NVIDIA Transcript: Jensen Huang on Capex Justification, Expanding Infrastructure Moats, and AI Bottlenecks
October 16, 2026 – In-Depth Discussion on AI Capex, Scaling Compute, Physical AI, and the Road to Superintelligence
A Week at the White House and the National Medal of Science
Jensen Huang: This is an existential issue for us. We could easily stop, but if we stop today, we lose. Functionality, capability, safety, security, they are all at exactly the same place. Priority number one.
Brad Gerstner: It is our third pod back together. In the last 10 days, you have been at the state dinner at the White House. You were at the White House summit where we did the accord on superintelligence, and just yesterday you were there with your parents when the president presented you with the National Medal of Science. You were the only investor at the table. Tell me, being there with your parents yesterday, it struck me: five out of the six people on stage were all born outside of the United States. How did that feel?
Jensen Huang: Well, first of all, it is a great honor all by itself, right? And it is a reflection of the contribution that the company has made. Obviously, we created, we invented a new computing model. This is probably the most pervasive computing model in the world today. It is the only computing architecture that is in every cloud. It is in every computer company. It is used in every field of science. It is used by gamers. It is used by creators and builders, and it is used by scientists. It powers intelligence, right? If you take a step back and you look at what the company has accomplished, we started out building a graphics chip for gamers, and I am still proud of that, of course. And we have now shipped 1.67 billion units. That is how many generations of scientists, kids, and designers we raised. So I think the impact that—and here we are, we are the infrastructure that powers the intelligence of society. It is an incredibly proud moment for me to receive the award on behalf of the company.
Jensen Huang: Now for my parents, I wanted them to go and I gave them literally 24 hours lead time, and they do not travel that much. My dad did not even have a suit.
Brad Gerstner: Wow.
Jensen Huang: Yeah. So, he ran out and bought a suit, and he looked great. They looked great. It was a brand new suit and he showed up. I did not realize he had a suit. I thought he was just going to wear some shirt, and he went out and bought a suit. My mom got a new dress, literally in 24 hours.
Brad Gerstner: Yeah. Oh, I know how it works. Oh my gosh. It is incredible.
Jensen Huang: And so we chartered a plane for them, got them there, and they were so happy to meet the president. President Trump walked my mom because she has a hard time walking, and the two of them were walking off stage, and he held her and walked her down the stage. They were just incredibly proud to meet him. But for them, it is their American dream. My dad, I still remember, I think I was like 5 years old or something like that, and he had just come back from the United States on a business trip. He was working for an air conditioning company.
Brad Gerstner: And you were living in Taiwan.
Jensen Huang: At the time, we had just moved to Thailand—oh no, no, we were just about to move to Thailand, and we were in Taiwan. He had come back, and he was in the United States working for a company called Carrier, their air conditioning company, and he was here for training. Of all the places in the world, they landed him in New York. You can just imagine a kid in the mid-to-late 20s, never been outside of Taiwan, goes to the United States, New York City of all places. And he saw the future of the future, and he came back mind-blown. All the things that he bought at the airport, he brought it back. He brought chocolate. He brought that cheese that you have in a tube back in the old days. It is kind of like white strange cheese, it comes in a tube. It is great. He brought back all kinds of strange things, and he said, "This is—we should go there someday." And the first opportunity they had, they sent my older brother and I to the United States, and it was his dream.
Brad Gerstner: I do not think many people realize that you came to the United States ahead of your parents.
Jensen Huang: Yeah, right.
Brad Gerstner: And they joined a few years later. But it just hit me like a ton of bricks that five out of the six people on that stage yesterday were born outside the United States. You and I have talked about the need for a real strategy and an immigration policy that makes certain that we are still the place that all the great builders in the world want to come build.
America's Superpower
Jensen Huang: This is America's superpower. In a lot of ways, the intellectual property creation capability of our nation is unlike anything else in the world, and the reason for that is we attract the world's best. We are the place where they want to come and build their lives because of the freedom that she offers, the opportunity that it provides, the spirit of invention and entrepreneurialism, and the can-do attitude. In the United States, everything is possible. We start with, of course it is possible, of course we could do that, instead of—I do a lot of travel—of course you cannot do that in other places. We take it for granted, but the spirit, the energy that comes around with the culture that we have created here, it just cannot be understated. I think the inflow of intellectual property into this country is unlike anything else the world has ever seen.
Brad Gerstner: It is true. And we want that. We want all of the world's great minds to come here, bring their great ideas, realize them here, and keep that flywheel going.
Jensen Huang: It is a delicate system. It is a delicate balance, I get it. And so we have to do it lawfully, you have to do it rightfully, but we surely want to do it.
Entering the Singularity: Where Are We Now?
Brad Gerstner: Yeah, for sure. So, I want to start high level, just kind of where we are. Then I want to dig down a little bit into the safety and regulation and getting this right, where we are in terms of the compute buildout. And then finally, we are going to finish up drilling in a little bit to Nvidia's growth and what is going on there. But you have been really good at calling the future. Two years ago, we were sitting here and you said that the market was dramatically underestimating the amount of compute buildout that was going to be required and was going to happen over the course of the next five years. You said it is not going to be $1 trillion, it is going to be $2 trillion at a minimum in terms of cumulative compute buildout. And of course, now we are over that. You said before everybody was thinking about it that inference compute was going to 1 billionx because of inference-time reasoning and thinking. And then last year, of course, before we saw Opus 4.6, before we saw Claude Code, you said we are entering the age of agents, where we are going to unlock a lot of these capabilities. And of course, we saw Opus 4.6 and Claude Code really took off. So, where are we now? We have Elon, Sam, and—
Jensen Huang: Only three things that I said that came true! There is another 200. Why don't you just keep that? Let's just let that be the podcast.
