0:00single worst outcome and that outcome comes from two that outcome could happen if we talk about the technology with fear with uncertainty. the leadership um talks about it uh with so much concern, so much fear and uh all the conversations around safety and security. It it has to be balanced with uh a lot of conversation about prosperity and a lot of conversation about what it can do. And so so I don't think it's about one or the other. We have just make sure that we're balanced
0:29in talking about it. Um, you you probably know that that uh in the history of of all of the technology advancements, one of the one of the stories I was just reading about recently uh has to do with the the airline industry. There was a time when all of them spent, you know, a lot of their advertising dollars and and marketing campaigns on my plane is safer than everybody else's plane. Turns out nobody wanted to hear about that. They wanted the c they wanted the companies to take care of that. What they wanted
0:59to hear was that that the airlines could take them to new places, expand their horizon, make the world smaller so that we could learn about each other's culture, create new economies. That's what they wanted to hear about. They wanted it to be our responsibility to make it safe, not to make it their burden that to think about safety all day all day long. And so I think this is the same thing that's hap it happened in the auto industry. It happens to to the power generation industry. It happened
1:28to the airline industry. And so I think it's a responsibility for all of our all of the leaders all of us to make sure that we take safety and security and building the technology properly our responsibility and uh working with regulators and of course every single there's so many regulations and so many laws already. Intelligence is a digital digital version of what we do and so you would think digital version of some task that humans do. So in fact, humans are
1:58have regulations across so many different domains already, whether it's the FDA, FDA or Nitsa or whatever it is, there's so many different agencies that already have regulations. I'm sure there's a way to formulate AI into that.
2:10And so uh I think the the balance of of precise regulation uh don't regulate hypothetical theoretical harm, regulate actual and pragmatic harm. um and give this technology which is at the at the early stages of the S-curve innovation an opportunity to advance quickly so that it could be safe. And so some of the people uh to if you just go back and think about some of the and remind
2:38yourself of some of the ideas of slowing down technology advancement three years ago when it was just being formed. In fact, the advancement is what made it safer. The advancement made it not hallucinate grounded on real truth. um iterate until the the answer uh is uh uh uh scientifically sound. Um all of those technologies made were made possible through technology advancements. And so we're we're making it safer but advancing it more.
3:07So let let's turn a little bit to your business for a moment. Um your your chips uh GPUs, they're they're different from what other people are hoping to build, right? And so maybe can you talk to the audience about about your products and and why are GPUs different than sort of any other letter before the PU if you will? Well, today we started out we invented a technology called a GPU
3:36and basically a GPU does parallel processing tasks and it solves things like fluid dynamics and particle physics and quantum chemistry and one of the computer graphics, image processing, um image reconstruction.
3:51Uh and of course one of the things that it does incredibly well is artificial intelligence. Um our architecture, our technology is general purpose meaning that uh any we support every single closed models, we support every single open model, we support languages, we support large models, small models, we support um physical models, biological models and so our architecture is fungeible in the sense that it runs
4:19every AI model, runs all data processing and and parallel processing tasks. As a result, we're everywhere. We're in every cloud. We're on prem. We could be at the edge. We're in robotic systems and self-driving cars. Um, and the com the company's been around. We've been working on this for so long. We're in our 14th generation of our architecture.
4:39And, and so, so Nvidia's basically everywhere. One of the advantages of doing that is that these infrastructures are so expensive. One gigawatt is about 50 60 billion dollars. You know, back in the old days, people thought building a fab $25 billion was a giant investment.
4:57And now, literally one gigawatt is twice that. And we're building between now and the end of the decade 100 gigawatts because we want to produce intelligence for the world. And so when you're building investing in infrastructure that's that costly you want to make sure that it is fungeable and it's durable because otherwise if you build something that's too specialized and the world's architecture and algorithm changes and these AI models and technologies are
5:26changing so fast if it changes then all a sudden it could obsolete the investment that you've made. And so the reason why Nvidia's uh growth is accelerating is because people are starting to realize that we're the safest architecture to invest in. It's fungeable. It is durable and it continues to improve over time through software and uh it runs everything and so we run every American model, every international model and every modality of models that you have.
5:54So people are always interested on where the world is going. if if you were going to give us sort of your big picture view where where are we going um what should we be thinking about that we don't always talk about you know what you you see everything really all the companies of the world sort of come and visit with you and sort of tell you where they're going how how would you sort of give us a big picture of a set of ideas that we don't usually hear about
6:23well there's some things that that we hear and and then I'll I'll I'll juxtapose that um against what I expect. And so, for example, in the next couple of years, uh we are going to achieve essentially what people call AGI.
6:39And and um uh and that's not an illogical thing to me. In fact, I would argue that we're practically there today. Okay? And and so the question is what does that mean? Um it either means a lot or it doesn't mean anything. And so, let me give you an example.
6:56If we said that in the next couple of years we're going to achieve artificial general intelligence um and every and there's no question everybody's everybody's tasks the things that we do whether it's in um just learning or working or uh every single industry will be impacted it will benefit everybody.
7:15However the part that is that is not not thought through is in fact uh the thought experiment let me just give you a thought experiment. Suppose and you know that several se several hundred years ago uh humans in order to have a more civil society uh manufactured one of the most important things in the world which is called education intelligence we manu through universities and schools we manufacture intelligence and we manufacture stability at scale without it how would
7:45we have civilization and so now we're creating the digital version of that we're manufacturing now intelligence digitally at scale So that even even countries and people who don't have access to the highest levels of education now have the benefits of the highest levels of education. Okay? And so that's the simplest way of thinking about what is happening right now. So the question is what's the benefit to society as a result of that? Well, let's pretend let's do a thought experiment.
8:12Suppose every single uh every single you know young person that we hire out of school uh is incredibly well educated.
8:20Well, that's pretty much consistent with everybody I hire. They come out of Stanford or Harvard or MIT or you know ETH or and these are incredibly smart kids. However, when these kids come to come to our company and every every company and every industry, we now still have to invest quite a bit in surrounding them with context. So, as it turns out, even when we get to AGI, it's not as if every company's problems are solved.
8:52It's not as if all of a sudden in two years time I say, "Okay, let me connect it to this a a this AI service and I just sit sit back and all of a sudden my company becomes more and more productive overnight." That's just not going to happen at all. You're going to spend all of your energy just like you currently do onboarding a new college grad with a PhD from MIT which is pro surrounding them with context, purpose, relevance, access to, you know, all of these other
9:22things which is essentially if you will the harness around these AIs, the environment around these AIs so that they could do productive work. All of that harnessing, all of that environment creation is what companies do, is what leaders do. And as it turns out, our jobs won't change.
9:43Our jobs is defined by purpose, context, meaning. That's what a job is. In that job, there's a lot of typing and talking and, you know, things like that. Those skills are going to be automated by AI. the tasks are going to be automated, but our our jobs purpose re remain. And so I think what's going to happen is that we're all just going to be supercharged just like we're currently supercharged.
10:10Nvidia is we use AI across the whole company today. We use off-the-shelf AIs like like anthropics and open AIs and cursor and um and we build a lot of our own custom AI and I encourage everybody to apply the same the same idea. You should use AI off the shelf wherever you can. But don't forget, you have to invest in building your own intelligence. Every country, every company, you can't outsource all of your intelligence to somebody else. And so,
10:37you have to use as much of it as you can build as much as you need to. And so, uh, we do that at at the company. And and, uh, and so that's the the first idea is just that we're going to be supercharged. um the idea that somehow all the jobs are going to be eliminated, removed, that's just nonsense. Okay? And the reason for that is because of all the things that I've just said. Now, the other thing that that is the case is that all of our ambitions are going to be supercharged.
11:04Just as you know, we are more ambitious today than any time in human civilization. We think about doing things that are people a hundred years ago would just that's insane. That's got to be science fiction.
11:21And yet we perfectly expla perfectly expected. And so just imagine how ambitious we're going to be in 5 10 years. That's the exciting time. You know, I've been working a long time and surely I don't have to work but I don't want to miss I I'm definitely not going to miss the next 10-15 years. The ambition of the industry, the ambition of my colleagues, my own ambition has really been supercharged. And this is another way of saying that all of all of you, all of the countries, your
11:49ambitions have to be supercharged. This is the time when you have to feel enlightened, inspired, you know, supercharged, lifted because you now have this incredible technology behind you. And so, you know, what do I expect in the next five, ten years? Uh, things that takes 10 years going to take one. things that take one will take a month.
12:10And so, um, you know, if you if you think that that this is going to cost a million dollars to do, uh, well, back a long time ago, just, you know, years ago, a million dollars seems like a big investment, but these days, just think about the level of investments. We're talking about trillions of dollars in investment. And the reason for that is this.
12:31It is now possible for average humans like ourselves to think that we can make an impact on a hundred trillion dollar industry. For the first time, you're hearing people, isn't that right? For the first time, you're hearing innovators, people, companies, countries talking about elevating a hundred trillion dollar industry. That's ambition. To think that you're going to add 20 50 trillion dollars of economic benefit around the world, that's ambition.
13:06No time in history have we been able to say things like that. So my goal for today when we're together is growth. Right? That's our goal. That we can all grow and benefit our countries and benefit society. And I think what you've done, uh, Jensen, which we really appreciate. I mean, you are an American treasure.
13:32Right. Your company, uh, is leading us in a path that none of us when we were young ever dreamed possible. Maybe we dreamed it a little bit. That's what, you know, sort of sci-fi was about, but now we're there. I think what you've laid out for us is a vision of that harnessing AI to using it both from uh surrounding it, contextualizing it and directing it is how we're going to
14:00succeed. That each of the countries here needs to invest in its infrastructure in order to capture that for themselves to make sure that their citizens and their society benefits from it. They've got to make that infrastructure investment. They've got to build the data centers. They've got to harness the AI for themselves.
14:20Because what you've said is that uh education has been democratized. But it's made available for everybody. You now can have a PhD in everything.
14:31At your fingertips and you don't have to find that person because you've helped find them and sort of capture that education in a bot.
14:40You are that person now, right? Everybody is that person now. So I think you know our goal is to uh is to have that idea of growth, that ambition for growth. And what you've done is you've laid us out that our dreams need to be higher because they can be higher because if you have all of this intelligence at your fingertips, how could you not dream higher?
15:04That's right. And I think really for a company and for a leader of a company to have such a uh such a brilliantly positive vision for where we're going, I think was the exact way to lead off today. So I want to thank you for coming everybody. I would like to thank Jensen W for coming and joining us. [applause] Thank you.
15:32[applause] Thank you everybody. It bootstraps you, it elevates you, it enables you, you know, it's a great great equalizer.
15:41So what advice would you give to all of our uh global partners here? How how can they best harness this opportunity going forward for their countries? What do they have to build inside their countries? what investments do they have to make so that they go home and they think about if we want to participate in the in the AI you know revolution what's the best way for them to do it well the first thing is to recognize
16:09what is AI you know I've been using this word infrastructure but when you break it all down AI is if I could simply is a five layer cake at the lowest layer is energy you can't produce something without energy it transforms electricity into mathematics and so first layer is energy the second layer are chips that's the world that I'm in the third layer is infrastructure basically these large data centers land
16:38power shell that incorporates all of the computer technology and energy is plugged into it and money starts spewing out okay so it's kind of the next layer is is essentially infrastructure the layer above that is where AI models live and most people think that that's what AI is but in fact AI is a five layer cake and AI models uh you know I for many of the countries that are here I work across many of your countries already today whether it's in startups
17:07established companies researchers and each one of the regions around the world are specialized in something very something really important in artificial intelligence because it's not just language artificial intelligence can learn and understand and predict almost any information that has structure So biology has structure, chemicals has structure, the laws of physics has structure. And so each one of the regions around the world specializes in different areas in AI and AI cuts across
17:35all of it. Different languages, different technologies, different information theories and so AI cuts across all of that. And there's a whole bunch of AI models. But the most important part of AI is the fifth layer, the layer above that, which is data and applications. the way to benefit you know United States is more deeply invested in uh inventing building uh developing the
18:04entire five layer cake. However, every single region, every single country should decide which one of the layers they want to invest in. You don't have to win in every single one of them. You don't have to develop all of them. The most important part is that every single country should advocate for AI diffusion or the adoption of AI into your industries and the industries that we're talking about span everything from education to healthcare to manufacturing
18:33to sciences you know every single industry in between and so the way to think about AI is just to think about it from the five layer stack and United States is going to lead in the development of AI across Ross these five layer stacks but we'd love to partner with everybody so that everybody could take advantage of this. Nvidia has built it in such a way that every single model and every single field of science and every single modality uh runs and so it
19:00runs American American uh models it runs international models it runs models in biology and chemicals and you know physics and robotics.
19:08So when if I go to a dinner party Yeah. And I say AI. Everybody thinks it's software.
19:16But what we're seeing now is really AI at the edge. It's it's hardware. It's robotics. It's advanced manufacturing.
