of applause for our pregame show superstars. Ladies and gentlemen, look how adorable they are. The superstars of Taiwan. There are so many of you here today. We are broadcasting this right now to 70 other watch parties across Taiwan. 70 different conferences are going at the same time. Everybody is watching this keynote. We have so much to tell you and I have so many partners to thank. It is incredible how large our ecosystem in Taiwan has become. Most of the time when people think about ecosystem, they think about our software stack. They think about the developer ecosystem above the computing systems that Nvidia builds, but Nvidia's ecosystem spans all the way upstream to all of our supply chain here in Taiwan where it all begins and downstream all the way to data centers and eventually to end users. Today, we're going to talk about almost all of the ecosystem. There's so many people to thank. I love my ecosystem here. I mean, there's so many companies here and some of my favorite ecosystem partners. So many. Taiwan's rich ecosystem, the richest ecosystem, the world's best supply chain ecosystem. Unbelievable. Well, thank you all for being here and this year our businesses together are growing incredibly. In fact, somebody told me last night that the annual GDP of Taiwan is going to grow almost 10%. Unbelievable. Well, we have a lot to talk about. Let's get going.
Two years ago when I was here I started to talk to you about how AI has moved from generative AI and the other waves of AIs that are coming. The next wave of AI was agentic AI and today we can say that agentic AI has arrived that useful AI has arrived. Now what does this mean? This is GitHub. This is of course one of the first applications of Agentic AI is software coding. One of the most valuable professions, incredibly large ecosystem, 30 million, 40 million professional software developers, probably another couple hundred who are students and enthusiasts, but say 30 to 40 million software developers in the world code for a living. And this represents most of them. This is GitHub. The pull request is when they download software, they modify it, and commit is when they push it back up. And so if you look at this in 2023, the number of commits was 300 million. 2024, 400 million. 2025, 500 million commits in the first few months. In the first few months of 2026, it has nearly tripled. Now, what does that mean? 30 million software developers representing about $3 trillion worth of GDP producing that's what they are paid. $3 trillion worth of salaries per year, which is generating economic growth for the rest of the industries. Say a hundred trillion dollars of the world's industries is impacted by $3 trillion worth of salary. That $3 trillion of salary is now producing nearly three times as much output. It's effectively a $9 trillion productivity from $3 trillion of salaries. The difference is absolutely extraordinary. This is the potential. This is the promise of AI. The number of software engineers is actually increasing. People talk about AI reducing jobs. Complete nonsense. It's causing more software engineers to be hired. The reason is simple: if you can hire a software engineer and generate $9 trillion worth of productive work, why wouldn't you hire more? If that line was flat, people would hire fewer. But because the output is incredible, people want to hire more. This will show up in our economy soon. So the first thing is useful AI has arrived. Now from the industry's perspective, tokens are now in extraordinary demand because if you can do this, you want to produce more. And because tokens are now profitable units of revenue. The AI companies want to build a lot more tokens, generate more tokens, build more AI factories, which is why compute demand in Taiwan has skyrocketed. It is the reason why all of you are so busy and your businesses are doing so well. In fact, that looks like some of your stock price.