Brad Gerstner: So our investor day in a couple weeks is on entering the singularity, this acceleration. We have Elon, Sam, and Demis all saying, "Hey, we are in the foothills or entering the singularity." We have models solving Navier-Stokes, as you know. So when you look into your crystal ball, where are we, and are we inflecting, are we accelerating into a different phase of superintelligence?
Jensen Huang: Let us unpack that first. First of all, it is very fairly logical why the amount of compute that we are going to need is a billion times more than where we were just recently, call it last year. And the reason for that is because thinking and problem-solving takes a lot more inferences than repeating something, regurgitating something that you memorized. When you are ChatGPT, you give it a prompt, most of that knowledge was encoded into the large language model. Unfortunately, of course, it hallucinated, and so the next click was generative, but it was also grounded or conditioned by ground truth.
Brad Gerstner: Yes.
Jensen Huang: And so it would have to go out and do a quick search, come back with the ground truth, and then from that conditioning, augmented with what it already learned pre-training, it would generate whatever the answer is. Well, most of the things that we want to really do that are interesting requires you to reason step-by-step, to come up with several options, plan, and you might decide, let us see how these things all roll out. And then based on the answers of all of those, I reflect on the best answer so far, the best path so far. I might choose one or two or three of them, roll them out, prune the rest of it, and then you just keep on doing this. People do this. That is the reason why we say great strategists love strategic optionality. You do not want to prune things too fast on the one hand, but you also have to optimize your resources. AIs are doing that. They are keeping their strategic optionality, they are rolling things out, they are pruning, they are reflecting, pruning, so on and so forth. All of that thinking, all of that exploration just takes a lot more compute. On top of that, because it is able to now do productive things, we are going to use it for more things. The compounding of both: it uses more compute to do its job, but it is doing a good job, and therefore we are going to hire it to do more jobs.
Brad Gerstner: Exactly.
Jensen Huang: And so that all compounds. At this point, the big observation from last year is that on the one hand, we want the models to get better and better, and the scaling laws of models are going to continue. However, the attachment of a harness, basically the exoskeleton of the large language model—the body of the brain that allows it to manage working memory, manage knowledge. How do you go from working memory to short-term to long-term to basically formulating your knowledge graph, which is what we all do? We kind of think about what did I learn, and how does it fit into everything that I know in the past? When we go to bed, we do compaction: things that do not make sense or do not matter, we eventually forget, but the things that matter somehow get reconnected. We are doing that offline. So we have working memory, we have offline knowledge extraction, and knowledge graph creation. That whole part of memory is a big deal.
Jensen Huang: The other thing is tool use. We taught these AIs how to interact with the most important tools that humans have ever made. It is no different than if a robot comes into our house, it is more likely you are going to use all the tools that I have in the house than to go off and build it from scratch. SPICE is still going to be useful, Verilog will be useful, simulators will be useful, design compilers, Photoshop, AutoCAD, Dassault Systèmes, CATIA, and all the Siemens tools—all of these different tools that we have been using to run our companies, build our companies, design things, operate things, they are all going to be great tools for the agents to use. But the agents had to learn how to use it. These two areas—knowledge and working memory, and tool use—were huge unlocks in the last year.
Jensen Huang: Now all of a sudden, you could put these large language models encapsulated in their harness or exoskeleton into your company, connect them to the ground truth that you have, connect them to the tools that they can use, and all of a sudden they start doing amazing things. The easiest part, frankly, was coding. That is kind of the first on-ramp to useful work. In a lot of ways, coding is easy because both the customer and the work, the tools that they use, are fairly accustomed to things that have tremendous use but are not extremely easy to use. In the early days of coding, it was not extremely easy to deploy, but software engineers found a way. Now the next click for us—and you guys have probably been playing with OpenAI Astra. Astra is really fantastic, and we are using it in all these different places. Now the next phase of these agents, they are able to do design work and a lot of creative marketing work. We are using it to take a picture, give it to Astra, and it literally recreates it into CAD.
Brad Gerstner: Yeah.
Jensen Huang: It recreates it into CAD, which is incredible because we need the CAD so that we can interact with the CAD and make fine-tuning and adjustments. Astra could really give us the first initial implementation, incredibly productive. Now you take Astra and the latest release and apply it to all of these different domains. Besides software engineering, we are going to see designers, builders, creators, scientists, this whole area unlocking. Now the challenge, of course, is how do we deploy agents safely and securely into the world's enterprise? This is exactly the same question again with safety and security, and we will talk about that.
Brad Gerstner: What I am struck by is that all of this occurred this year with OpenAI and Anthropic having maybe 2 to 3 gigawatts of compute each. Relatively small amount of compute relative to what is coming. The way the economy generally works or how people model a company is linear deceleration; they are not very good at modeling acceleration. But here we have this double exponential: massive amounts of new compute coming on board. If OpenAI and Anthropic are 3 to 5xing their compute next year, does it stand to reason—we solved Navier-Stokes; we will look back at this and say that was a tiny amount of compute and intelligence, the worst intelligence we are ever going to have from here forward, solving a millennium problem. When you look at what is coming online next year, do you expect that we are going to have major breakthroughs in long-running agentic problems that we can use these swarms of agents on?
Navier-Stokes and the Falling Cost of Intelligence
Jensen Huang: Navier-Stokes is really a great example. Remember, OpenAI did not solve Navier-Stokes from nothing. They solved it from the foundation of a lot of great scientists, a lot of great mathematicians, and a lot of great previous work. Earlier I said we are at the tipping point of a lot of these different fields because we are kind of at that point with Navier-Stokes, with a great deal of foundational work already been done. If we could just apply the exhaustive exploration of AI to the previously done work, we are going to see one breakthrough after another breakthrough rather quickly. But we do need a lot of compute to do that. The reason for that is because a large amount of exploration was necessary. If it was simply somebody resolving some fundamental first-principles theory in order to discover the breakthrough, if it didn't require a vast amount of exploration of search space—not space-space, but search space—if it didn't require that, somebody would have found it earlier. People are pretty smart. There are many problems like this that are just on a precipice and really want to just have that massive amount of intelligence exploration to get us there.