19:25So how do you see AI and the difference between the physical infrastructure of factories and sort of the general perception that AI is software out there?
19:37Well, AI is intelligence and and starts with software. Um and in fact if you look at AI in the last year we started by talking about large language models. This LLM you know this concept that the world would know what LLM stands for.
19:51It's kind of an amazing thing. And so large language models what made AI useful is putting an exoskeleton around the LLM. That exoskeleton is what is called an agent harness. ess essentially gives the brain the the parts necessary to uh retrieve knowledge, have working memory, uh use tools, be able to collaborate, so on so forth, and so solve problems. And so the
20:21first thing is that the digital version of AI is a is what is called an agent. When you take that digital version, you could put it instead of using a tool like spreadsheet or PowerPoint or Excel or web browsers, you could make it use a tool essentially a machinery. And so this agent would use uh would be embodied inside a physical physical body and all of a sudden it becomes a robot.
20:45You could take this agent and put it inside uh a machine that has four wheels, self-driving cars. You could uh put this agent inside a manipulator and that becomes a pick and place or a manufacturing robotics arm. Uh you could put this uh agent inside something that that essentially is a grocery delivery or or a logistics delivery vehicle or a surgical robot or a autonomous uh drug discovery laboratory. You know, all of
21:14these different versions are basically the same idea. Large language models with an agentic system around it. you could use this this agent agentic system could operate um whether it's digital tools or physical tools. Now in the end what what every single country will have to do is you need to recognize that this is infrastructure just as it's water, roads, electricity, the internet. Every single country needs to build infrastructure so that you could support
21:44your own local economy. It becomes the capacity, the intellectual capacity, the digital intellectual capacity for you to support your researchers and students and society and industries and startups and every single every single country that we've worked with and there's so many here that we've partnered with already when the moment we create an infrastructure locally all of a sudden the startups the researchers just really get activated and so it's really a fantastic investment to make. We've had the benefit in the United States uh to
22:14have built one of the largest infrastructures and and um uh because because of the pro-energy growth and the the speed um and the regulatory environments in the United States, we've been able to build out really quite substantially. And you you look at what's going on in the United States, AI is now generating jobs across every single sector from chip fabs to computer plants to AI factories. Um boy just the number of jobs that we're creating is just hundreds of thousands and as you
22:42know and and uh this year alone we're going to invest close to a trillion dollars into the world's into the United States infrastructure and that's generating enormous economic growth and of course a lot of jobs. So the change makes people nervous. Yeah.
22:59You know and gets all sorts of people to say all sorts of uh of things. you know this AI change how how do you rank it as compared to the other change all the way back to industrial revolution you know where do you put it is it small is it big from your perspective and and what's the best way for uh for our our friends here to embrace it from your perspective well as with all new technology uh at first it see it seems uh uh magical you
23:28could just imagine in the beginning of the production of electricity that somehow uh You could be far far away and all of a sudden you know a light could turn on. It seems like magic and uh in the beginning of all technology innovations it seems like magic but the fact of the matter is all of us who are building it this is engineering there's computers there's software there's mathematics involved um there's probabilities involved and and uh uh
23:56there's a lot there just as in every technology industry when we're building something whether it's a car or a plane it is the responsibility of the people building it to do it safely and responsibly and to continue to advance the technology so that as it advances it becomes safer. I prefer to drive today's car with so much more technology in it than to drive a car from 100 years ago.
24:19I prefer to fly in a plane today. Um it's with all of the technology advances than a plane 100 years ago. And so in a lot of ways we want to accelerate the advancement of the technology so it could be more functionally um uh able to deliver on its promise so that it's safer to to uh to use. And so the same thing is is going on with AI. We have to rem remind the industry um that and I think the dialogue around safety and
24:46security is all fantastic. Uh however, if you think about each one of your countries, what is the worst outcome that could happen in this technology industrial revolution? The worst outcome is that you don't take advantage of it.
25:02That you are left behind. That is the single worst outcome. And that outcome comes from two that outcome could happen if we talk about the technology with fear with uncertainty. The leadership um talks about it uh with so much concern so much fear and uh all the conversations around safety and security. It it has to be balanced with a lot of conversation about prosperity and a lot of conversation about what it can do. And so so I don't think it's
25:31about one or the other. We have to make sure that we're balanced in talking about it. Um, you probably know that that uh in the history of of all of the technology advancements, one of the one of the stories I was just reading about recently uh has to do with the the airline industry. There was a time when all of them spent, you know, a lot of their advertising dollars and and marketing campaigns on my plane is safer than everybody else's plane. Turns out nobody wanted to hear about that. They
26:00wanted the c they wanted the companies to take care of that. What they wanted to hear was that that the airlines could take them to new places, uh, expand their horizon, make the world smaller so that we could learn about each other's culture, create new economies. That's what they wanted to hear about. They wanted it to be our responsibility to make it safe, not to make it their burden that to think about safety all day all day long. And so I think this is the same thing that's hap it happened in
26:29the auto industry. It happens to to the power generation industry. It happened to the airline industry. And so I think it's a responsibility for all of our all of the leaders all of us to make sure that we take safety and security and building the technology properly our responsibility and uh working with regulators and of course every single there's so many regulations and so many laws already. Intelligence is a digital digital version of what we do and so you
26:57would think digital version of some task that humans do. So in fact, humans are have regulations across so many different domains already, whether it's the FDA, FDA or Nitsa or whatever it is, there's so many different agencies that already have regulations. I'm sure there's a way to formulate AI into that.
27:15And so, uh, I think the the balance of of precise regulation, uh, don't regulate hypothetical theoretical harm, regulate actual and pragmatic harm. um and give this technology which is at the at the early stages of the S-curve innovation an opportunity to advance quickly so that it could be safe. And so some of the people uh to if you just go back and think about some of the and remind yourself of some of the ideas of
27:45slowing down technology advancement three years ago when it was just being formed. In fact, the advancement is what made it safer. The advancement made it not hallucinate grounded on real truth. um iterate until that the the answer uh is uh uh uh scientifically sound. Um all of those technologies made were made possible through technology advancements. And so we're we're making it safer but advancing it more.
28:13So let let's turn a little bit to your business for a moment.
28:17um your your chips uh right and so maybe can you talk to the audience about about your products and and why are GPUs different than sort of any other letter before the PU if you will. Well, today we started out, we invented a technology called a GPU and basically a GPU does parallel processing tasks and it solves things like fluid dynamics and particle physics and quantum chemistry and one of the
28:46computer graphics, image processing, um image reconstruction. Uh and of course, one of the things that it does incredibly well is artificial intelligence. Um our architecture, our technology is general purpose meaning that uh any we support every single closed models, we support every single open model, we support languages, we support large models, small models, we support um physical models, biological models and so our architecture is fungeible in the sense that it runs
29:20every AI model, runs all data processing and and parallel processing tasks. As a result, we're everywhere. We're in every cloud. We're on prem. We could be at the edge. We're in robotic systems and self-driving cars. Um, and the com the company's been around. We've been working on this for so long. We're in our 14th generation of our architecture.
29:39And, and so, so Nvidia's basically everywhere. One of the advantages of doing that is that these infrastructures are so expensive. One gigawatt is about 50 60 billion dollars, right? You know, back in the old days, people thought building a fab $25 billion was a giant investment. And now, literally one gigawatt is twice that.
30:01And we're building between now and the end of the decade 100 gigawatts because we want to produce intelligence for the world. And so when you're building investing in infrastructure that's that costly you want to make sure that it is fungeable and it's durable because otherwise if you build something that's too specialized and the world's architecture and algorithm changes and these AI models and technologies are changing so fast if it changes then all of a sudden it could obsolete the
30:30investment that you've made. And so the reason why Nvidia's uh growth is accelerating is because people are starting to realize that we're the safest architecture to invest in. It's fungeable. It is durable and it continues to improve over time through software and uh it runs everything and so we run every American model, every international model and every modality of models that you have.
30:54So people are always interested on where the world is going. If if you were going to give us sort of your big picture view, where where are we going? Um what should we be thinking about that we don't always talk about? You know, what's you you see everything really?
31:12All the companies of the world sort of come and visit with you and sort of tell you where they're going. How how would you sort of give us a big picture of a set of ideas that we don't usually hear about?
31:23Well, there's some things that that we hear and and then I'll I'll I'll juxtapose that. um against what I expect. And so, for example, in the next couple of years, we are going to achieve essentially what people call AGI.
31:39And and um uh and that's not an illogical thing to me. In fact, I would argue that we're practically there today. Okay? And and so the question is what does that mean? Um it either means a lot or it doesn't mean anything. And so, let me give you an example.
31:56If we said that in the next couple of years we're going to achieve artificial general intelligence um and every and there's no question everybody's everybody's tasks the things that we do whether it's in um just learning or working or uh every single industry will be impacted it will benefit everybody.
32:16However the part that is that is not not thought through is in fact uh the thought experiment let me just give you a thought experiment. Suppose and you know that several se several hundred years ago uh humans in order to have a more civil society uh manufactured one of the most important things in the world which is called education intelligence we manu through universities and schools we manufacture intelligence and we manufacture civility at scale this without it how would we
32:45have civilization and so now we're creating the digital version of that we're manufacturing now intelligence digitally at scale So that even even countries and people who don't have access to the highest levels of education now have the benefits of the highest levels of education. Okay? And so that's the simplest way of thinking about what is happening right now. So the question is what's the benefit to society as a result of that? Well, let's pretend let's do a thought experiment.
33:13Suppose every single uh every single you know young person that we hire out of school uh is incredibly well educated.
33:20Well, that's pretty much consistent with everybody I hire. They come out of Stanford or Harvard or MIT or you know ETH or and these are incredibly smart kids. However, when these kids come to come to our company and every every company and every industry, we now still have to invest quite a bit in surrounding them with context. So, as it turns out, even when we get to AGI, it's not as if every company's problems are solved.
33:52It's not as if all of a sudden in two years time I say, "Okay, let me connect it to this a a this AI service and I just sit sit back and all of a sudden my company becomes more and more productive overnight." That's just not going to happen at all. You're going to spend all of your energy just like you currently do onboarding a new college grad with a PhD from MIT which is pro surrounding them with context, purpose, relevance, access to, you know, all of these other
34:22things which is essentially if you will the harness around these AIs, the environment around these AIs so that they could do productive work. All of that harnessing, all of that environment creation is what companies do, is what leaders do. And as it turns out, our jobs won't change.
34:44Our jobs is defined by purpose, context, meaning. That's what a job is. In that job, there's a lot of typing and talking and, you know, things like that. Those skills are going to be automated by AI. the tasks are going to be automated, but our our jobs purpose re remain. And so I think what's going to happen is that we're all just going to be supercharged just like we're currently supercharged.
35:10NVIDIA is we use AI across the whole company today. We use off-the-shelf AIs like like anthropics and open AIs and cursor and um and we build a lot of our own custom AI and I encourage everybody to apply the same the same idea. You should use AI off the shelf wherever you can. But don't forget, you have to invest in building your own intelligence. Every country, every company, you can't outsource all of your intelligence to somebody else. And so,
35:37you have to use as much of it as you can build as much as you need to. And so, uh, we do that at at the company. And and, uh, and so that's the the first idea is just that we're going to be supercharged. um the idea that somehow all the jobs are going to be eliminated or removed that's just nonsense. Okay?
35:54And the reason for that is because of all the things that I've just said. Now the other thing that that is the case is that all of our ambitions are going to be supercharged. Just as you know we are more ambitious today than any time in human civilization. We think about doing things that people a hundred years ago would just that's insane. That's got to be science fiction.
36:21And yet we perfectly explained and so just imagine how ambitious we're going to be in 5 10 years. That's the exciting time. You know, I've been working a long time and surely I don't have to work but I don't want to miss I I'm definitely not going to miss the next 1015 years. The ambition of the industry, the ambition of my colleagues, my own ambition has really been supercharged. And this is another way of saying that all of all of you, all of the countries, your ambitions have to be
36:50supercharged. This is the time when you have to feel enlightened, inspired, you know, supercharged, lifted because you now have this incredible technology behind you. And so, you know, what do I expect in the next five, 10 years? Uh, things that takes 10 years going to take one. things that take one will take a month.
37:10And so, um, you know, if you if you think that that this is going to cost a million dollars to do, uh, well, back a long time ago, just, you know, years ago, a million dollars seems like a big investment, but these days, just think about the level of investments. We're talking about trillions of dollars in investment. And the reason for that is this. It is now possible for average humans like ourselves to think that we
37:37can make an impact on a hundred trillion dollar industry. For the first time, you're hearing people, Isn't that right? For the first time, you're hearing innovators, people, companies, countries talking about elevating a hundred trillion dollar industry. That's ambition. To think that you're going to add 20 50 trillion dollars of economic benefit around the world, that's ambition. No time in history have we been able to say things like that.