The compute pattern has changed. Everything has changed. So the first idea is that useful AI has arrived. AI is now a profit generator. AI is now a GDP generator. Behind it is a whole new kind of computing pattern. Not just a large language model, but an agent. Today almost everything we're going to talk about is based on this. Let me show you what I'm talking about. This is an agent application. In the old days, this would be application, code, operating system. Today it is an agent which consists of a large language model or many sitting inside a harness that orchestrates it to do productive work. This is the input. When input comes, it has to understand, observe, reason, act, use tools. That tool could be a spreadsheet, web browser, data processing engine, database engine. This is orchestrated by the harness. Every time it touches processing context, understanding, reasoning, coming up with a plan, it acts. This is fundamentally an agent. It deals with short-term memory called working memory, long-term memory. Memory management is incredibly important. This entire system is called an agent. The large language model does the thinking and the harness connects everything like an operating system. This is the new computing model. The simultaneous convergence of large language models that can think, reason, plan, use tools, and harnesses that manage memory and orchestrate tools allows us to do amazing things. For example, this is a prompt: create a GIF of NVIDIA gen green dots on black scatter form Taiwan 101 building, morph to GTC Taipei 2026, morph to Nvidia logo, then scatter and repeat. That was the prompt. Here's another: I lost my remote control battery clip. It looks like this. Create a CAD file ready for 3D printing to create a new one. This is the new computing pattern. Instead of launching an application and clicking and typing, we explain our intent and the AI generates the code or uses tools to produce the output. This is how computers will work in the future. This is agentic AI. For two years we've been building towards this and now it has arrived. A big breakthrough is tool use. Some say AI will put software companies out of business. I say exactly the opposite because there will be so many agents. The world is no longer limited by the number of people. Those agents will use more tools than ever. This is an incredible time to be a software company. But the software has to be presented to the agent in a way it can use. Nvidia's treasure is our CUDA X libraries. Today we can present these libraries to agents who can use them much more effectively than even humans. This is a wonderful time for CUDA X libraries. Let's take a look.
20 years ago, we built CUDA, a single architecture for accelerated computing. We reinvented computing. A thousand CUDA X libraries help developers make breakthroughs in every field of science and engineering. CUDAX libraries are tools for agents. CU litho for computational lithography, Coop for decision optimization, CDSS for direct sparse solvers, AIQ for deep research across structured and unstructured documents, Aerial for AI ran, warp for differentiable physics, parabrics for genomics. At their foundation are algorithms and they are beautiful.
Heat. Heat. Heat up here. Heat up here. Heat up here.
A round of applause for math. Math is beautiful.
The computing pattern of software is going to change. Let's come back to this. This is the agent. It is the ultimate disaggregated and distributed computing model. So many different computers are activated to process this agent. The agent consists of model, harness, tools, skills, and a runtime. All running in different places in a data center. Think of the model as the brain, the harness as the body, the tools as a workshop. The large language model thinks, processes context, observes, understands, reasons, plans, and acts. Every time that happens, an entire rack of Grace Blackwell MVLink72 is activated. Whenever it uses a tool, a CPU is used. That tool could be a C compiler, Python, JavaScript, or accelerated computing. Today's agents are simple tool users. Tomorrow they will be sophisticated. That's why CUDA X libraries will be incredibly popular with agents. They solve important problems. All our CUDA X libraries now come with skills that AI can learn. The tools run on CPUs, GPUs, and large language models. The security harness runs on CPUs and a DPU (Nvidia BlueField). Orchestration runs on a CPU. The hardest part is memory. Working memory is KV caching. The memory system will revolutionize storage. Every aspect of this computing model is fundamentally different. This is why we built Vera Rubin. Vera Rubin is not one chip. It starts with the GPU but is an entire system: GPUs, Vera CPUs, storage, CX9, DOA software, security processor. Everything is encrypted. This system obeys confidential computing. Vera Rubin is the most ambitious endeavor in our history. 40,000 engineers worked on it. All of you participated. Vera Rubin is a miracle. It's more than a chip. Nvidia has evolved from a GPU company to a systems company to an infrastructure company. We help customers build AI factories. Let's take a look.
The world is racing to build AI factories. The largest infrastructure buildout in human history. AI factories are complex. Every layer must be designed together because compute is revenues. NVIDIA DSX is the blueprint. It starts with DSX SIM, an Omniverse blueprint where partners design and validate a Vera Rubin AI factory before a single rack lands. They plan layout, simulate power and cooling, design the network, test changes in the digital twin. The factory powers on. DSXOSS provisions, operates, monitors, and remediates. Today's AI factories overprovision power by 40%. DSX Max LPS lets operators deploy more GPUs within the same power budget, adding billions in revenue. Hot liquid cooling at 45°C uses less water and energy. Dynamic power allocation recovers stranded watts. In-rack power smoothing flattens spikes. AI agents continuously balance cooling and power. DSX factories are flexible energy assets that cooperate with the grid via DSX Flex. 100 gigawatts of AI factories will come online before the end of the decade.