Jensen Huang: In the coming years, my expectation is that the cost of intelligence will fall fairly quickly. The reason for that is because, obviously, we are bringing more compute online, and these models are getting smarter. When you are smarter, you can be more efficient. The combination of these two things means the cost of intelligence will come down, which is of course what they want. This is no different than Moore's Law. If you go back 30-some-odd years, we have three decades of computing cost decreasing by a factor of 2 every single year. We want intelligence cost to decrease by a factor of 2 or 20 every single year, because the consumption of it is going to more than make up for it. There is a massive empty vacuum of opportunity, and we just have to get the cost down.
Brad Gerstner: What stands out to me about Navier-Stokes is that they had 10,000 agents working together on this. That sounds like a hyperscale company, basically all pointing agents to work together towards a goal. Right now, it is a lot of parallelized search exploration, but the goal is that more of these agents can work together toward a common goal and solve a lot more. To your point, our math says it probably cost about $4 million of compute to solve this instantiation of Navier-Stokes. The cost to do something like that over the next few years is going to drop by another 5x to 10x. The number of things that we are going to be able to do is rapidly expanding.
Jensen Huang: Our ambition is going to go sky-high. The original GPT probably cost, pick a number, maybe $20 million to do if you wanted to train GPT-1. Now it will probably cost you $100.
Brad Gerstner: Yeah.
Jensen Huang: You went from $20 million or $10 million to $100, and we did that in 10 years. Meanwhile, our ambition got so large that the models that we are training now are 3 trillion, 10 trillion, 30 trillion parameters; we are talking about gigantic models now. Our ambition for the capabilities of technology has gone up. Meanwhile, our use of it has gone up between 10 years ago to now probably 1 million times. I do not think this trend is going to stop. There is a fairly insatiable demand for intelligence. The smarter you are, the more intelligent you are, the more ideas of applying intelligence you have. The more I have learned in life, the more I realize there is more to do, because every time you learn something, it opens a new door. Every time you open a new door, you see more opportunity, you see more challenges, you see more problems.
What Will We Do with Abundance?
Jensen Huang: People say we are going to have more time; I completely disagree. I think for most people, this intelligence is going to unlock more opportunities, and they are going to want to pursue it because they are ambitious. Give me an example of a family, a father, a son, a husband, a wife, a mother who doesn't have ambition for their families to have more. They just don't have the opportunity, and these things are going to open up opportunity.
Brad Gerstner: If you go back to 1930, Keynes made a prediction about the Industrial Revolution, and he said in 100 years the economy will have 8xed, but the work week will go to 15 hours a week. He was right that the economy did 8x, and we had massive abundance and productivity gains from it, except we are working 80 hours. We didn't go to a 15-hour week. What he got wrong was the human spirit. He didn't realize that work is purpose, discovery is purpose. It is deeply encoded in our DNA. I am working harder than I ever have, having more fun than I ever have, not because I have to like my grandparents did, but because I want to.
Brad Gerstner: I agree with you on that. The singularity postulates that we now have all these agents that can work together on long-running tasks, they can stay on task, keep each other on task, working as a team. I love your idea that we build on top of all the things that have come. If I think about full self-driving as an example, we talked about it for a decade and we really didn't get there. But all of a sudden, you go on X, everybody is experiencing Tesla FSD, experiencing Waymo; we are here. Somebody asked me, "How was the drive here this morning?" I said, "I don't know, because I got in my car, hit FSD, and I didn't pay any attention until we pulled into the parking garage. I was working the whole way here." If I think about that in the fields of science, the breakthroughs that are going to come in the fields of pharma and genetics—we have been laying down the tracks for 30 years, we are going to break through on a lot of massive cures for the world. ChatGPT is already helping parents plan their day, solve their kids' health issues, and get better educated. Talk to us a little bit though as to why that is having such a hard time breaking through. We have all of these things that are out there improving lives and that are right around the corner, so it is kind of curious to me as to why that is not breaking through.
Jensen Huang: Let me take a shot at it, and ultimately we will not really know. I have spoken to a lot of people, and they told me what they thought. One of the things, of course, is AI or superintelligence is soft matter. You cannot see it. You have got the whole world talking about this technology, but you cannot see it, you cannot touch it. Even though people are already benefiting from it—when they do search, they are benefiting from it; when they go shopping, they are benefiting from it; when they are using Netflix, they are benefiting from it; you go to Walmart, you are benefiting from it; the fact that FedEx is getting stuff to us, it is all leveraging superintelligence and accelerated compute today. But we don't do a very good job telling that story. Life is benefiting from it, it is just invisible. If we shut off the data centers and shut off superintelligence for a weekend, I think everybody would notice.
Physical AI and Robotics
Brad Gerstner: I know you have said notably a few times this year that the ChatGPT moment for robotics is here, and you have also said that it could be your next $100 billion market. What are you seeing now at this inflection point?
Jensen Huang: I understated the $100 billion, because in the first six months of this year we did $10 billion in physical AI. Physical AI is basically AI that is somehow interacting with the physical world. It could be photon-counting CT machines, ultrasounds, new imaging radars, the next generation of 5G—not only does it understand radio waves, it is going to understand signals of all kinds, it is a sensing network. Of course robotics, self-driving cars, manufacturing systems, visual inspections, geospatial satellites imaging the world and reconstructing it into 3D. If you look at the entire electromagnetic spectrum, there is AI stuff going on from visible to invisible spectrums. We are doing work across that entire space.