38:10So my goal for today when we're together is growth. Right? That's our goal that we can all grow and benefit our countries and benefit society. And I think what you've done, uh, Jensen, which we really appreciate. I mean, you are an American treasure. Thank you.
38:32Right. Your company, uh, is leading us in a path that none of us when we were young ever dreamed possible. Maybe we dreamed it a little bit. That's what, you know, sort of sci-fi was about, but now we're there. I think what you've laid out for us is a vision of that harnessing AI to using it both from uh surrounding it, contextualizing it and directing it is how we're going to
39:01succeed. That each of the countries here needs to invest in its infrastructure in order to capture that for themselves to make sure that their citizens and their society benefits from it. They've got to make that infrastructure investment. They've got to build the data centers. They've got to harness the AI for themselves.
39:20Because what you've said is that uh education has been democratized. But it's made available for everybody. You now can have a PhD in everything.
39:31At your fingertips and you don't have to find that person because you've helped find them and sort of capture that education in a bot. You are that person now, right? Everybody is that person now. So I think you know our goal is to uh is to have that idea of growth, that ambition for growth. And what you've done is you've laid us out that our dreams need to be higher because they can be higher because if you have all of
40:00this intelligence at your fingertips, how could you not dream higher?
40:04That's right. And I think really for a company and for a leader of a company to have such a uh such a brilliantly positive vision for where we're going, I think was the exact way to lead off today. So I want to thank you for coming everybody. I would like to thank Jensen W for coming and joining us. [applause] Thank you.
40:29This is why I really encourage everybody to engage artificial intelligence. It bootstraps you, it elevates you, it enables you. You know, it's great great equalizer.
40:39So, what advice would you give to all of our uh global partners here? How how can they best harness this opportunity going forward for their countries? What do they have to build inside their countries? what investments do they have to make so that they go home and they think about if we want to participate in the in the AI you know revolution what's the best way for them to do it well the first thing is to recognize
41:07what is AI you know I've been using this word infrastructure but when you break it all down AI is if I could simply is a five layer cake at the lowest layer is energy you can't produce something without energy it transforms electricity into mathematics and so first layer is energy the second layer are chips that's the world that I'm in the third layer is infrastructure basically these large data centers land
41:36power shell that incorporates all of the computer technology and energy is plugged into it and money starts spewing out okay so it's kind of the next layer is is essentially infrastructure the layer above that is where AI models live and most people think that That's what AI is. But in fact, AI is a five layer cake. And AI models uh you know I for many of the countries that are here, I work across many of your countries already today. Whether it's in startups,
42:05established companies, researchers, and each one of the regions around the world are specialized in something very something really important in artificial intelligence because it's not just language. artificial intelligence can learn and understand and predict almost any information that has structure. So biology has structure, chemicals has structure, the laws of physics has structure and so each one of the regions around the world specializes in different areas in AI and AI cuts across
42:34all of it. Different languages, different technologies, different information theories. And so AI cuts across all of that. And there's a whole bunch of AI models. But the most important part of AI is the fifth layer the layer above that which is data and applications. The way to benefit you know United States is more deeply invested in uh inventing building uh developing the
43:02entire five layer cake. However, every single region every single country should decide which one of the layers they want to invest in. You don't have to win in every single one of them. You don't have to develop all of them. The most important part is that every single country should advocate for AI diffusion or the adoption of AI into your industries. And the industries that we're talking about span everything from education to healthcare to manufacturing
43:31to sciences, you know, every single industry in between. And so the way to think about AI is just to think about it from the five layer stack. and United States is going to lead in the development of AI across these five layer stacks. But we'd love to partner with everybody so that everybody could take advantage of this. Nvidia has built it in such a way that every single model in every single field of science and every single modality uh runs and so it runs American American uh models, it
44:01runs international models, it runs models in biology and chemicals and you know physics and robotics. So when if I go to a dinner party, Yeah.
44:10and I say AI, everybody thinks it's software.
44:15But what we're seeing now is really AI at the edge. It's it's hardware. It's robotics. It's advanced manufacturing.
44:24So how do you see AI and the difference between the physical infrastructure of factories and sort of the general perception that AI is software out there? Well, AI is intelligence and and starts with software. Um, and in fact, if you look at AI in the last year, we started by talking about large language models. This LLM, you know, this concept that the world would know what LLM stands for. It's kind of an amazing thing. And so, large language models,
44:53what made AI useful is putting an exoskeleton around the LLM. That exoskeleton is what is called an agent harness. E essentially gives the brain the the parts necessary to uh retrieve knowledge, have working memory, uh use tools, be able to collaborate, so on so forth and so solve problems. And so the first thing is that the digital version of AI is a is what is called an agent.
45:25When you take that digital version, you could put it instead of using a tool like spreadsheet or PowerPoint or Excel or web browsers, you could make it use a tool essentially a machinery. And so this agent would use uh would be embodied inside a physical physical body and all of a sudden it becomes a robot.
45:44You could take this agent and put it inside uh a machine that has four wheels, self-driving cars. You could uh put this agent inside a manipulator and it becomes a pick and place or a manufacturing robotics arm. Uh you could put this uh agent inside something that that essentially is a grocery delivery or or logistics delivery vehicle or a surgical robot or a autonomous uh drug discovery laboratory. You know, all of
46:12these different versions are basically the same idea. Large language models with an agentic system around it. you could use this this agent agentic system could operate um whether it's digital tools or physical tools. Now in the end what what every single country will have to do is you need to recognize that this is infrastructure just as it's water, roads, electricity, the internet. Every single country needs to build infrastructure so that you could support
46:42your own local economy. It becomes the capacity, the intellectual capacity, the digital intellectual capacity for you to support your researchers and students and society and industries and startups. And every single every single country that we've worked with, and there's so many here that we've partnered with already, when the moment we create an infrastructure locally, all of a sudden the startups, the researchers just really get activated. And so, it's really a fantastic investment to make.
47:10We've had the benefit in the United States uh to have built one of the largest infrastructures and and um uh because because of the pro-en energy growth and the the speed um and the regulatory environments in the United States, we've been able to build out really quite substantially. And you you look at what's going on in the United States, AI is now generating jobs across every single sector from chip fabs to computer plants to AI factories. Um boy just the number of jobs that we're
47:39creating is just hundreds of thousands and as you know and and uh this year alone we're going to invest close to a trillion dollars into the world's into the United States infrastructure and that's generating enormous economic growth and of course a lot of jobs. So the change makes people nervous. Yeah.
47:57You know and gets all sorts of people to say all sorts of uh of things. you know this AI change how how do you rank it as compared to the other change all the way back to industrial revolution you know where do you put it is it small is it big from your perspective and and what's the best way for uh for our our friends here to embrace it from your perspective well as with all new technology uh at first it see it seems uh uh magical you
48:26could just imagine in the beginning of the production of electricity that somehow You could be far far away and all of a sudden you know a light could turn on. It seems like magic and in the beginning of all technology innovations it seems like magic but the fact of the matter is all of us who are building it this is engineering. There's computers, there's software, there's mathematics involved.
48:51Um there's probabilities involved and and uh uh there's a lot there just as in every technology industry when we're building something whether it's a car or a plane, it is the responsibility of the people building it to do it safely and responsibly and to continue to advance the technology so that as it advances it becomes safer. I prefer to drive today's car with so much more technology in it than to drive a car from a 100 years ago. I prefer to fly in a plane today.
49:20Um it's with all of the technology advances than a plane 100 years ago. And so in a lot of ways we want to accelerate the advancement of the technology so it could be more functionally um uh able to deliver on its promise so that it's safer to to uh to use. And so the same thing is is going on with AI. We have to rem remind the industry um that and I think the dialogue around safety and security is all fantastic. Uh however,
49:49if you think about each one of your countries, what is the worst outcome that could happen in this technology industrial revolution? The worst outcome is that you don't take advantage of it.
50:01That you are left behind. That is the single worst outcome. And that outcome comes from two that outcome could happen if we talk about the technology with fear with uncertainty. The leadership um talks about it uh with so much concern so much fear and uh all the conversations around safety and security it has to be balanced with a lot of conversation about prosperity and a lot of conversation about what it can do and so so I don't think it's about one or
50:30the other. We have to make sure that we're balanced in talking about it. Um you you probably know that that uh in the history of of all of the technology advancements, one of the one of the stories I was just reading about recently uh has to do with the the airline industry. There was a time when all of them spent, you know, a lot of their advertising dollars and and marketing campaigns on my plane is safer than everybody else's plane. Turns out nobody wanted to hear about that. They
50:58wanted the c they wanted the companies to take care of that. What they wanted to hear was that that the airlines could take them to new places, expand their horizon, make the world smaller so that we could learn about each other's culture, create new economies. That's what they wanted to hear about. They wanted it to be our responsibility to make it safe, not to make it their burden that they think about safety all day all day long. And so I think this is the same thing that's hap it happened in
51:28the auto industry. It happens to to the power generation industry. happened to the airline industry. And so I think it's a responsibility for all of our all of the leaders all of us to make sure that we take safety and security and building the technology properly our responsibility and uh working with regulators and of course every single there are so many regulations and so many laws already. Intelligence is a digital digital version of what we do and so you would think digital version
51:56of some task that humans do. So in fact humans are have regulations across so many different domains already whether it's the FDA FDA or Nitsa or whatever it is there's so many different agencies that already have regulations. I'm sure there's a way to formulate AI into that.
52:14And so, uh, I think the the balance of of precise regulation, uh, don't regulate hypothetical theoretical harm, regulate actual and pragmatic harm. Um, and give this technology, which is at the at the early stages of the Scurve innovation, an opportunity to advance quickly so that it could be safe. And so some of the people uh to if you just go back and think about some of the and remind yourself of some of the ideas of
52:43slowing down technology advancement three years ago when it was just being formed. In fact the advancement is what made it safer. The advancement made it not hallucinate grounded on real truth.
52:55Um iterate until the the answer uh is uh uh uh scientifically sound. Um, all of those technologies made were made possible through technology advancements and so we're we're making it safer but advancing it more. So let let's turn a little bit to your business for a moment. Um, your your chips uh GPUs, they're they're different from what other people are hoping to build, right?
53:23And so maybe can you talk to the audience about about your products and and why are GPUs different than so any other letter before the PU if you will? Well, today we started out we invented a technology called a GPU and basically a GPU does parallel processing tasks and it solves things like fluid dynamics and particle physics and quantum chemistry and one of the computer graphics, image processing, um
53:53image reconstruction.
53:55Uh and of course one of the things that it does incredibly well is artificial intelligence. Um our architecture, our technology is general purpose meaning that uh any we support every single closed models, we support every single open model, we support languages, we support large models, small models, we support um physical models, biological models and so our architecture is fungeible in the sense that it runs
54:23every AI model, runs all data processing and and parallel processing tasks. As a result, we're everywhere. We're in every cloud. We're on prem. We could be at the edge. We're in robotic systems and self-driving cars. Um, and the com the company's been around. We working on this for so long. We're in our 14th generation of our architecture. And, and so, so Nvidia's basically everywhere.
54:46One of the advantages of doing that is that these infrastructures are so expensive. One gawatt is about 50 60 billion dollars. You know, back in the old days, people thought building a fab $25 billion was a giant investment. And now, literally one gigawatt is twice that. And we're building between now and the end of the decade 100 gigawatts because we want to produce intelligence for the world. And so when you're building investing in infrastructure
55:14that's that costly, you want to make sure that it is fungeable and it's durable because otherwise if you build something that's too specialized and the world's architecture and algorithm changes and these AI models and technologies are changing so fast, if it changes then all a sudden it could obsolete the investment that you've made. And so the reason why Nvidia's uh growth is accelerating is because people are starting to realize that we're the
55:41safest architecture to invest in. It's fungeible. It is durable and it continues to improve over time through software and uh it runs everything. And so we run every American model, every international model and every modality of models that you have.
55:57So people are always interested on where the world is going. If if you were going to give us sort of your big picture view, where where are we going? Um what should we be thinking about that we don't always talk about? You know, what you you see everything really all the companies of the world sort of come and visit with you and sort of tell you where they're going. How how would you sort of give us a big picture of a set of ideas that we don't usually hear about?
56:26Well, there's some things that that we hear and and then I'll I'll I'll juxtapose that. um against what I expect. And so, for example, in the next couple of years, uh we are going to achieve essentially what people call AGI.
56:42And and um uh and that's not an illogical thing to me. In fact, I would argue that we're practically there today. Okay? And and so the question is what does that mean? Um it either means a lot or it doesn't mean anything. And so, let me give you an example.
57:00If we said that in the next couple of years we're going to achieve artificial general intelligence um and every and there's no question everybody's everybody's tasks the things that we do whether it's in um just learning or working or uh every single industry will be impacted it will benefit everybody.