I've shown you ecosystem slides of the past. That was a computing ecosystem. This is an AI factory ecosystem. Upstream of me is all of you. Downstream is this ecosystem. Nvidia is not just building a GPU or a system. We help customers build these AI factories. Each one gigawatt level started at $30-30 billion, now $50-60 billion, soon $80-100 billion per gigawatt. It must work the first time. The complexity is incredible. We used to design a chip inside a computer. Now everything is built in Omniverse. We can build these systems in a digital world before breaking ground. This is our ecosystem: RTX for GPU, DGX for systems, DSX for infrastructure. Because of our work, small companies can become world-class AI clouds. CoreWeave is worth $50-70 billion. We worked with Nebius. Each cloud has incredible customers like Cursor, Black Mountain Labs, World Labs, Revolute, Shopify. Nscale has British Telecom and Google. Neighbor Cloud in Korea has Bank of Korea and Hyundai. Yoda in India. Together AI in Singapore. GMI in Taiwan. All need the computing stack. But the AI cloud is so complex that Nvidia has become an AI infrastructure company. Compute is revenue. Compute is profit. Absence of revenue and profit is loss. The curve of AI infrastructure matters. Our fully integrated capability means faster time to first token, higher throughput per watt, and better reliability. We design everything from the ground up and use extreme code design. If your data center has one gigawatt, throughput per watt is revenue. Choosing the wrong architecture because chips are cheaper doesn't make sense. The more you buy, the more you make. Tokens per watt. Finally, reliability. These data centers have millions of moving parts. Our experience with large scale matters. The lifetime of systems is also important. Software changes every four years. Our architecture's flexibility and rich ecosystem ensure a long useful life. Nvidia systems are used worldwide. Software developers start with CUDA. The asset's life is longer. The TCO is low. The more you buy, the more you make.
Now, all of you are experiencing this with me. Isn't that right? All your demand, your factories are working so hard across Taiwan because everybody wants to make money. They realize useful AI is here. Profitable AI is here. Compute demand is incredibly high and it's the constraint. So let's work super hard and help the world stand up AI factories everywhere. I'm so happy to stand in front of you. Vera Rubin is in full production. The supply chain for Vera Rubin is twice as large as Grace Blackwell. It used to take two hours to assemble a Grace Blackwell rack. Now it takes five minutes. Capacity is higher, throughput is faster. The ecosystem is extraordinary. Millions of square feet have been put online to support Grace Blackwell and now Vera Rubin. I want to thank all of you. Vera Rubin is now in full production. Thank you.
Large language models generate answers. Now AI agents can do work. Processing agentic AI is different. Agents observe, reason, plan, use tools, manage context, working memory, and long-term memory. They spin up subagents. NVIDIA Vera Rubin is a multi-rack pod-scale system built for agentic AI, now in full production. The manufacturing automation across the supply chain is a miracle. Our journey started with DGX1. Over a decade we pushed every chip and system to the limit: Pascal, Grace Blackwell, and now Vera Rubin. It starts at TSMC. Seven new chips on 3nm process with HBM4 memory. The compute board has 6 trillion transistors and 18,000 components. Vera Rubin NVL72 does thinking, reasoning, planning. New modular compute with PCB midplane, X9 NICs, Bluefield 4 DPUs, all cableless. 18 compute trays, 9 NVLink switch trays, liquid cooled bus bars carrying 5,000 amps. 1.3 million components form the third generation MGX rack. Congratulations to Microsoft, Dell, CoreWeave for their engineering racks. The Vera CPU rack has 256 CPUs in one liquid cooled rack. Vera Bluefield 4 STX handles memory and storage. Spectrum X Ethernet switch with 200 Gbit co-packaged optics. Vera Rubin: five connected rack systems. 150 supply chain partners in Taiwan. Millions of square feet of factory floor. Extreme code design. We did this with Taiwan. Together we reinvented computing for the age of AI. Thank you, Taiwan.