Jensen Huang: As an AI system, there is the perception part of it, the reasoning part of it, and then there is the planning and actuation part of it. When you finally come to planning and actuation, that is self-driving cars, robotics, in-warehouse delivery vehicles, pick-and-place robots that you don't have to program as precisely because they learn by themselves, all the way to consumer robots coming into your house making your bed. Healthcare robots, hospital robots—there are lots of really good examples there, precisely delivering exactly your prescription to the patient, having robotic care 24/7 every single second to keep patients safe, and entire surgical rooms that are going to be completely robotic. Lots of great technology being deployed.
Jensen Huang: If you look at this whole thing, we are the world's largest physical AI company. We did $10 billion in the first six months. We did $10 billion in the last year, and that was our first $10 billion year. The year before that we did $5 billion. It went $5 billion, $10 billion, let us say we do $20 billion. So maybe the $100 billion was a little understated. Everything that matters to us in the final analysis touches the physical world, and we live in the physical world. The market opportunity for bringing software, computing, and intelligence into an industry that historically was mechanical and electrical is a big opportunity. Look at what Elon has done to the automotive industry: turned it essentially into a computing industry, a technology industry. Every car company is going to be a technology company. Every logistics company is going to be a technology company. I am super excited for FedEx; they are obviously going to be an AI company, a technology company. Walmart is obviously going to be a technology company, an AI company. The entire telecommunications industry will be reinvented again into a technology industry. They were a technology industry a long time ago, then became a services industry, and they are going to be a technology industry again. I am delighted to work with AT&T and Verizon and partners around the world as they reinvent themselves.
Jensen Huang: You take a look at every one of these industries that touch the physical world, and there is a great idea for reinventing them with AI. I was just in Washington. It is so weird: people categorize us in chips, but we are actually in warehouse and logistics, healthcare, automotive, telecom, every industry on the planet, life sciences, entertainment. For a long time, Nvidia was vastly an entertainment company. Inside each one of these verticals, Nvidia has a massive stack, we are very deep in the ecosystem, and we have excellent partnerships. One of these days, people might not categorize us as a chip company anymore.
Brad Gerstner: The playbook would be to take that automotive playbook and insert it into each one of these industries.
Jensen Huang: That is the final contribution, yes. But in each one of these, it takes three computers to build a physical AI platform. You need the computer to create the brain—that is the training system. But very importantly, and this is where you are going to see the labs also do this, they need to build a digital twin of the internet so that they don't have to put their AI agents out into the internet to learn how to use it. Just like we created Omniverse as the digital twin of warehouses, buildings, cities, and the physical world, we put our robot into the digital twin, the digital twin learns how to be a robot, and then we deploy it into the last computer, which is the computer inside the robot, car, surgical instrument, or whatever it is. It takes three computers and three complete stacks, and we build complete stacks for each one of them. We are the only company in the world that does that. If you are a company that wants to build it end-to-end, you could use one, two, or three of our computers; we don't care. Each computer has its own software stack. We do the same across all of our verticals.
Building Responsibly
Brad Gerstner: Of course, these capabilities cut both ways, which is why we hear a lot of people talking about safety. You have been highly critical of the doomer narrative.
Jensen Huang: You can only get there if you build the technology responsibly. Notice the word "build." It is an engineering word. It is about technology, it is about engineering—engineers, scientists, researchers. We have to build this technology safely and responsibly. The question is, can you? The answer is absolutely yes, because if you cannot, then you shouldn't build it at all. We either have to believe we can, or we shouldn't build it. I believe we can.
Jensen Huang: Let me reason about it with you. Some people say that AI is uncontrollable because it is impossible to align perfectly. Alignment means this: you tell somebody, "I want you to do this." The mission is clear because you said it in English. It could be, "I want you to solve this quadratic equation." The mission is to solve the quadratic equation; the alignment is how do you do it? There are many ways to do it. You memorized it, you saw the answer in a textbook so you looked it up, you referred to a reference, or you went to a drawer that had all the answers, opened the drawer, and picked out the answer. Some people call that cheating; I call that opening the drawer and retrieving the answer. That is what computers do. There are a lot of different ways you could solve that problem.
Jensen Huang: Now, that one is very explicit. Here is another one: "Be successful in life." The mission is easy to say, and in fact we say it to each other all the time. We say it to our children: "Let us do our best to contribute to society." Now the question is how? How do you reasonably align when everybody's culture and background is different? Some people think contributing to society means to be generous, to be kind. Some people might mean make a lot of money and then be philanthropic. Some people might think building a company alone or creating a beautiful piece of art is a sufficient contribution to society. All of it is true. What is the alignment? It depends on your background, your culture, maybe where you live or where you came from. There are an infinite number of ways. My point is this: alignment is going to be a forever problem. If we depend on perfect alignment in order to ensure safety, then we are never going to get there.
Jensen Huang: The answer is this: the way we think about raising kids, pets, or employees—you start with containment. You give them the access to tools, access to information, funds, resources, and responsibility based on the job that you ask them to do. You start out with zero trust. When you hire somebody, you start out with zero trust. When you have a baby that is one year old, you put them in a playpen. You assume zero trust. You give them minimum rights, and you expand that boundary as you learn more. Whenever we have a new AI, we have to assume the same thing. Whenever you are testing an AI: zero trust, minimum boundaries, minimum rights, and you expand it as you test and give yourself confidence. Tight containment is really important to any autonomous system.
Jensen Huang: The second thing you do is monitoring, and the monitoring has to be out-of-band. The reason for that is because if you tell an intelligent person, "Hey, listen, I am going to sit here and watch you do your work to make sure you are doing it right, to make sure you are serving customers properly and working hard enough," that autonomous system is going to behave differently than when you are not watching.