57:19However, the part that is that is not not thought through is in fact uh the thought experiment. Let me just give you the thought experiment. Suppose and you know that several several hundred years ago uh humans in order to have a more civil society uh manufactured one of the most important things in the world which is called education intelligence. We manu through universities and schools we manufacture intelligence and we manufacture stability at scale. is without it, how would we have
57:49civilization? And so now we're creating the digital version of that. We're manufacturing now intelligence digitally at scale so that even even countries and people who don't have access to the highest levels of education now have the benefits of the highest levels of education. Okay? And so that's the simplest way of thinking about what is happening right now. So the question is what's the benefit to society as a result of that? Well, let's pretend let's do a thought experiment. Suppose
58:16every single uh every single, you know, young person that we hire out of school uh is incredibly well educated. Well, that's pretty much consistent with everybody I hire. They come out of Stanford or Harvard or MIT or, you know, ETH or and these are incredibly smart kids. However, when these kids come to come to our company and every every company and every industry, we now still have to invest quite a bit in
58:44surrounding them with context. So, as it turns out, even when we get to AGI, it's not as if every company's problems are solved.
58:55It's not as if all of a sudden in two years time, I say, "Okay, let me connect it to this a a this AI service." and I just sit sit back and all of a sudden my company becomes more and more productive overnight. That's just not going to happen at all. You're going to spend all of your energy just like you currently do onboarding a new college grad with a PhD from MIT which is pro surrounding them with context, purpose, relevance, access to, you know, all of these other
59:25things which is essentially if you will the harness around these AIs, the environment around these AIs so that they could do productive work. All of that harnessing, all of that environment creation is what companies do, is what leaders do. And as it turns out, our jobs won't change.
59:47Our jobs is defined by purpose, context, meaning. That's what a job is. In that job, there's a lot of typing and talking and, you know, things like that. Those skills are going to be automated by AI. The tasks are going to be automated. But our our jobs purpose re remain. And so I think what's going to happen is that we're all just going to be supercharged just like we're currently supercharged.
1:00:13Nvidia is we use AI across the whole company today. We use off-the-shelf AIs like like anthropics and open AIs and cursor and um and we build a lot of our own custom AI and I encourage everybody to apply the same the same idea. You should use AI off the shelf wherever you can. But don't forget, you have to invest in building your own intelligence. Every country, every company, you can't outsource all of your intelligence to somebody else. And so,
1:00:40you have to use as much of it as you can build as much as you need to. And so, uh, we do that at the company. And and, uh, and so that's the the first idea is just that we're going to be supercharged. um the idea that somehow all the jobs are going to be eliminated, removed, that's just nonsense. Okay? And the reason for that is because of all the things that I've just said. Now, the other thing that that is the case is that all of our ambitions are going to be supercharged.
1:01:07Just as you know, we are more ambitious today than any time in human civilization. We think about doing things that people a hundred years ago would just that's insane. That's got to be science fiction.
1:01:25And yet we perfectly expla perfectly expected. And so just imagine how ambitious we're going to be in five 10 years. That's the exciting time. You know, I've been working a long time and surely I don't have to work but I don't want to miss I I'm definitely not going to miss the next 10-15 years. The ambition of the industry, the ambition of my colleagues, my own ambition has really been supercharged. And this is another way of saying that all of all of you, all of the countries, your
1:01:52ambitions have to be supercharged. This is the time when you have to feel enlightened, inspired, you know, supercharged, lifted because you now have this incredible technology behind you. And so, you know, what do I expect in the next five, ten years? Uh, things that takes 10 years going to take one. things that take one will take a month.
1:02:13And so, um, you know, if you if you think that that this is going to cost a million dollars to do, uh, well, back a long time a just, you know, years ago, a million dollars seems like a big investment, but these days, just think about the level of investments.
1:02:30We're talking about trillions of dollars in investment. And the reason for that is this. It is now possible for average humans like ourselves to think that we can make an impact on a hundred trillion dollar industry.
1:02:46For the first time, you're hearing people, Isn't that right? For the first time, you're hearing innovators, people, companies, countries talking about elevating a hundred trillion dollar industry. That's ambition. To think that you're going to add 2050 trillion dollars of economic benefit around the world, that's ambition. No time in history have we been able to say things like that.
1:03:13So my goal for today when we're together is growth. Right? That's our goal. That we can all grow and benefit our countries and benefit society. And I think what you've done uh Jensen, which we really appreciate, I mean, you are an American treasure. Thank you.
1:03:35Right. Your company, uh, is leading us in a path that none of us when we were young ever dreamed possible. Maybe we dreamed it a little bit. That's what, you know, sort of sci-fi was about, but now we're there. I think what you've laid out for us is a vision of that harnessing AI to using it both from uh surrounding it, contextualizing it and directing it is how we're going to
1:04:04succeed. That each of the countries here needs to invest in its infrastructure in order to capture that for themselves to make sure that their citizens and their society benefits from it. They've got to make that infrastructure investment. They've got to build the data centers. They've got to harness the AI for themselves.
1:04:23Because what you've said is that uh education has been democratized. But it's made available for everybody. You now can have a PhD in everything.
1:04:35At your fingertips and you don't have to find that person because you've helped find them and sort of capture that education in a bot. You are that person now, right? Everybody is that person now. So I think you know our goal is to uh is to have that idea of growth, that ambition for growth. And what you've done is you've laid us out that our dreams need to be higher because they can be higher because if you have all of
1:05:04this intelligence at your fingertips, how could you not dream higher?
1:05:08That's right. And I think really for a company and for a leader of a company to have such a uh such a brilliantly positive vision for where we're going, I think was the exact way to lead off today. So I want to thank you for coming everybody. I would like to thank Jensen W for coming and joining us. [applause] Thank you.
1:05:33That's why I really encourage everybody to engage artificial intelligence. It bootstraps you, it elevates you, it enables you, you know, it's a great great equalizer.
1:05:43So, what advice would you give to all of our uh global partners here? How how can they best harness this opportunity going forward for their countries? What do they have to build inside their countries? what investments do they have to make so that they go home and they think about if we want to participate in the in the AI you know revolution what's the best way for them to do it well the first thing is to recognize
1:06:11what is AI you know I've been using this word infrastructure but when you break it all down AI is if I could simply is a five layer cake at the lowest layer is energy you can't produce something without energy it transforms electricity into mathematics and so first layer is energy the second layer are chips that's the world that I'm in the third layer is infrastructure basically these large data centers land
1:06:40power shell that incorporates all of the computer technology and energy is plugged into it and money starts spewing out okay so it's kind of the next layer is is essentially infrastructure the layer above that is where AI models live and most people think that That's what AI is. But in fact, AI is a five layer cake. And AI models uh you know I for many of the countries that are here, I work across many of your countries already today. Whether it's in startups,
1:07:08established companies, researchers, and each one of the regions around the world are specialized in something very something really important in artificial intelligence because it's not just language. artificial intelligence can learn and understand and predict almost any information that has structure. So biology has structure, chemicals has structure, the laws of physics has structure and so each one of the regions around the world specializes in different areas in AI and AI cuts across
1:07:37all of it. Different languages, different technologies, different information theories. And so AI cuts across all of that. And there's a whole bunch of AI models. But the most important part of AI is the fifth layer the layer above that which is data and applications. The way to benefit you know United States is more deeply invested in uh inventing building uh developing the
1:08:05entire five layer cake. However, every single region every single country should decide which one of the layers they want to invest in. You don't have to win in every single one of them. You don't have to develop all of them. The most important part is that every single country should advocate for AI diffusion or the adoption of AI into your industries. And the industries that we're talking about span everything from education to healthcare to manufacturing
1:08:35to sciences, you know, every single industry in between. And so the way to think about AI is just to think about it from the five layer stack. and United States is going to lead in the development of AI across these five layer stacks. But we'd love to partner with everybody so that everybody could take advantage of this. Nvidia has built it in such a way that every single model in every single field of science and every single modality uh runs and so it runs American American uh models, it
1:09:04runs international models, it runs models in biology and chemicals and you know physics and robotics. So when if I go to a dinner party, Yeah.
1:09:14and I say AI, everybody thinks it's software.
1:09:18But what we're seeing now is really AI at the edge. It's it's hardware. It's robotics. It's advanced manufacturing.
1:09:27So how do you see AI and the difference between the physical infrastructure of factories and sort of the general perception that AI is software out there? Well, AI is intelligence and and it starts with software. Um, and in fact, if you look at AI in the last year, we started by talking about large language models. This LLM, you know, this concept that the world would know what LLM stands for. It's kind of an amazing thing. And so, large language
1:09:56models, what made AI useful is putting an exoskeleton around the LLM. That exoskeleton is what is called an agent harness. It's essentially gives the brain the the parts necessary to uh retrieve knowledge, have working memory, uh use tools, be able to collaborate, so on so forth and so solve problems. And so the first thing is that the digital version
1:10:25of AI is a is what is called an agent. When you take that digital version, you could put it instead of using a tool like spreadsheet or PowerPoint or Excel or web browsers, you could make it use a tool essentially a machinery. And so this agent would use uh would be embodied inside a physical physical body and all of a sudden it becomes a robot.
1:10:47You could take this agent and put it inside uh a machine that has four wheels, self-driving cars. You could uh put this agent inside a manipulator and that becomes a pick and place or a manufacturing robotics arm. Uh you could put this uh agent inside something that that essentially is a grocery delivery or or a logistics delivery vehicle or a surgical robot or a autonomous uh drug discovery laboratory. You know, all of
1:11:16these different versions are basically the same idea. Large language models with an agentic system around it. you could use this this agent agentic system could operate um whether it's digital tools or physical tools. Now in the end what what every single country will have to do is you need to recognize that this is infrastructure just as it's water, roads, electricity, the internet. Every single country needs to build infrastructure so that you could support
1:11:46your own local economy. It becomes the capacity, the intellectual capacity, the digital intellectual capacity for you to support your researchers and students and society and industries and startups.
1:11:58And every single every single country that we've worked with and there's so many here that we've partnered with already when the moment we create an infrastructure locally, all of a sudden the startups, the researchers just really get activated. And so it's really a fantastic investment to make. We've had the benefit in the United States uh to have built one of the largest infrastructures and and um uh because because of the pro-energy growth and the the speed um and the regulatory environments in the United States, we've
1:12:27been able to build out really quite substantially. And you you look at what's going on in the United States, AI is now generating jobs across every single sector from chip fabs to computer plants to AI factories. Um boy just the number of jobs that we're creating is just hundreds of thousands and as you know and and uh this year alone we're going to invest close to a trillion dollars into the world's into the United States infrastructure and that's generating enormous economic growth and
1:12:54of course a lot of jobs. So the change makes people nervous. Yeah.
1:13:01You know it gets all sorts of people to say all sorts of uh of things. you know this AI change how how do you rank it as compared to the other change all the way back to industrial revolution you know where do you put it is it small is it big from your perspective and and what's the best way for uh for our our friends here to embrace it from your perspective well as with all new technology uh at first it see it seems uh uh magical you
1:13:30could just imagine in the beginning of the production of electricity that somehow You could be far far away and all of a sudden you know a light could turn on.
1:13:40It seems like magic and uh in the beginning of all technology innovations it seems like magic but the fact of the matter is all of us who are building it this is engineering there's computers there's software there's mathematics involved um there's probabilities involved and and uh uh there's a lot there just as in every technology industry when we're building something whether it's a car or a plane it is the responsibility of the people building it to do it safely and responsibly and to
1:14:10continue to advance the technology so that as it advances it becomes safer. I prefer to drive today's car with so much more technology in it than to drive a car from 100 years ago. I prefer to fly in a plane today. Um it's with all of the technology advances than a plane 100 years ago. And so in a lot of ways we want to accelerate the advancement of the technology so it could be more functionally um uh able to deliver on its promise so that it's safer to to uh
1:14:38to use. And so the same thing is is going on with AI. We have to rem remind the industry um that and I think the dialogue around safety and security is all fantastic. Uh however, if you think about each one of your countries, what is the worst outcome that could happen in this technology industrial revolution? The worst outcome is that you don't take advantage of it.