Ladies and gentlemen, Vera Rubin. Vera Rubin was not built just to run AI. It was built to run agents. This is an agentic system. It has taken years for agents to realize their potential. The computer that runs it is the most advanced in the world. Can we bring out Vera Rubin? And Janine, do we have the racks? This is Vera Rubin MVLink 72. This is the Gro LPX. This is Vera CPU rack with 256 liquid cooled CPUs. This is Vera Bluefield storage and security system. And our Melanox networking, the world's first CPO. When we built Hopper, it was for pre-training. Then Grace Blackwell for inference. People said inference is easy. But inference equals money. The models are complicated. We created MVLink 72. Today Nvidia's token cost is the lowest by orders of magnitude. With Vera Rubin, it's beyond inference. It's inference in an agentic system. No cables, no hoses, no fans. What used to take two hours now takes five minutes. The reliability is off the charts. This is our Vera CPU tray, the most advanced CPUs. Two Vera CPUs, four CX9. This is our Gro system for low latency inference. The Vera Rubin MVLink switch tray is revolutionary. Nvidia is now the largest networking company. I'm so proud. Let's talk about CPUs. Vera CPUs built for AI. All previous CPUs were built for humans. Agents live in nanoseconds. They need low latency. So we created Vera CPU. Inside our system, it's used three ways: in Vera Rubin for thinking, with two CPUs orchestrating GPUs and managing KV cache; Grace BlueField for security; Vera compute for harness, orchestration, tool use, database access. Vera BlueField is the fastest storage system. The economics of the AI factory is tokens. Vera CPU is built for agents. Four key properties: instructions per clock (IPC) is the highest in the world at 10 per clock; bandwidth per core and total bandwidth are world class; energy efficiency. The fabric connects all cores at speed of light 3.6 TB/s. No chiplet taxes. It's the first PCIe Gen 6 CPU with LPDDR5 at 1.2 TB/s. Three times the bandwidth of the highest performance CPUs. The number of CPUs will be high because there will be billions of agents. These CPUs must be performant and energy efficient. Vera is off the charts compared to x86. This kind of performance speedup is unheard of. This is Nvidia Vera. Let's take a look.
Agentic AI changes the role of the CPU. The CPU is now the conductor and the GPU is the orchestra. Traditional CPUs were built for a different era: maximizing cores, virtualizing, renting by the hour. In the age of agents, the CPU is a bottleneck to GPU utilization, affecting token throughput and latency. NVIDIA Vera is the CPU built for the agentic loop. It combines NVIDIA's custom data center CPU core with a scalable coherency fabric for the right balance of performance, cores, and bandwidth to maximize AI factory output. At the heart of Vera is the NVIDIA Olympus core built for modern data center workloads, branch-heavy Python runtimes, tool calls.
And sandbox code execution. Each core is tuned for throughput. A neural branch predictor evaluating two taken branches per cycle. A 10-wide decode engine, a large out-of-order engine, advanced prefetchers with a novel graph engine. Vera is the first CPU to use LPDDR5X memory while correcting multiple errors simultaneously. Vera achieves 40% lower peak memory latency versus x86. NVIDIA's second-generation scalable coherency fabric unifies all 88 Olympus cores on a monolithic mesh with separate dies for memory and IO, enabling 50% faster core-to-core communication. Memory-coherent NVLink chip-to-chip connects GPUs directly to the fabric and can scale Vera up to multiple sockets. Vera delivers 1.8 times the agentic sandbox performance of x86 CPUs. Standalone Vera racks run agent sandboxes, tools, code, and data pipelines. Tightly coupled to Reuben GPUs, Vera keeps accelerated workflows moving. NVIDIA Vera Bluefield 4 STX powers context memory and AI storage. Compute, networking, storage. Vera is the CPU for the age of agents.