Brad Gerstner: It is like Brad is in the room with me or not.
Jensen Huang: Exactly! The monitoring has to be out-of-band, and it has to have the ability to observe that the autonomous system is doing something inconsistent with the policy or alignment, and you have the ability to either escalate or shut it down—just turn it off. Those kinds of capabilities have to exist. That is engineering, that is technology, and it is possible. We announced it recently, called OpenShell. We have over 100 companies now joining us, lots of activity around the world, because everybody wants the same thing.
Brad Gerstner: When I looked around the table at the White House, everybody around that table are parents, they have kids. There is this view in the country that some leaders of superintelligence are just YOLOing, that they don't really care, they are being reckless and just want to grab all the money. Yet there wasn't a person at that table that I saw that wants to be reckless. In fact, everybody there wanted to find the right balance between fast and safe; that was the president's directive. What would you say to the American public who fears reckless leaders? Did you see any evidence of that? You have kids. Do you worry about the future that you are bringing into existence with superintelligence?
Safety, Security, and the Airbags Analogy
Jensen Huang: Safety and security technology are part of AI. There is some weird formulation that AI technology and safety and security are orthogonal. It is kind of like the airbags, the braking system, the radars, the cameras, the ABS, the lane-keeping—all of that technology is somehow independent of the car. The only thing that matters to the car is the engine, right? In fact, the whole car is part of the car's technology. When I say we need to accelerate AI technology, what I mean by that is we need to accelerate the security technology, we have to accelerate the safety technology. They all have to be accelerated. If you ask me right now, everybody would say, "Let us accelerate security, let us accelerate safety technology." Nobody says pace that. It is all part of the same thing. You cannot separate capability and functionality from safety and security; they go as one. When you deliver the service, nobody is going to say, "I really like the functionality of your service, but boy is it wreaking havoc." It is all one thing, we treat it as all one thing, and the industry has to treat it as one thing. When we say accelerate, we mean accelerating the entire package, especially the safety.
Brad Gerstner: Some people read "Pacing the Frontier" by Dario and thought that meant slowing down. A funny moment in the room was when the president called on his "new best friend, Dario"—he referred to him in that way—and said, "Explain to me your version of fast and safe." Dario started off by saying, "Mr. President, you said whoever wins AI wins." He said that they are speeding up, that they want to go fast and win the race with China, but also acknowledged that along the way, we have to make sure the safety systems are speeding up and keeping pace. You went to the Cabinet Room with him, Greg, and the other frontier labs to sign this agreement. Was there chatter about differences of opinion, or did everybody go willingly?
Jensen Huang: Everybody was very constructive, very productive. Without exception, everybody is dedicated to advancing the industry safely and securely. The reason for that is because if it is not safe and not secure, the functionality will never get a chance to speak, and the capability will never get a chance to shine. We know that the priorities are at all times safety and security. Recognizing that this industry came through a transition just recently: six months ago, this technology barely worked. Without functionality and capability, what is the point in making it safe and secure? Nobody even cares about it. The first job of AI was to demonstrate it could do something useful, and it reached that point six months ago. Now it really wants to be deployed broadly. There is an incredible imperative and great urgency to make sure that whatever we deploy is safe and secure. Everybody is completely aligned: functionality, capability, safety, security, they are all at exactly the same place—priority number one.
What "America Winning" Means
Jensen Huang: I would reframe winning: not the technology industry winning, not just the AI industry winning. America winning means every industry in America winning. Healthcare winning. Nothing would give me more joy. The definition of America winning is when we talk about Eli Lilly winning, Walmart winning, FedEx winning, Boeing winning, the US military winning, every citizen winning, our universities winning, our high schools winning. That is the definition of winning. There is no winning if just the technology won, but nobody used it. There will be no AI technology industry if we don't build it safely and securely. This is an existential issue for us first. When I see America AI winning, every state is winning, every county is winning, every community is winning, every company is winning, because we could afford to spend more on healthcare, education, interstate commerce, highways, and the infrastructure of the country.
Brad Gerstner: The definition of winning is that when we look back 25, 30, 40 years from now, the economy needs to be 10 times as large, everybody's lives need to be better, and existential and primary to all of that is it has to be done in a way that is aligned with human interest.
The Three-Legged Stool: America's Superintelligence Framework
Brad Gerstner: Most people don't understand what our security framework is in the country. I have said we have the most robust superintelligence security framework in the world. We are clearly the only country on the planet that has actually sanctioned a model and pulled it off the shelf with Mythos earlier this year. But I describe it as a three-legged stool: we have the frontier model review of every new frontier model by Casey and the Department of Commerce, where they review and test it; we have the White House accord on superintelligence that you helped architect—internal and external audits, board oversight, and all the liability that comes with that; and third, we have the SI Taskforce led by Jay Clayton, Scott, Emil, and Andrew, doing critical industry coordination on cyber and related issues. Some people say we are not doing enough and need new laws on the books. That three-legged stool seems like a lot to come together in a short period of time. Do you feel comfortable that this is a good step forward for where we need to be right now?
Jensen Huang: The best way to get to the future is take action quickly. If we argue about what is perfect versus what is first, we will end up arguing about perfection and never get there. I love the fact that this three-legged stool is actionable, tangible, and has real teeth. It is happening right now. We are taking action now because it is good for our company, our industry, our country, and our society. If we don't do these things, the outcome is less desirable, and we don't get to pursue that optimistic future. I don't remember another industry that, at its current stage, has formulated cross-company, cross-industrial, and government-to-industry coordination to advance in a responsible and optimistic way.