1:15:04That you are left behind. That is the single worst outcome. And that outcome comes from two that outcome could happen if we talk about the technology with fear with uncertainty. The leadership um talks about it uh with so much concern so much fear and uh all the conversations around safety and security. It it has to be balanced with a lot of conversation about prosperity and a lot of conversation about what it can do. And so so I don't think it's
1:15:33about one or the other. We have to make sure that we're balanced in talking about it. Um, you probably know that that uh in the history of of all of the technology advancements, one of the one of the stories I was just reading about recently uh has to do with the the airline industry. There was a time when all of them spent, you know, a lot of their advertising dollars and and marketing campaigns on my plane is safer than everybody else's plane. Turns out nobody wanted to hear about that. They
1:16:02wanted the c they wanted the companies to take care of that. What they wanted to hear was that that the airlines could take them to new places, expand their horizon, make the world smaller so that we could learn about each other's culture, create new economies. That's what they wanted to hear about. They wanted it to be our responsibility to make it safe, not to make it their burden that to think about safety all day all day long. And so I think this is the same thing that's hap it happened in
1:16:31the auto industry. It happens to to the power generation industry. it happened to the airline industry and so I think it's a responsibility for all of our all of the leaders all of us to make sure that we take safety and security and building the technology properly our responsibility and uh working with regulators and of course every single there are so many regulations and so many laws already intelligence is a digital digital version of what we do and so you would think digital version
1:17:00of some task that humans do so in In fact, humans are have regulations across so many different domains already, whether it's the FDA, FDA or NIST or whatever it is. There's so many different agencies that already have regulations. I'm sure there's a way to formulate AI into that. And so, uh, I think the the balance of of precise regulation, uh, don't regulate hypothetical theoretical harm, regulate
1:17:28actual and pragmatic harm. um and give this technology which is at the at the early stages of the S-curve innovation an opportunity to advance quickly so that it could be safe. And so some of the people uh to if you just go back and think about some of the and remind yourself of some of the ideas of slowing down technology advancement three years ago when it was just being formed. In fact, the advancement is what made it safer. The advancement made it not
1:17:56hallucinate grounded on real truth. um iterate until the the answer uh is uh uh uh scientifically sound. Um all of those technologies made were made possible through technology advancements. And so we're we're making it safer but advancing it more.
1:18:14So let let's turn a little bit to your business for a moment. Um your your chips uh GPUs, they're they're different from what other people are hoping to build, right? And so maybe can you talk to the audience about about your products and and why are GPUs different than sort of any other letter before the PU if you will? Well, today we started out, we invented a technology called a GPU.
1:18:43And basically a GPU does parallel processing tasks and it solves things like fluid dynamics and particle physics and quantum chemistry and one of the computer graphics, image processing, um, image reconstruction.
1:18:58Uh, and of course, one of the things that it does incredibly well is artificial intelligence. Um our architecture, our technology is general purpose meaning that uh any we support every single closed models, we support every single open model, we support languages, we support large models, small models, we support um physical models, biological models and so our architecture is funible in the sense that it runs
1:19:26every AI model, runs all data processing and and parallel processing tasks. As a result, we're everywhere. We're in every cloud. We're on prem. We could be at the edge. We're in robotic systems and self-driving cars. Um, and the com the company's been around. We've been working on this for so long. We're in our 14th generation of our architecture.
1:19:46And, and so, so Nvidia's basically everywhere. One of the advantages of doing that is that these infrastructures are so expensive. One gigawatt is about 50 60 billion dollars. You know, back in the old days, people thought building a fab $25 billion was a giant investment.
1:20:04And now, literally one gigawatt is twice that. And we're building between now and the end of the decade 100 gigawatts because we want to produce intelligence for the world. And so when you're building investing in infrastructure that's that costly, you want to make sure that it is fungeable and it's durable because otherwise if you build something that's too specialized and the world's architecture and algorithm changes and these AI models and
1:20:32technologies are changing so fast if it changes then all of a sudden it could obsolete the investment that you've made. And so the reason why Nvidia's uh growth is accelerating is because people are starting to realize that we're the safest architecture to invest in. It's fungeable. It is durable and it continues to improve over time through software and uh it runs everything. And so we run every American model, every international model and every modality of models that you have.
1:21:01So people are always interested on where the world is going. if if you were going to give us sort of your big picture view.
1:21:11I'm really uh excited for our conversation. So, I'd like to introduce you all to Tom Brown, one of the founders of Anthropic. So, Tom, thank thanks for coming and joining us. Would you mind giving um my my uh my partner ministers uh sort of a background of yourself and Anthropic because I don't think they know you so well. and Tropic of course is famous, but a little a little story would always be helpful.
1:21:38Yeah, I'd love to. Um, and yeah, just first off, thank you so much for uh for inviting me to this. It's an honor to be here and to get to speak with you all.
1:21:46Um, maybe I so I'll talk through a little bit about myself and my background. Um, maybe even before that, I think like probably three things that I'd love all of you guys to take away from uh this conversation. one is that uh AI capabilities have been increasing very rapidly obviously um I expect them to continue to be increasing really rapidly and consistently um two uh as a result of
1:22:15that we are seeing this demand that's not just fast but also like exponential so like increasing in multiples every year um and then I think three as a result of that uh we have this incredible industrial buildout the likes of which we've never seen in the history of humanity, even more than the the railroads of the 1800s. And so that's a place where um any country that wants to be involved has a chance to to be
1:22:43involved. Um so I guess my my background uh is um I was an engineer originally.
1:22:50Uh I'm now the um chief comput officer at Anthropic. I kind of got into this back in um 2014 where uh a few of my friends came to me and they were like hey I I had done startups for many years they came to me and they were like Tom uh it seems like we might end up getting uh human level intelligence from AI within our
1:23:16lifetimes. So in the next 50 years. Um, and I was like, "Wow, that like seems kind of crazy." But if it does happen, like that would be the biggest most important thing that ever happens to humanity. Like the things that makes us special is our minds. It's not our body.
1:23:33The industrial revolution transformed all of society. But really, it was like like the second most important thing about humans is our bodies. The main thing is our is our minds. Um, and so that that got me to think like, okay, maybe I'll switch from engineering to try to help out learning about AI. Um, and so, uh, I did I started doing kind of self-study on that. And then about a year after that, um, uh, OpenAI was
1:24:03founded. Um, and so I knew some of the folks from that. I'd long been a fan of, uh, of Elon Musk. And so I went to see if I could help out in some way. And so I joined I joined them as like one of the first 20 employees basically to do engineering work. Um and then uh uh kind of learned more about that. And then I think the the key thing for me where I was like, "Oh, wow. this actually is going to be a big deal that's not just
1:24:31in 50 years but faster was um the scaling laws paper um which was in 2019 and that was basically looking at um within AI people had seen from like 20 uh for for a decade before 2019 that if you like the models were getting bigger and bigger people were using more and more compute and the quality of the models was improving over time um but they didn't have a good understanding of
1:25:00how like what the what the mechanism was for taking the bigger models and making them smarter. And so this was a thing where some of my collaborators um uh now anthropic co-founders um went and tried to do the same kind of scientific research that you would do if you were like in the early days of a combustion engine and you were like okay like we see combustion engines bigger engines provide more power but uh how much power
1:25:29do you get from an engine and if you like scale up the size of the engine or change the ratio of how much fuel versus oxygen you put in the engine, how much power do you get? And so those studies that they did then gave a recipe of what exact combinations of compute data and model size resulted in a given level of intelligence. And so um those types of
1:25:59algorithms and those types of recipes have been improving over time. And then also just the amount of compute that we've been putting into the models has been scaling up. And so that's been kind of the the underlying thing that got me to be like, okay, wow, we we we can look and see try it out over eight orders of magnitude like many many orders of magnitude um and see this consistent trend. And now we've seen it over even more like um over yeah over over 15
1:26:26orders of magnitude. And so uh um that gave us kind of the confidence that this stuff would continue to advance um at the rapid rate that it that it has been which ended up leading to the to the founding of Anthropic because we saw that this was going to be something that was happening not in 50 years [clears throat] but in on the order of uh of under 10 years. Um so that that kind of was my origin story of getting
1:26:55into this entire thing. Now, Anthropic is focused on making sure that as we go through this very rapid um kind of a like AI revolution, things go well for everyone. And so, I've been really grateful to get to work with uh with you, Secretary, and with um with the rest of government to make sure that this stuff is benefiting everybody. And now is a chance also, I think, for all of you guys to to be included in this uh
1:27:24incredible once in a once in a generation perhaps probably once in the entire history of humanity um uh technology change.
1:27:35So anthropic has managed to stay at the frontier, right? And that uh you know we we toss around the word frontier, but there's no real frontier. It's just that which you're creating that's new tomorrow. How how have you managed uh to stay, you know, focused and and how are you selecting where to go because you're doing things different than the other companies are doing?
1:28:04So, how what's the thinking behind it?
1:28:06And so, that'll give us a little sense of where you're going next or not going next.
1:28:12Yeah. Um, that's a great question. And so I think from the beginning we've wanted to make um we've wanted to make the AIS great at scientific innovation and entrepreneurship. Um and so from the very get-go uh initially the the the first thing that took off with with the models um was uh was chat GPT um basically kind of like Google a better Google. Um and so you could ask them
1:28:42questions and you could they would go and answer a question much better than what you could do before. Um I think from the beginning we were thinking about like okay the main underlying source of all of the value like this this amazing room all of the prosperity that we have is uh the hard work of of our um uh of of people doing like intellectual labor. And so we wanted to make sure that the models could help
1:29:09with that. Um and so that ended up resulting in our focus on coding where that was a place where um if you imagine if I imagined just like a very smart uh friend who was uh like wanted to do work but they were trapped in a computer like what are the things that they could do to help out and like add a huge amount of of uh of their of value to the world there. And so a programmer is like a very clear way that you could do that.
1:29:38you don't need to be running like physical experiments like a a biologist would or a physicist. Um you can just go ahead and do it in the computer. Um and so that that was kind of why we initially focused on that. It also has the benefit now where our models are able to do um to help out with the research that we do where um the vast majority of all of the work that [clears throat]
1:30:06uh anthropic does itself is software engineering um internal software engineering solving bugs in the model training. Um and so that that focus then was both useful for um like promoting things externally and also for helping us out internally.
1:30:25So my ministerial partners my my objective for today is uh growth global growth uh that they can go home and and have ideas that will help their societies their citizens grow and be more successful. So in your experience trying to figure out how to have AI around the world, what what's the infrastructure you think they need?
1:30:52What are the moves um that our you know G20 ministerial partners need to make so that they can participate with you in the in the coming benefits from AI?
1:31:05Oh, that's that's a great question. So, um I think I I I think that the biggest thing by far is uh building more making it possible to build more um data centers, more compute in your countries. It's I think it's it's it's very clearly the bottleneck for all of our progress.
1:31:25We're in a huge shortage of uh the power and labor needed to support um the demand for AI. Um, I think I think I I I really loved President Trump's uh post from from earlier this week where I think it's um it's very it he was pointing out that um uh the data centers are just an enormous source of prosperity. They produce a ton of jobs.
1:31:48They um uh they reduce taxes. Now, the way that we design them, we actually bring on more power to the grid. So um the like rateayer um pledge uh is something that we've signed up to as have uh the major other major labs and that makes it so that we bring on power behind the grid um behind behind the meter and we can give it back to the grid as well. So I think that's a place
1:32:15where um uh uh like every country can quickly can quickly be be helping out and having a big part to play here. So, you know, what I what I try to have people think about is the red carpet, right? That if your country rolls out the red carpet to great investment, they're more likely to have that investment than if it's a fight to get in.
1:32:43Yes, sounds right. So, um, you know, so in in our conversation last night, uh, Tom was suggesting that, you know, I'd really like to build in countries around the world. Can you help me? So, I said, well, actually, you're going to have the opportunity to talk directly yourself. So, what what do you you need data centers and infrastructure? Is that basically um and you should tell them what you need.
1:33:10Yeah. Yeah. No, that that that's definitely right. So, um uh I think the the the biggest things that we need in order to to build data centers are um we need permitting um for the land to be able to to build there and then we need to figure out how to get uh enough labor to do the construction um there as well.
1:33:32Um and uh then the obvious benefits to the uh to the country too is like um like now there it's a huge investment in the country. It produces jobs. Um it can be like it can be taxed which provides more revenue to the country. Um and so yeah like I I think that uh there there's many things that can block [laughter] data centers from being built so that we we could get a very long list of it. Um but once they are built they they provide huge amounts of value.
1:34:01All right. So uh our our partners build data centers. They um they embrace AI.
1:34:12What are the what's the deliveries? You know what's the outcome? How how can AI make their countries better and make the world a better place? Can you give us sort of your vision from anthropics perspective?
1:34:26what what do you bring that's different and what can you um help them achieve?