This is going to be our new major growth driver. The reviews are already coming out and it's pretty good. Now remember, Grace and Vera are also the most highly qualified CPUs in the world of AI because every data center, every cloud, every enterprise that works with Nvidia on AI has already qualified Grace. The entire software stack has been optimized for Grace. Every company will be qualifying Vera. Vera will be the most optimized agentic CPU in the world simply because it's going with Vera Rubin and we made the big hard switch from external CPU x86 to Grace Blackwell. That transition was extremely dangerous, but we did it with incredible execution. Now Grace is synonymous with Grace Blackwell. Here comes Vera. I'm super excited. Look at the performance numbers. SQL running three times faster. Real-time stream processing for the New York Stock Exchange: Vera CPU is six times faster all because of bandwidth, single-threaded instruction execution, bandwidth between cores, bandwidth outside. Vera is completely revolutionary. That's Vera. You know, X factors are something you talk about when you're talking about GPUs. It is quite rare that somebody talks about X factors on real workload associated with CPU. So I'm so proud of the team. We have an extraordinary roadmap coming. But what's really exciting is almost everybody is supporting Vera. Agents is a new workload. We built CPUs for humans in the past. We need CPUs for agents. Agentic systems. Their properties are different. We are building millions and millions of units to go to market with Taiwan's ODMs, computer makers, and OEMs. The early adopters are the agent companies. This is the beginning of a new market. CPU for agents. This market will surely be larger than the last because there will be a lot more agents than there are people and the agents are very impatient. So Nvidia Vera CPU, thank you.
This is the most important slide. The takeaway is that this is the application pattern, the computing pattern of the next decade. Agents harnesses orchestrating large language models. Every company will run it. Every company will be an agent company. Every company will have agents running inside. Every company will need its own operating system. So we have the NVIDIA agent toolkit for enterprise AI. You've seen me build this in plain sight. There are four things companies need to build agents as a service: models (large language models), a harness to orchestrate, tools with skills (like CUDA X libraries), and a runtime (operating system). This Nvidia toolkit includes models you can modify, Nvidia's world-class open models. You can run agents from anybody inside this harness called open shell, which is highly secure for the enterprise. Open shell is open source and will be adopted everywhere. Red Hat, Canonical, Microsoft. The runtime is fully optimized for the Nvidia AI platform, which is everywhere. So you can run open shell in any cloud, on prem, or even on device. One of my favorite use cases is chip designers. We partnered with Cadence to build a chip design super agent. Cadence and Nvidia are partnering to build chip design agents. Hundreds of thousands of Nvidia chips come together to make AI factories. Designing these chips is one of the hardest engineering challenges. From weeks to hours, Nvidia has thousands of chip designers. We are going to hire hundreds of thousands of Cadence super agents to accelerate our company. The agents are impatient. They want the answer immediately. Models, harnesses, accelerated CUDA libraries and tools, and the runtime all come together. It starts with a great model that Cadence could modify. We call it Neotron. Today we are announcing the Neotron 3 Ultra, our next open model. It is smart. The Neotron models give you the model and all the data we used to train it. Neotron 3 Ultra is five times faster, 30% cheaper, and completely open. We're dedicated to open models. This entire toolkit from models, harnesses, tools and skills, and runtimes is why every enterprise company in the world can now create their own agents. People always said Jensen, the agents are going to disrupt these markets. I said completely opposite. Agents are going to create the largest opportunity ever for my partners and friends. We have the Nvidia agentic toolkit for enterprise AI to help them. So there you go.
First, Vera Rubin in full production. Two, Vera CPU built for a new generation for agents. And three, Nvidia's enterprise AI toolkits so that every enterprise and every enterprise software company can build agents.
My relationship with you started here. And many of you, many of my friends and partners here in Taiwan, your companies started here. This is in a lot of ways the beginning of the modern computer industry. 40 years now. Nvidia is 33 years old. The PC industry was already starting with Windows and Apple. By the time we came along, Windows 3.1 was the PC, and Windows 95 made PC personal. This computing platform did several things incredibly smart. Windows was properly abstracted: systems BIOS, open chipsets, the operating system with drivers, and an abstraction layer with a multimedia API. 40 years later, Microsoft and Nvidia are going to reinvent the PC. This is going to be the new PC. Tomorrow night I will be with Satya Nadella to talk more about the work we are doing together. It took three years to completely reinvent how the PC works so that we could be ready for this moment. The compute pattern called the agent will run in AI clouds, inside enterprises, and on your PC. What happens to that PC when it has an autonomous agent? An agent that helps you, understands you, you can talk to it, it can look at you, read files, do research. The new operating system is the old operating system plus large language models. LLMs are the modern version of DirectX. The application will be replaced by an agentic runtime. Let's take a look at what it can do.