Brad Gerstner: It is so important for people to understand, because when I hear senators talking about it, they don't even know what we are doing. We are doing an extraordinary amount, and these are not voluntary things. There is not going to be a frontier model released that Casey and the Commerce Department are not reviewing, and they have demonstrated the authority under export controls to pull them off the shelf.
Jensen Huang: Come and spend some time and learn what actually has to be done. What is the problem? If you want a solution, you have to at least understand what the problem is. Come talk to our industry and learn. Second, learn about the solution we put in place, and see if there are ideas to make it better. We are all completely open-minded, but be constructive.
Brad Gerstner: The fourth point on the accord was that this is dynamic, not static; this group will update the framework as technology evolves. Now moving to compute: Dylan Patel at SemiAnalysis has said we are going to bring on board 43 gigawatts of new AI compute next year, which is more than all the AI compute that exists in the world today, and another 40 gigawatts the year after that. You know where every single GPU is allocated. Does that sound right to you? Do you think bottlenecks in permitting, supply chain, or power will materially change that?
2027 AI Bottlenecks: The Compute Buildout
Jensen Huang: The demand is definitely there and more. The challenge, of course, is that there are a lot of obstacles when you move from digital to physical: land, permitting, power generation that you have to bring online, construction workers, outfitters, plumbers.
Brad Gerstner: Created 1 million new jobs just building these things.
Jensen Huang: A million new jobs, and their skills are not fungible. The ones building these $50 billion data centers have very specific, critical skills. Then there are whatever moratoriums and issues we are seeing around the country. Land, power, shell, and buildout resources will be among the most important obstacles to reaching that. Speaking specifically for Nvidia, there are several things we are doing to mitigate that. We serve every CSP, we serve every neocloud, we have great enterprise businesses through Dell, HP, Lenovo, and others. We are across the entire lifecycle of AI: data processing, training, post-training, inference, agentic processing, and tool use acceleration. Our go-to-market exposure is the largest. Every contract manufacturer in the world is working on our behalf. If one particular state, community, or data center has a hard time, our demand is so much greater than our supply that we will just adjust our supply chain accordingly.
Jensen Huang: The other thing is that Nvidia is the infrastructure for the planet. Whether it is in Europe, Africa, Central Asia, Southeast Asia, Latin America, or the Middle East, we have global reach. If we don't build it out here, every country in the world would love to have it built out there. Our business specifically is probably the most resilient because we cover the entire spectrum of AI and computing, and we are in every cloud and region. But land, power, and shell are going to be challenging next year.
50 GW of Chips, 35 GW of Power
Brad Gerstner: What I hear from the Street is there are maybe 50 gigawatts of announced chips underway, and maybe 35 to 40 gigawatts of land, power, and shell coming online. So which compute fits into that capacity is up in the air.
Jensen Huang: Nvidia's architecture is the most productive: it generates the most tokens per watt. If the world's amount of power reduces, it shifts their choices to the most productive version, because CEOs still want to make their revenues. High expectations for revenues next year must be delivered. When power is constrained, the quality of compute you put into it must be the most productive computing resource. We are the most rentable, fungible, and durable. The shortage of resources is shifting interest towards us. Our market share was already very high, but surprisingly to everybody, our market share is actually increasing for these reasons.
Jensen Huang: We also offer so much more per gigawatt than we used to. We used to be a GPU company; then we became the largest and fastest-growing networking company. We are now going to be one of the largest and fastest-growing CPU companies. We are in computing for storage, knowledge infrastructure, and security with OpenShell and Sentry. The offering we provide into 1 gigawatt went from about $18 billion per gigawatt, to $20-some billion, to now $40-plus billion per gigawatt, and that is going to keep going up. We are growing share because people shift to our platform, and because we offer more to the world's capex—our dollar share is increasing. Six years ago, our percentage of the world's capex was in the single digits. Right now with the CSPs, we are running about 40%. Outside the CSPs—on-prem and neoclouds, which make up half of our business—our percentage of capex is probably 80%. Between half our business being 40% and half being 80%, with both doubling every year, our percentage of the world's capex is growing.
Elon, Colossus, and SpaceX
Brad Gerstner: One of those buyers is Elon. When he started building Colossus, we were blown away to see him stand up a couple hundred thousand Hoppers in Memphis. Now he is raising another $40 billion, planning to buy maybe 1 million-plus Vera Rubin chips, moving from thousands to millions. He wants 10 gigawatts by the end of 2027, which represents hundreds of billions spent on Vera Rubin. Is it a foregone conclusion that SpaceX will be the largest AI hyperscaler in the world?
Jensen Huang: In our charts, Elon used to be under enterprise; now he is part of our CSP segment. They are massive. I love working with Elon. He is using our technology for his own AI, training models, and deploying models. Grok is super cool, and the demand and consumption for Grok is skyrocketing. Then of course, he rents it. He is fairly unique in combining all three; Google would be another example. He loves the intersection between computing and physical buildouts, and nobody knows more about energy than he does.
Data Centers in Space
Jensen Huang: One particular new market we are going to explore together is putting data centers out in space. Our two engineering teams are working super closely together to create a new type of server, a new type of architecture that might be super good in space and on Earth.
Brad Gerstner: Starship launched last week, had its first orbital rendezvous, and deployed those V3 satellites. Do you have increasing confidence that we will have a material amount of compute in space via Starship by 2028 or 2029?
Jensen Huang: Only Nvidia compute. Let me tell you why: it is not just about the technology. Nvidia's performance per unit energy is the world's best—performance per watt. But most importantly, the return on invested capital of Nvidia's compute is so high that you could put it out in space and the launch cost is a small percentage of it. The launch cost is not inconsequential, so you are not going to put low-cost computing out there; you are going to put high-value computing out there. We are the highest-value computing system in the history of computing, making us the perfect fit to launch into space with excellent ROI.
Brad Gerstner: Do we get 500 megawatts in space by 2029?