1:34:33Yeah. Um so I think there's probably two there's probably two main two main areas. One is one is just the um like re-industrialization that's caused by building huge amounts of of valuable physical infrastructure. Um and so yeah we we talked about that. I think that's that's pretty clear. Um and then a separate thing there is that uh the reason that people want to build [laughter] the like build more compute is because the models end up then adding
1:35:01value. Um and so uh the places that that's happened so far has been um mostly by uh like um scaling up the effective labor force or intellectual labor. Um which is yeah like the the key the key component of the overall economy is how many smart people do we have doing good work. um the AIs now are like not quite people, but they're uh doing intellectual labor also. And so they
1:35:31they feed into that same um like economic calculus that that improves that like raises all boats. Um and then I think like very concretely uh uh some of the most important problems that have plagued humanity like um uh like disease, cancer, etc. are all blocked by not having enough smart, dedicated people who can be constantly working to solve those problems. Um, like I think I
1:36:00think that there's no doubt that if we had uh like a million times more uh oncology researchers, we could make progress that's not just like uh like 10% faster, it would be like multiples faster. Um and so that that I think is the is the the the thing that will come from this very large scale up of more and more smart models. Um and and we see
1:36:28that now with um with math. So uh historically the models were doing more wrote tasks like you would ask them to do um like uh simple things that like maybe like a parallegal work or something like that or like a basic software engineering task. Now they're doing much more complex software engineering tasks. Also with math they traditionally could do okay with with math and then they got to the point where they would be like um like an undergraduate level mathematician
1:36:55graduate level. Now over the last three months they've gotten to the point where they're like a a once in a generation genius on math like any of the frontier models now are I I think they're like a top 100 current mathematician in many of the sub fields of mathematics. um which is just astonishingly fast. Um and so many aspects of science are kind of a combination of engineering plus like
1:37:23mathematics. And so my expectation is that over the next 12 months we'll see kind of that same thing where like now they're like okay scientists or biologists. But um I my prediction is that yeah within the next 12 months we'll get the same thing where like they could be a a once in a generation um scientists for key scientific fields. So it's your view then that using
1:37:51anthropics model uh you can really be aspirational right meaning that what what are the what are the problems facing your economy your people and use the models to try to really deeply examine what could be solutions to those models and use the the sort of bottled logic of anthropic to uh to achieve that. So how how would one think about
1:38:20that from from your perspective because you you know it's capable in anthropic and you know where it's going. So maybe give us some sense on how to solve those kind of problems.
1:38:31Yeah. Yeah. And I I think that's exactly right. I think that um uh that um often people are not ambitious enough in what they ask the models to do. And so I think that uh um like because the way that the model started out was like you would ask them a question uh that like like you would ask Google and it wouldn't have that much context. It would give you like a a a page long response or a paragraph response. People often will take one of the new models
1:38:59from the last like 3 months and do that same thing and they're like ah it's like slightly better. It's like cool I guess.
1:39:06Um, but the way that uh I see the the best people who get the most value out of the models using them now is instead they treat them like they would treat a um a collaborator uh like a a co-orker in their organization. And so they do the same thing that you would need to do if you were like, I've hired the smartest person I know. I want them to be as productive as possible within my organization. And so you go and you make sure that they have access to all of the
1:39:36documents that they need. Um you make sure that they have access to uh the tools. If they're a software engineer, you're like, I'm going to make sure that you have um all of the documentation for the code that you have access to the actual servers so that you can go and run things. If they're a scientist, you make sure that they have access to like the machines that can actually move stuff around. with a model like that would now need a an API to move the machines, but that's the thing that you can get for them. Now, um you you also
1:40:05need to make sure so like in in government, you would make need to make sure that they have the right permissions to access the right information that they would need to go and do the work. Um and so I think once you do that, then you end up with a model that has all the things that a great co-orker would need in order to to do work. And we see that that's now actually how the majority of the um like intellectual labor within anthropic is
1:40:32done is by these models that have the same permissions that a human co-orker
1:41:31intelligence, every country, every company. You can't outsource all of your intelligence to somebody else. And so you have to use as much of it as you can build as much as you need to. And so, uh, we do that at at the company and and, uh, and so that's the the first idea is just that we're going to be supercharged. Um, the idea that somehow all the jobs are going to be eliminated, removed, that's just nonsense. Okay? And the reason for that is because of all the things that I've just said. Now, the other thing that that is the case is
1:42:01that all of our ambitions are going to be supercharged. Just as you know we are more ambitious today than any time in human civilization. We think about doing things that are people 100 years ago we just that's insane. That's got to be science fiction.
1:42:23And yet we perfectly expla perfectly expected. And so just imagine how ambitious we're going to be in five ten years. That's the exciting time. You know, I've been working a long time and surely I don't have to work, but I don't want to miss I I'm definitely not going to miss the next 10-15 years. The ambition of the industry, the ambition of my colleagues, my own ambition has really been supercharged. And this is another way of saying that all of all of you, all of the countries, your
1:42:50ambitions have to be supercharged. This is the time when you have to feel enlightened, inspired, you know, supercharged, lifted because you now have this incredible technology behind you. And so you know what do I expect in the next five t years? Uh things that takes 10 years going to take one. Things that take one will take a month.
1:43:11And so um you know if you if you think that that this is going to cost a million dollars to do uh well back a long time a just you know years ago a million dollars seems like a big investment but these days just think about the level investments we're talking about trillions of dollars in investment. And the reason for that is this. It is now possible for average humans like ourselves to think that we
1:43:38can make an impact on a hundred trillion dollar industry. For the first time, you're hearing people, isn't that right? For the first time, you're hearing innovators, people, companies, countries talking about elevating a hundred trillion dollar industry. That's ambition. To think that you're going to add 2050 trillion dollars of economic benefit around the world, that's ambition. No time in history have we been able to say things like that.
1:44:11So my goal for today when we're together is growth. Right? That's our goal. That we can all grow and benefit our countries and benefit society. And I think what you've done uh Jensen, which we really appreciate, I mean, you are an American treasure. Thank you.
1:44:33Right. Your company, uh, is leading us in a path that none of us when we were young ever dreamed possible. Maybe we dreamed it a little bit. That's what, you know, sort of sci-fi was about, but now we're there. I think what you've laid out for us is a vision of that harnessing AI to using it both from uh surrounding it, contextualizing it and directing it is how we're going to
1:45:02succeed. That each of the countries here needs to invest in its infrastructure in order to capture that for themselves to make sure that their citizens and their society benefits from it. They've got to make that infrastructure investment. They've got to build the data centers. They've got to harness the AI for themselves.
1:45:21Because what you've said is that uh education has been democratized. But it's made available for everybody. You now can have a PhD in everything.
1:45:33At your fingertips and you don't have to find that person because you've helped find them and sort of capture that education in a bottle. You are that person now, right? Everybody is that person now. So I think you know our goal is to uh is to have that idea of growth, that ambition for growth. And what you've done is you've laid us out that our dreams need to be higher because they can be higher because if you have
1:46:01all of this intelligence at your fingertips, how could you not dream higher?
1:46:06That's right. And I think really for a company and for a leader of a company to have such a uh such a brilliantly positive vision for where we're going, I think was the exact way to lead off today. So I want to thank you for coming everybody. I would like to thank Jensen Wong for coming and joining us [applause]
1:46:31for our conversation. So, I'd like to introduce you all to Tom Brown, one of the founders of Anthropic. So, Tom, thank thanks for coming and joining us. Would you mind giving um my my uh my partner ministers uh sort of a background of yourself and Anthropic because I don't think they know you so well. Anthropic of course is famous, but a little a little story would always be helpful.
1:46:55Yeah, I'd love to. Um and yeah, just first off, thank you so much for uh for inviting me to this. It's an honor to be here and to get to speak with you all.
1:47:03Um maybe so I'll talk through a little bit about myself and my background. Um maybe even before that I think like probably three things that I'd love all of you guys to take away from uh this conversation. One is that uh AI capabilities have been increasing very rapidly.
1:47:24Obviously um I expect them to continue to be increasing really rapidly and consistently. Um two uh as a result of that we are seeing this demand that's not just fast but also like exponential.
1:47:38So like increasing in multiples every year. Um and then I think three as a result of that uh we have this incredible industrial buildout the likes of which we've never seen in the history of humanity even more than the the railroads of the 1800s. And so that's a place where um any country that wants to be involved has a chance to to be involved. Um, so I guess my my background uh is um I was an engineer
1:48:07originally. Uh I'm now the um chief compute officer at Anthropic. I kind of got into this back in um 2014 where uh a few of my friends came to me and they were like, "Hey, I I've had done startups for many years. They came to me and they were like, "Tom, uh it seems like we might end up getting uh human level intelligence from AI within
1:48:33our lifetimes." So, in the next 50 years. Um and I was like, "Wow, that like seems kind of crazy, but if it does happen, like that would be the biggest most important thing that ever happens to humanity." Like the things that makes us special is our minds. It's not our body. The industrial revolution transformed all of society, but really it was like like the second most important thing about humans is our bodies. The main thing is our is our
1:49:02minds. Um and so that that got me to think like okay maybe I'll switch from engineering to try to help out learning about AI. Um, and so, uh, I did I started doing kind of self-study on that. And then about a year after that, um, uh, OpenAI was founded. Um, and so I knew some of the folks from that. I'd long been a fan of, uh, of Elon Musk.
1:49:27And so I went to see if I could help out in some way. And so I joined I joined them as like one of the first 20 employees basically to do engineering work. Um, and then, uh, uh, kind of learned more about that. And then I think the the key thing for me where I was like, "Oh, wow. This actually is going to be a big deal that's not just in 50 years, but faster was um, the scaling laws paper um, which was in
1:49:562019. And that was basically looking at um within AI people had seen from like 20 uh for for a decade before 2019 that if you like the models were getting bigger and bigger. People were using more and more compute and the quality of the models was improving over time. Um but they didn't have a good understanding of how like what the what the mechanism was for taking the bigger models and making them smarter. And so
1:50:25this was a thing where some of my collaborators um uh now anthropic co-founders um went and tried to do the same kind of scientific research that you would do if you were like in the early days of a combustion engine and you were like okay like we see combustion engines bigger engines provide more power but uh how much power do you get from an engine and if you like scale up the size of the engine or
1:50:53change the ratio of how much fuel versus oxygen you put in the engine, how much power do you get? And so those studies that they did then gave a recipe of what exact combinations of compute data and model size resulted in a given level of intelligence. And so um those types of algorithms and those types of recipes have been improving over time. And then also just the amount of compute that
1:51:22we've been putting into the models has been scaling up. And so that's been kind of the the underlying thing that got me to be like, okay, wow, we we we can look and see try it out over uh eight orders of magnitude like many many orders of magnitude um and see this consistent trend. And now we've seen it over even more like um over yeah over over 15 orders of magnitude. And so uh um that gave us kind of the confidence that this
1:51:50stuff would continue to advance um at the rapid rate that it that it has been which ended up leading to the to the founding of anthropic because we saw that this was going to be something that was happening not in 50 years but in on the order of uh of under 10 years. Um so that that kind of was my origin story of getting into this entire thing. Now, Anthropic is focused on making sure that
1:52:16as we go through this very rapid um uh kind of a like AI revolution, things go well for everyone. And so, I've been really grateful to get to work with uh with you, Secretary, and with um with the rest of government to make sure that this stuff is benefiting everybody. And now is a chance also, I think, for all of you guys to to be included in this uh incredible once in a once in a
1:52:44generation, perhaps probably once in the entire history of humanity um uh technology change.
1:52:53So, Anthropic has managed to stay at the frontier, right? And that uh you know, we we toss around the word frontier, but there's no real frontier. It's just that which you're creating that's new tomorrow. How how have you managed uh to stay, you know, focused and and how are you selecting where to go because you're doing things different than the other companies are doing?
1:53:22So, how what's the thinking behind it?
1:53:24And so, that'll give us a little sense of where you're going next or not going next.
1:53:30Yeah. Um that's a great question. And so I think from the beginning we've wanted to make um we've wanted to make the AIS great at scientific innovation and entrepreneurship. Um and so from the very get-go uh initially the the the first thing that took off with with the models um was uh was chat GPT um basically kind of like Google a better Google um and so you could ask them
1:54:00questions and you could they would go and answer a question much better than what you could do before. Um I think from the beginning we were thinking about like okay the main underlying source of all of the value like this this amazing room all of the prosperity that we have is uh the hard work of of our um uh of of people doing like intellectual labor. And so we wanted to make sure that the models could help
1:54:27with that. Um and so that ended up resulting in our focus on coding where that was a place where um if you imagine if I imagined just like a very smart uh friend who was uh like wanted to do work but they were trapped in a computer like what are the things that they could do to help out and like add a huge amount of of uh of their of value to the world there. And so a programmer is like a very clear way that you could do that.
1:54:55you don't need to be running like physical experiments like a a biologist would or a physicist. Um you can just go ahead and do it in the computer. Um and so that that was kind of why we initially focused on that. It also has the benefit now where our models are able to do um to help out with the research that we do where um the vast majority of all of the work [clears throat] that uh Anthropic does
1:55:25itself is software engineering um internal software engineering solving bugs in the model training. Um and so that that focus then was both useful for um like promoting things externally and also for helping us out internally.
1:55:42So my ministerial partners my my objective for today is uh growth global growth uh that they can go home and and have ideas that will help their societies their citizens grow and be more successful. So in your experience trying to figure out how to have AI around the world, what what's the infrastructure you think they need?
1:56:10What are the moves um that our you know G20 ministerial partners need to make so that they can participate with you in the in the coming benefits from AI?
1:56:23Oh, that's that's a great question. So, um I think I I I think that the biggest thing by far is uh building more making it possible to build more um data centers, more compute in your countries. It's I think it's it's it's very clearly the bottleneck for all of our progress.
1:56:43We're in a huge shortage of uh the power and labor needed to support um the demand for AI. Um, I think I think I I I really loved President Trump's uh post from from earlier this week where I think it's um it's very it he was pointing out that um uh the data centers are just an enormous source of prosperity. They produce a ton of jobs.