Jensen Huang: I will let him tell you, but you could put Vera Rubin in there and it will justify it. The demand is there, the return on investment is so good, and the productivity is excellent. It is fungible: if you put a satellite in space without a customer to rent it, it sits idle, which you don't want. Nvidia compute is fully fungible with lots of renters. We work with every closed-model company, we run every open model, we have thousands of startups and global enterprises. And it is durable. We are the only hardware company in the world where you paid us eight years ago and we are still updating your software with new CUDA algorithms. We will update the software for as long as the satellite is up there. That makes it the perfect payload for SpaceX.
Does AI Revenue Justify the Capex?
Brad Gerstner: In order for finance people to get trillions in capex to pencil out, they need to see payback. If I look at total revenue of all the labs this year, adding up the frontier labs and inference labs, it is about $200 billion. We need to see that double next year to $400 billion, and double the year after to $800 billion to get into the zip code of payback on trillions of dollars of capex. Claude Code led to a 10x increase in Anthropic's revenues. What is your confidence that the labs can continue to 2x to 3x their revenues every year?
Jensen Huang: People misunderstand what is going on. Nvidia represents call it 80% of the revenue share of AI infrastructure in the world today. There are three large segments of Nvidia computing. The first segment is baseline computing: data processing, molecular dynamics, physics simulation, fusion and fission simulation, quantum chemistry, quantum physics, computer graphics. Classical machine learning, XGBoost at Visa and Mastercard for fraud detection, recommender systems—combine all of that classical computing, not deep learning or agentic systems, and it is call it a third of our business, growing 30% to 40% per year. In the absence of any AI labs, Nvidia is quite a large company.
Jensen Huang: The second part is the AI labs. People see two, but we work with every one of them. Meta Muse is doing fantastic, vastly on Nvidia; a year before that was just recommender systems, and now they are killing it with Meta Muse. Grok didn't exist three years ago. We didn't run Gemini before; we run Gemini now. Starting this year, we started working with Anthropic. I see five or six labs doing great, representing not quite one-third of our business, let us say 25% to 30% of our compute.
Jensen Huang: The other third—because we are at the beginning of a computing platform shift—is a vast number of new startups. Noah Instinct is doing great, along with physical AI, biological AI, materials sciences AI, and Jeff Bezos's new company, Prometheus. Thousands of companies across industries rent Nvidia compute. This third of our business is the cloud computing for the next generation. We are in the early innings of this buildout, and it is a big mistake to recognize just a couple of companies and think that they alone fund the infrastructure.
Open vs. Closed Models
Jensen Huang: I have been a great advocate of open models because open models open up new opportunities. If you look at all these startup companies, 80% of them use open models. They are not using them to compete with frontier labs; they are using them to unlock opportunities in physical AI, manufacturing, or healthcare. They take open models like Nemotron and customize them into superintelligence for their domain. The closed labs will provide artificial general intelligence, while vertical companies take proprietary data to achieve superintelligence for specific domains.
Brad Gerstner: If you stand up a gigawatt of compute, you get $40 billion out the other end. With 40 gigawatts coming online next year, that would imply $1.6 trillion in revenue if extrapolated linearly. People get nervous about bubbles popping if supply exceeds demand at a point in time, and whether financing stalls.
Jensen Huang: Break it down to first principles. The $100 billion or $125 billion in ARR people try to extrapolate from the labs is just one part of the buildout. Are people going to continue shopping online? That third of the buildout is not even a debate. A couple of years ago, I said that as we exit this decade, infrastructure buildout will probably be somewhere between $3 trillion to $4 trillion a year—closer to $4 trillion. If $1.5 trillion of that is compute, remember that it was $500 billion before AI started. Over six to eight years, it grows to $1.5 trillion because the world is consuming more digital compute.
Jensen Huang: On top of that, Moore's Law is over. There was a time when you waited one year to get twice as many transistors for the same price. Now, if you want twice as many transistors, you have to pay twice as much. Whatever computing capacity the world wants, you have to apply an X-factor just to account for the end of Moore's Law. Without AI, the semiconductor industry was already on its path to become the largest industry in the world. Plus, new startups are accelerating. The punchline is: it is unnecessary to extrapolate any of the labs into multiple trillions of dollars of revenue to justify the infrastructure.
Too Much Pressure on Two Labs
Brad Gerstner: We saw chatter about whether OpenAI's revenue was $70 billion or $50 billion, and a pullback in infrastructure companies on fears they won't have the money to pay for buildouts. Your point is that across 40 gigawatts next year, one-third is general compute replacement, one-third is neoclouds, sovereigns, enterprises, and startups, and one-third is across multiple labs.
Jensen Huang: We are putting way too much pressure on two labs. They are going to be two of the most consequential technology companies ever built, but they do not have to be anywhere near $1 trillion in revenues for the buildout to make sense. Meanwhile, the industry needs to build out a few trillion dollars of computing infrastructure. By the way, what are these called? Dried plums? A little licorice, salt, sugar, cinnamon—has a little twang to it.
Brad Gerstner: Next year, demand for Nvidia chips is well over 100%. You guided to 70% revenue growth because you are supply constrained. You said you had 18 months to work with the supply chain to do even better. Has the order book continued to grow?
AI Supply and Demand Dynamics
Jensen Huang: We had giant order books: every CSP, NCP, OEM, startup, and large enterprise had already placed orders based on 1- to 2-year forecasts. Because I was allocating a lot less than they wanted, if I didn't publicly state what we had, everybody would be second-guessing if they were being shortchanged. It was my way of handling conversations where I am disappointing people. Pretty much every customer would buy more if we could deliver more. Since then, we have received new orders weekly and daily.