1:57:06They um uh they reduce taxes. Now, the way that we design them, we actually bring on more power to the grid. So um the like rateayer um pledge uh is something that we've signed up to as have uh the major other major labs and that makes it so that we bring on power behind the grid um behind behind the meter and we can give it back to the grid as well. So I think that's a place
1:57:32where um uh uh like every country can quickly can quickly be be helping out and having a big part to play here. So, you know, what I what I try to have people think about is the red carpet, right? That if your country rolls out the red carpet to great investment, they're more likely to have that investment than if it's a fight to get in.
1:58:01Yes, sounds right. So, um, you know, so in in our conversation last night, uh, Tom was suggesting that, you know, I'd really like to build in countries around the world. Can you help me? So, I said, well, actually, you're going to have the opportunity to talk directly yourself. So, what what do you you need data centers and infrastructure? Is that basically um and you should tell them what you need.
1:58:27Yeah. Yeah. No, that that that's definitely right. So, um uh I think the the the biggest things that we need in order to to build data centers are um we need permitting um for the land to be able to to build there and then we need to figure out how to get uh enough labor to do the construction um there as well.
1:58:50Um and uh then the obvious benefits to the uh to the country too is like um like now there it's a huge investment in the country. It produces jobs. Um it can be like it can be taxed which provides more revenue to the country. Um and so yeah like I I think that uh there there's many things that can block [laughter] data centers from being built so that we we could get a very long list of it. Um but once they are built they they provide huge amounts of value.
1:59:19All right. So uh our our partners build data centers. They um they embrace AI.
1:59:30What are the what's the deliveries? You know what's the outcome? How how can AI make their countries better and make the world a better place? Can you give us sort of your vision from anthropics perspective?
1:59:44what what do you bring that's different and what can you um help them achieve?
1:59:51Yeah. Um so I think there's probably two there's probably two main two main areas. One is one is just the um like re-industrialization that's caused by building huge amounts of of valuable physical infrastructure. Um and so yeah we we talked about that. I think that's that's pretty clear. Um and then a separate thing there is that uh the reason that people want to build [laughter] the like build more compute is because the models end up then adding
2:00:19value. Um and so uh the places that that's happened so far has been um mostly by uh like um scaling up the effective labor force or intellectual labor. um which is yeah like the the key the key component of the overall economy is how many smart people do we have doing good work. Um the AIS now are like not quite people but they're uh doing
2:00:47intellectual labor also and so they they feed into that same um like economic calculus that that improves that like raises all votes. Um and then I think like very concretely uh uh some of the most important problems that have plagued humanity like um uh like disease, cancer, etc. are all blocked by not having enough smart dedicated people who can be constantly working to solve
2:01:16those problems. Um, like I think I think that there's no doubt that if we had uh like a million times more uh oncology researchers, we could make progress that's not just like uh like 10% faster, it would be like multiples faster. Um, and so that that I think is the is the the the thing that will come from this very large scale up of more and more smart models. Um and and we see that now
2:01:46with um with math. So uh historically the models were doing more wrote tasks like you would ask them to do um like uh simple things that like maybe like a parallegal work or something like that or like a basic software engineering task. Now they're doing much more complex software engineering tasks. Also with math they traditionally could do okay with with math and then they got to the point where they would be like um like an undergraduate level mathematician graduate level. Now over
2:02:15the last three months they've gotten to the point where they're like a a once in a generation genius on math. Like any of the frontier models now are I I think they're like a top 100 current mathematician in many of the sub fields of mathematics. um which is just astonishingly fast. Um and so many aspects of science are kind of a combination of engineering plus like mathematics. And so my expectation is
2:02:43that over the next 12 months we'll see kind of that same thing where like now they're like okay scientists or biologists. But um I my prediction is that yeah within the next 12 months we'll get the same thing where like they could be a a once in a generation um scientists for key scientific fields. So it's your view then that using anthropics model uh you can really be
2:03:12aspirational right meaning that what what are what are the problems facing your economy your people and use the models to try to really deeply examine what could be solutions to those models and use the the sort of bottled logic of anthropic to uh to achieve that. So how how would one think about that from from your perspective because
2:03:40you you know it's capable in anthropic and you know where it's going. So maybe give us some sense on how to solve those kind of problems.
2:03:48Yeah. Yeah. And I think that's exactly right. I think that um uh that um often people are not ambitious enough in what they ask the models to do. And so I think that uh um like because the way that the model started out was like you would ask them a question uh that like like you would ask Google and it wouldn't have that much context. It would give you like a a a page long response or a paragraph response. People often will take one of the new models
2:04:16from the last like three months and do that same thing and they're like ah it's like slightly better. It's like cool I guess. Um, but the way that uh I see the the best people who get the most value out of the models using them now is instead they treat them like they would treat a um a collaborator uh like a a co-orker in their organization. And so they do the same thing that you would need to do if you were like, I've hired the smartest person I know. I want them
2:04:45to be as productive as possible within my organization. And so you go and you make sure that they have access to all of the documents that they need. Um you make sure that they have access to uh the tools. If they're a software engineer, you're like, I'm going to make sure that you have um all of the documentation for the code that you have access to the actual servers so that you can go and run things. If they're a scientist, you make sure that they have access to like the machines that can actually move stuff around. with a model
2:05:15like that would now need a an API to move the machines, but that's a thing that you can get for them. Now, um you you also need to make sure so like in in government, you would make need to make sure that they have the right permissions to access the right information that they would need to go and do the work. Um and so I think once you do that, then you end up with a model that has all the things that a great co-orker would need in order to to do work. And we see that that's now
2:05:43actually how the majority of the um like intellectual labor within anthropic is done is by these models that have the same permissions that a human co-orker
2:06:48That's why I really encourage everybody to engage artificial intelligence. It bootstraps you, it elevates you, it enables you. You know, it's great great equalizer.
2:06:58So, what advice would you give to all of our uh global partners here? How how can they best harness this opportunity going forward for their countries? What do they have to build inside their countries? what investments do they have to make so that they go home and they think about if we want to participate in the in the AI you know revolution what's the best way for them to do it well the first thing is to recognize
2:07:26what is AI you know I've been using this word infrastructure but when you break it all down AI is if I could simply is a five layer cake at the lowest layer is energy you can't produce something without energy it transforms electricity into mathematics and so first layer is energy the second layer are chips that's the world that I'm in the third layer is infrastructure basically these large data centers land
2:07:55power shell that incorporates all of the computer technology and energy is plugged into it and money starts spewing out okay so it's kind of the next layer is is essentially infrastructure the layer above that is where AI models live and most people think that That's what AI is. But in fact, AI is a five layer cake. And AI models uh you know I for many of the countries that are here, I work across many of your countries already today. Whether it's in startups,
2:08:24established companies, researchers, and each one of the regions around the world are specialized in something very something really important in artificial intelligence because it's not just language. artificial intelligence can learn and understand and predict almost any information that has structure. So biology has structure, chemicals has structure, the laws of physics has structure and so each one of the regions around the world specializes in different areas in AI and AI cuts across
2:08:52all of it. Different languages, different technologies, different information theories. And so AI cuts across all of that. And there's a whole bunch of AI models. But the most important part of AI is the fifth layer the layer above that which is data and applications. The way to benefit you know United States is more deeply invested in uh inventing building uh developing the
2:09:20entire five layer cake. However, every single region every single country should decide which one of the layers they want to invest in. You don't have to win in every single one of them. You don't have to develop all of them. The most important part is that every single country should advocate for AI diffusion or the adoption of AI into your industries. And the industries that we're talking about span everything from education to healthcare to manufacturing
2:09:50to sciences, you know, every single industry in between. And so the way to think about AI is just to think about it from the five layer stack. and United States is going to lead in the development of AI across these five layer stacks. But we'd love to partner with everybody so that everybody could take advantage of this. Nvidia has built it in such a way that every single model in every single field of science and every single modality uh runs and so it runs American American uh models, it
2:10:20runs international models, it runs models in biology and chemicals and you know physics and robotics. So when if I go to a dinner party, Yeah.
2:10:29and I say AI, everybody thinks it's software.
2:10:33But what we're seeing now is really AI at the edge. It's it's hardware. It's robotics. It's advanced manufacturing.
2:10:42So how do you see AI and the difference between the physical infrastructure of factories and sort of the general perception that AI is software out there? Well, AI is intelligence and and it starts with software. Um, and in fact, if you look at AI in the last year, we started by talking about large language models. This LLM, you know, this concept that the world would know what LLM stands for. It's kind of an amazing thing. And so, large language
2:11:11models, what made AI useful is putting an exoskeleton around the LLM. That exoskeleton is what is called an agent harness. It's essentially gives the brain the the parts necessary to uh retrieve knowledge, have working memory, uh use tools, be able to collaborate, so on so forth, and so solve problems. And so the first thing is that the digital version
2:11:40of AI is a is what is called an agent. When you take that digital version, you could put it instead of using a tool like spreadsheet or PowerPoint or Excel or web browsers, you could make it use a tool essentially a machinery. And so this agent would use uh would be embodied inside a physical physical body and all of a sudden it becomes a robot.
2:12:02You could take this agent and put it inside uh a machine that has four wheels, self-driving cars. You could uh put this agent inside a manipulator and it becomes a pick and place or a manufacturing robotics arm. Uh you could put this uh agent inside something that that essentially is a grocery delivery or or a logistics delivery vehicle or a surgical robot or a autonomous uh drug discovery laboratory. You know, all of
2:12:31these different versions are basically the same idea. Large language models with an agentic system around it. you could use this this agent agentic system could operate um whether it's digital tools or physical tools. Now in the end what what every single country will have to do is you need to recognize that this is infrastructure just as it's water, roads, electricity, the internet. Every single country needs to build infrastructure so that you could support
2:13:01your own local economy. It becomes the capacity, the intellectual capacity, the digital intellectual capacity for you to support your researchers and students and society and industries and startups and every single every single country that we've worked with and there's so many here that we've partnered with already when the moment we create an infrastructure locally all of a sudden the startups the researchers just really get activated and so it's really a fantastic investment to make. We've had the benefit in the United States uh to
2:13:31have built one of the largest infrastructures and and um uh because because of the pro-en energy growth and the the speed um and the regulatory environments in the United States, we've been able to build out really quite substantially. And you you look at what's going on in the United States, AI is now generating jobs across every single sector from chip fabs to computer plants to AI factories. Um boy just the number of jobs that we're creating is just hundreds of thousands and as you
2:13:59know and and uh this year alone we're going to invest close to a trillion dollars into the world's into the United States infrastructure and that's generating enormous economic growth and of course a lot of jobs.
2:14:12So the change makes people nervous.
2:14:16You know it gets all sorts of people to say all sorts of uh of things. you know this AI change how how do you rank it as compared to the other change all the way back to industrial revolution you know where do you put it is it small is it big from your perspective and and what's the best way for uh for our our friends here to embrace it from your perspective well as with all new technology uh at first it see it seems uh uh magical you
2:14:45could just imagine in the beginning of the production of electricity that somehow uh You could be far far away and all of a sudden you know a light could turn on. It seems like magic and uh in the beginning of all technology innovations it seems like magic but the fact of the matter is all of us who are building it this is engineering there's computers there's software there's mathematics involved um there's probabilities involved and and uh uh
2:15:13there's a lot there just as in every technology industry when we're building something whether it's a car or a plane it is the responsibility of the people building it to do it safely and responsibly and to continue to advance the technology so that as it advances it becomes safer. I prefer to drive today's car with so much more technology in it than to drive a car from 100 years ago.
2:15:36I prefer to fly in a plane today. Um it's with all of the technology advances than a plane 100 years ago. And so in a lot of ways we want to accelerate the advancement of the technology so it could be more functionally um uh able to deliver on its promise so that it's safer to to uh to use. And so the same thing is is going on with AI. We have to rem remind the industry um that and I think the dialogue around safety and
2:16:03security is all fantastic. Uh however, if you think about each one of your countries, what is the worst outcome that could happen in this technology industrial revolution? The worst outcome is that you don't take advantage of it.
2:16:19That you are left behind. That is the single worst outcome. And that outcome comes from two that outcome could happen if we talk about the technology with fear with uncertainty. The leadership um talks about it uh with so much concern so much fear and uh all the conversations around safety and security. It it has to be balanced with a lot of conversation about prosperity and a lot of conversation about what it can do. And so so I don't think it's
2:16:48about one or the other. We have to make sure that we're balanced in talking about it. Um, you probably know that that uh in the history of of all of the technology advancements, one of the one of the stories I was just reading about recently uh has to do with the the airline industry. There was a time when all of them spent, you know, a lot of their advertising dollars and and marketing campaigns on my plane is safer than everybody else's plane. Turns out nobody wanted to hear about that. They
2:17:17wanted the c they wanted the companies to take care of that. What they wanted to hear was that that the airlines could take them to new places, expand their horizon, make the world smaller so that we could learn about each other's culture, create new economies. That's what they wanted to hear about. They wanted it to be our responsibility to make it safe, not to make it their burden that they think about safety all day all day long. And so I think this is the same thing that's hap it happened in
2:17:46the auto industry. It happens to to the power generation industry. it happened to the airline industry. And so I think it's a responsibility for all of our all of the leaders all of us to make sure that we take safety and security and building the technology properly our responsibility and uh working with regulators and of course every single there are so many regulations and so many laws already. Intelligence is a digital digital version of what we do and so you would think digital version
2:18:15of some task that humans do. So in fact, humans are have regulations across so many different domains already, whether it's the FDA, FDA or Nitsa or whatever it is, there's so many different agencies that already have regulations.