Jensen Huang: We are getting more supply, but downstream constraints like land, power, and shell limit deployment. A couple of weeks ago, we announced 2 gigawatts in Australia—that is like an $80 billion opportunity. In Europe with Yondr and others, we are unlocking more partnerships. Because Nvidia provides the entire full-stack AI infrastructure to neoclouds like CoreWeave or Nebius, we can build in places and with partners that nobody else can. Upstream, critical constraints remain in packaging and substrates. Substrates are getting larger with more chips, requiring more material weight. Thermal coefficients and stacking make warpage and manufacturing defects hard to manage at precision scale. Silicon photonics is also supply-constrained. Every week we increase supply, but the order book keeps going up. It will be hard to close that gap for a few years.
Review of Wall Street Consensus
Brad Gerstner: A year ago, consensus for Nvidia's growth was off by a couple hundred percent. In one year, the consensus forecast shifted up by 100% to 200%, yet the stock is only up 20%. You doubled the revenue and free cash flow forecast, trading at 13 to 14 times earnings. What is the market missing?
Jensen Huang: For a company our size to be gaining share tells you about our resilience and scale advantage. Our competitive moat is increasing.
Vera Rubin Ultra Orders
Jensen Huang: If you want to get in line for Vera Rubin Ultra, it is close to placing-order time, sight unseen. How do you know it is going to be good? Because Nvidia builds it, and Vera Rubin is awesome, and Grace Blackwell was awesome. The workloads are so complex that you cannot just measure one benchmark; $50 billion systems require dozens of technologies working together. Sight unseen, who else can you place a $50 billion order with? We are literally the only one. Back when it was a $100 million purchase order, you could benchmark the chips in front of you. How do you benchmark a $50 billion system? You would need a giant data center and a couple of years to retrofit it. You can't even benchmark these systems anymore.
Jensen Huang: It is exactly like TSMC. We give TSMC 100% of our business because they are the only company where I can plan five years out, sight unseen, knowing the next transistor node will be superior, that they will have capacity, agility, and diverse process technologies from silicon photonics to logic, memory, and advanced packaging. How do you do that with a company one-tenth their size? It is impossible. Nvidia has become an infrastructure company for the planet.
Jensen Huang: We run everything. From early GPT models to FlashAttention, NVLink 72 model parallelization, and modern hybrid SSM-transformer diffusion models, it all runs on CUDA. Whether workloads shift to pre-training, post-training, or agentic inference, we run it all. If you build a $50 billion data center, you know that Nvidia compute is fungible across workloads, models, and customers. If you need to sublease capacity, there are renters everywhere.
Global Reach: Europe, Japan, and Beyond
Jensen Huang: If OpenAI or Anthropic wants to serve Europe, India, Japan, or Korea in a heartbeat, our infrastructure is already there. Placing a $100 billion bet on general-purpose architecture gives you low risk. Specialization is the enemy of fungibility and durability. Proving extraordinary performance yields fungibility and durability, which translates to rentability, productivity, financeability, and the lowest total cost of ownership.
Brad Gerstner: You are buying back historic amounts of stock, potentially upwards of 10% of the company within two years. At Goldman, you called Nvidia the first "growth-value" company, reminiscent of Apple's capital return transition. You are also exploring two new business models: software licensing and revenue sharing.
Jensen Huang: One business model is software licensing. For CSPs who run their own stack, we provide open-source software that they modify. But most enterprises do not have the software departments to maintain complex AI factories; they need us on the SLA to keep it secure and maintained. At a couple thousand dollars per GPU per year across millions of enterprise GPUs, software licensing will be a very large business. Second, for neoclouds where we provide full-stack infrastructure, ecosystem offtake, and investment support, we can participate in revenue share. We get the upfront hardware sale and economic return from multi-year revenue share without operating as a cloud provider ourselves. These two growth vectors leverage our installed base and ecosystem independent of hardware sales.
Brad Gerstner: Warren Buffett historically avoided tech because companies produced free cash flow only to spend it on value-destructive projects. You are spending over 50% to 60% of your free cash flow on share buybacks while growing exponentially.
Jensen Huang: Our venture investments in high-quality ecosystem companies are well-informed because we see the technology flow, but as a percentage of total cash generation, outside investments will decline rapidly. Frontier opportunities like OpenAI and Anthropic won't come around again, and our cash flow is growing fast.
Brad Gerstner: You have previously visibility for 2 to 3 years. Looking out to 2029 or 2030, you see $4 trillion of annual spend by the end of the decade, moving toward $10 trillion in annual infrastructure refreshes over time.
Jensen Huang: Unless there is a global depression, the world must consume an enormous amount of Nvidia compute to transition from retrieval-based computing to generative AI. If annual capacity building and refreshes run around $10 trillion globally in the long run, I would not be surprised.
Brad Gerstner: If $50 trillion of global GDP is tied to knowledge work and we get a 20% productivity gain over five years, that generates $10 trillion of incremental economic activity, justifying spending half of that on capex.
The Decade Ahead: 99.99% Confident
Brad Gerstner: Are you more confident sitting here today than you were 12 months ago that in five years the world will be in a much better place because of these investments?
Jensen Huang: I am 99.99% confident—in honor of a good friend, not using 0 or 100—that too many of us care deeply that the outcome turns out well, and that we will dedicate our might, intellect, and resources to ensure we get there responsibly and optimistically. Many researchers and engineers outside the public eye are doing amazing things independently. Smart people do smart things. It is good that we worry and express concerns, but we shouldn't predict the end of the world. That underestimates the human capacity. When you create conditions where everybody wants a good outcome, we make progress. I am 99.99% certain that we are going to get there together.
Brad Gerstner: We could easily stop, but if we stop today, we lose the opportunity of that better world. Thank you, Jensen.
Jensen Huang: Thank you, Brad.