2:18:29I'm sure there's a way to formulate AI into that. And so uh I think the the balance of of precise regulation uh don't regulate hypothetical theoretical harm, regulate actual and pragmatic harm. um and give this technology which is at the at the early stages of the S-curve innovation an opportunity to advance quickly so that it could be safe. And so some of the people uh to if you just go back and think about some of
2:18:59the and remind yourself of some of the ideas of slowing down technology advancement three years ago when it was just being formed. In fact, the advancement is what made it safer. The advancement made it not hallucinate grounded on real truth. um iterate until the the answer uh is uh uh uh scientifically sound. Um all of those technologies made were made possible through technology advancements. And so we're we're making it safer but
2:19:28advancing it more.
2:19:30So let let's turn a little bit to your business for a moment. Um your your chips uh GPUs, they're they're different from what other people are hoping to build, right? And so maybe can you talk to the audience about about your products and and why are GPUs different than so any other letter before the PU if you will? Well, today we started out, we invented a technology called a GPU.
2:19:58And basically a GPU does parallel processing tasks and it solves things like fluid dynamics and particle physics and quantum chemistry and one of the computer graphics, image processing, um image reconstruction.
2:20:13Uh and of course, one of the things that it does incredibly well is artificial intelligence. Um our architecture, our technology is general purpose meaning that uh any we support every single closed models, we support every single open model, we support languages, we support large models, small models, we support um physical models, biological models and so our architecture is funible in the sense that it runs
2:20:42every AI model, runs all data processing and and parallel processing tasks. As a result, we're everywhere. We're in every cloud. We're on prem. We could be at the edge. We're in robotic systems and self-driving cars. Um, and the com the company's been around. We working on this for so long. We're in our 14th generation of our architecture. And and so, so Nvidia's basically everywhere.
2:21:05One of the advantages of doing that is that these infrastructures are so expensive. One gigawatt is about 50 60 billion dollars. You know, back in the old days, people thought building a fab $25 billion was a giant investment. And now, literally one gigawatt is twice that. And we're building between now and the end of the decade 100 gigawatts because we want to produce intelligence for the world. And so when you're building investing in infrastructure
2:21:33that's that costly you want to make sure that it is fungeible and it's durable because otherwise if you build something that's too specialized and the world's architecture and algorithm changes and these AI models and technologies are changing so fast if it changes then all of a sudden it could obsolete the investment that you've made. And so the reason why Nvidia's uh growth is accelerating is because people are starting to realize that we're the
2:22:00safest architecture to invest in. It's fungeable. It is durable and it continues to improve over time through software and uh it runs everything and so we run every American model, every international model and every modality of models that you have.
2:22:16So people are always interested on where the world is going. If if you were going to give us sort of your big picture view, where are we going? Um what should we be thinking about that we don't always talk about? You know, what you you see everything really all the companies of the world sort of come and visit with you and sort of tell you where they're going. How how would you sort of give us a big picture of a set of ideas that we don't usually hear about?
2:22:45Well, there's some things that that we hear and and then I'll I'll I'll juxtapose that. um against what I expect. And so, for example, in the next couple of years, we are going to achieve essentially what people call AGI.
2:23:01And and um uh and that's not an illogical thing to me. In fact, I would argue that we're practically there today. Okay? And and so the question is what does that mean? Um it either means a lot or it doesn't mean anything. And so, let me give you an example.
2:23:18If we said that in the next couple of years we're going to achieve artificial general intelligence um and every and there's no question everybody's everybody's tasks the things that we do whether it's in um just learning or working or uh every single industry will be impacted it will benefit everybody.
2:23:38However, the part that is that is not not thought through is in fact uh the thought experiment. Let me just give you the thought experiment. Suppose and you know that several several hundred years ago uh humans in order to have a more civil society uh manufactured one of the most important things in the world which is called education intelligence. We manu through universities and schools we manufacture intelligence and we manufacture stability at scale. is without it, how would we have
2:24:08civilization? And so now we're creating the digital version of that. We're manufacturing now intelligence digitally at scale so that even even countries and people who don't have access to the highest levels of education now have the benefits of the highest levels of education. Okay? And so that's the simplest way of thinking about what is happening right now. So the question is what's the benefit to society as a result of that? Well, let's pretend let's do a thought experiment. Suppose
2:24:35every single uh every single, you know, young person that we hire out of school uh is incredibly well educated. Well, that's pretty much consistent with everybody I hire. They come out of Stanford or Harvard or MIT or, you know, ETH or and these are incredibly smart kids. However, when these kids come to come to our company and every every company and every industry, we now still have to invest quite a bit in surrounding them
2:25:06So, as it turns out, even when we get to AGI, it's not as if every company's problems are solved.
2:25:14It's not as if all of a sudden in two years time, I say, "Okay, let me connect it to this a a this AI service." and I just sit sit back and all of a sudden my company becomes more and more productive overnight. That's just not going to happen at all. You're going to spend all of your energy just like you currently do onboarding a new college grad with a PhD from MIT which is pro surrounding them with context, purpose, relevance, access to, you know, all of these other
2:25:44things which is essentially if you will the harness around these AIs, the environment around these AIs so that they could do productive work. All of that harnessing, all of that environment creation is what companies do. It's what leaders do. And as it turns out, our jobs won't change.
2:26:06Our jobs is defined by purpose, context, meaning. That's what a job is. In that job, there's a lot of typing and talking and, you know, things like that. Those skills are going to be automated by AI. the tasks are going to be automated, but our our jobs purpose re remain. And so I think what's going to happen is that we're all just going to be supercharged just like we're currently supercharged.
2:26:32Nvidia is we use AI across the whole company today. We use off-the-shelf AIs like like anthropics and open AIs and cursor and um and we build a lot of our own custom AI and I encourage everybody to apply the same the same idea. You should use AI off the shelf wherever you can. But don't forget, you have to invest in building your own intelligence. Every country, every company, you can't outsource all of your intelligence to somebody else. And so,
2:26:59you have to use as much of it as you can build as much as you need to. And so, uh, we do that at the company. And and, uh, and so that's the the first idea is just that we're going to be supercharged. um the idea that somehow all the jobs are going to be eliminated or removed that's just nonsense. Okay?
2:27:16And the reason for that is because of all the things that I've just said. Now the other thing that that is the case is that all of our ambitions are going to be supercharged. Just as you know we are more ambitious today than any time in human civilization. We think about doing things that people a 100 years ago would just that's insane. That's got to be science fiction.
2:27:43And yet we perfectly expla perfectly expected. And so just imagine how ambitious we're going to be in 5 10 years. That's the exciting time. You know, I've been working a long time and surely I don't have to work but I don't want to miss I I'm definitely not going to miss the next 1015 years. The ambition of the industry, the ambition of my colleagues, my own ambition has really been supercharged. And this is another way of saying that all of all of you, all of the countries, your
2:28:11ambitions have to be supercharged. This is the time when you have to feel enlightened, inspired, you know, supercharged, lifted because you now have this incredible technology behind you. And so, you know, what do I expect in the next five, ten years? Uh, things that takes 10 years going to take one. things that take one will take a month.
2:28:32And so, um, you know, if you if you think that that this is going to cost a million dollars to do, uh, well, back a long time ago, just, you know, years ago, a million dollars seems like a big investment, but these days, just think about the level of investments. We're talking about trillions of dollars in investment. And the reason for that is this.
2:28:54It is now possible for average humans like ourselves to think that we can make an impact on a hundred trillion dollar industry. For the first time, you're hearing people, Isn't that right? For the first time, you're hearing innovators, people, companies, countries talking about elevating a hundred trillion dollar industry. That's ambition. To think that you're going to add 2050 trillion dollars of economic benefit around the world, that's ambition.
2:29:28No time in history have we been able to say things like that. So my goal for today when we're together is growth. Right? That's our goal that we can all grow and benefit our countries and benefit society. And I think what you've done, uh, Jensen, which we really appreciate, I mean, you are an American treasure. Thank you.
2:29:54Right. Your company, uh, is leading us in a path that none of us when we were young ever dreamed possible. Maybe we dreamed it a little bit. That's what, you know, sort of sci-fi was about, but now we're there. I think what you've laid out for us is a vision of that harnessing AI to using it both from uh surrounding it, contextualizing it and directing it is how we're going to
2:30:23succeed. That each of the countries here needs to invest in its infrastructure in order to capture that for themselves to make sure that their citizens and their society benefits from it. They've got to make that infrastructure investment. They've got to build the data centers. They've got to harness the AI for themselves.
2:30:42Because what you've said is that uh education has been democratized. But it's made available for everybody. You now can have a PhD in everything.
2:30:53At your fingertips and you don't have to find that person because you've helped find them and sort of capture that education in a bot. You are that person now, right? Everybody is that person now. So I think you know our goal is to uh is to have that idea of growth, that ambition for growth. And what you've done is you've laid us out that our dreams need to be higher because they can be higher because if you have all of
2:31:22this intelligence at your fingertips, how could you not dream higher?
2:31:26That's right. And I think really for a company and for a leader of a company to have such a uh such a brilliantly positive vision for where we're going, I think was the exact way to lead off today. So I want to thank you for coming everybody. I would like to thank Jensen W for coming and joining us [applause]
2:31:52for our conversation. So, I'd like to introduce you all to Tom Brown, one of the founders of Anthropic. So, Tom, thank thanks for coming and joining us. Would you mind giving um my my uh my partner ministers uh sort of a background of yourself and Anthropic because I don't think they know you so well. Anthropic of course is famous, but a little a little story would always be helpful.
2:32:16Yeah, I'd love to. Um and yeah, just first off, thank you so much for uh for inviting me to this. It's an honor to be here and to get to speak with you all.
2:32:24Um maybe I so I'll talk through a little bit about myself and my background. Um maybe even before that I think like probably three things that I'd love all of you guys to take away from uh this conversation. One is that uh AI capabilities have been increasing very rapidly.
2:32:44obviously um I expect them to continue to be increasing really rapidly and consistently. Um two uh as a result of that we are seeing this demand that's not just fast but also like exponential so like increasing in multiples every year. Um, and then I think three, as a result of that, uh, we have this incredible industrial buildout the likes of which we've never seen in the history of humanity, even more than the the
2:33:13railroads of the 1800s. And so that's a place where um, any country that wants to be involved has a chance to to be involved. Um, so I guess my my background uh, is um, I was an engineer originally. Uh I'm now the um chief compute officer at Anthropic. I kind of got into this back in um 2014 where uh a few of my friends came to me and they were like hey I I had done
2:33:43startups for many years. They came to me and they were like Tom uh it seems like we might end up getting uh human level intelligence from AI within our lifetimes. So, in the next 50 years, um, and I was like, "Wow, that like seems kind of crazy, but if it does happen, like that would be the biggest most important thing that ever happens to humanity." Like the things that makes us special is our minds. It's not our body.
2:34:12The industrial revolution transformed all of society. But really, it was like like the second most important thing about humans is our bodies. The main thing is our is our minds. Um, and so that that got me to think like, okay, maybe I'll switch from engineering to try to help out learning about AI. Um, and so, uh, I did I started doing kind of self-study on that. And then about a year after that, um, uh, OpenAI was
2:34:42founded. Um, and so I knew some of the folks from that. I'd long been a fan of, uh, of Elon Musk. And so I went to see if I could help out in some way. And so I joined I joined them as like one of the first 20 employees basically to do engineering work. Um and then uh uh kind of learned more about that. And then I think the the key thing for me where I was like, "Oh, wow. this actually is going to be a big deal that's not just
2:35:10in 50 years but faster was um the scaling laws paper um which was in 2019 and that was basically looking at um within AI people had seen from like 20 uh for for a decade before 2019 that if you like the models were getting bigger and bigger people were using more and more compute and the quality of the models was improving over time um but they didn't have a good understanding of
2:35:39how like what the what the mechanism was for taking the bigger models and making them smarter. And so this was a thing where some of my collaborators um uh now anthropic co-founders um went and tried to do the same kind of scientific research that you would do if you were like in the early days of a combustion engine. And you were like, okay, like we see combustion engines, bigger engines provide more power, but uh how much
2:36:08power do you get from an engine? And if you like scale up the size of the engine or change
