# NVIDIA's Jensen Huang on Building the Dynamo of the Intelligence Age https://www.youtube.com/watch?v=2UpQbeAZuqA Translation: zh-CN [00:00] Thank you so much, Jensen. So, uh we are [00:04] in the middle of a massive AI [00:08] revolution. Uh it is probably bigger and [00:13] faster than even the industrial [00:14] revolution [00:16] and you have called out what's happening [00:18] right now as the largest infrastructure [00:22] buildout in human history. At the center [00:25] of that buildout is the AI factory and [00:28] the company enabling all of that is [00:31] Nvidia. Can you tell us what is an AI [00:35] factory and why is it the best [00:37] investment for any enterprise in the [00:39] next decade? [00:42] >> Okay, so um you could you could [00:45] understand AI in a in a particular [00:47] number of ways. the the way that you [00:49] understand AI probably most is through a [00:52] chatbot through a web browser. Uh you're [00:55] interacting with it. You give it a [00:57] prompt. It says something back to you. [01:00] And and even those of you who have been [01:02] using AI for some time uh you've seen in [01:05] the last couple two three years a very [01:08] significant evolution improvement in the [01:11] capabilities of AI. Two years ago, you [01:14] heard about chat GBPT. And ChatGpt [01:17] basically is a computer software that [01:21] understands the input you give it. It [01:24] can perceive understand information [01:27] and it can translate and generate the [01:31] information into something else. Okay? [01:33] So, you can give it a prompting. You can [01:35] say, "Here's this PDF I gave you. Um, I [01:37] would like you now to summarize it. It [01:39] went from text to text." You could also [01:42] tell it, here's a PDF I gave you. Um, I [01:44] would like you to now generate an image [01:47] of that story. It go text to image. You [01:50] could use go from image to text, meaning [01:53] you could give it a picture and you can [01:55] what's happening inside this picture. It [01:57] goes text uh image to text. Does that [02:00] make sense? Anything to anything else. [02:02] And AI in the last in two years time, [02:05] two years ago was largely able to do [02:07] this translation. We called it [02:09] generation generative models. Okay, [02:12] generative AI. But the the thing that is [02:15] very big deal inside that word [02:18] generative AI is in order to do [02:20] something even more valuable than [02:23] generation p understanding and [02:26] generating is thinking. Well, you can't [02:29] think if you don't generate words. And [02:32] so the fir the foundation of generative [02:35] AI gave us the ability to generate [02:37] internal thoughts, thinking, reasoning, [02:40] step-by-step reasoning, problem solving. [02:42] It also allowed us to do another thing [02:45] that is now very important, which is [02:47] generate intelligence to control [02:50] something else to generate control to [02:54] use a tool. Does it make sense to use a [02:56] browser, use a spreadsheet, use [02:58] Photoshop, use PowerPoint, use [03:00] something, use AutoCAD, use another [03:02] tool? Now that tool today is digital, [03:06] but someday that tool will be [03:08] mechanical. So if I generate a command [03:11] to a mechanical system, that would be [03:13] called robotics. If I generate commands [03:15] for uh a machine with steering wheel, [03:18] that would be called self-driving cars. [03:20] Does that make sense? Okay. And so two [03:22] years ago, two years ago, in fact, you [03:25] saw the foundations. We call it chat [03:27] GBT. And everybody said, "Ah, you know, [03:29] it's fun. It's it's silly or it produced [03:32] a whole bunch of crazy hallucinated [03:34] text." That's all true, but it was the [03:37] foundational technology that led to all [03:39] of this. [03:41] Two years later, we now have agentic [03:43] systems. Now, that's one view of AI. I [03:49] just described the view which is what [03:51] can AI do right and so now all of you [03:55] realize you see it from chatg you see it [03:57] from codeex you see it from cloud code [03:59] you see you see that it's now able to [04:02] not just understand but it's able to do [04:05] work reason and do work now two years [04:08] ago when AI was able to understand you [04:11] and generate information that was [04:14] interesting novel [04:16] a little cute. Whenever you need a poem [04:19] written, great way to do it, right? Who [04:23] doesn't want to write a country song? [04:25] And so, so that was two years ago. But [04:28] now, because it's able to do work, AI is [04:31] valuable. Valuable meaning it can [04:34] generate information. It can generate [04:36] useful work, and it could be paid for [04:38] it. Because we pay for we're we're [04:40] interested in having friends that are [04:42] smart. We love people who are [04:44] know-it-alls, [04:46] but we don't pay them for it. We pay for [04:49] people who do work. Does that make [04:50] sense? All right. Which is what happened [04:52] in the last two years. AI went from [04:54] having this capability to now agent went [04:57] from not use not very valuable to now [05:00] producing useful work. So much useful [05:03] work that you and I are doing this every [05:05] day. We're paying AI by the hour, right? [05:09] And so we might pay them $30 an hour to [05:11] do the work, $20 an hour to do the work. [05:13] We're basically paying AI a lot of money [05:15] today. The fastest growing software [05:18] business in the history of mankind [05:20] because now it's doing useful work and [05:22] we can pay them to do it. Now that's one [05:24] view of AI which is what it can do. But [05:26] one other view of AI that's really [05:28] important to help reason through what [05:29] Constantine is saying. So for example, [05:32] the reason why some companies, some [05:34] people are able to build great [05:35] businesses and could maneuver themselves [05:38] into the center of very large industries [05:41] is because when they see this [05:42] capability, [05:44] this is very interesting. One [05:46] interesting thought is if we're able to [05:48] do this, what is the implication to this [05:51] downstream industries? That's an [05:54] interesting conversation we should have. [05:55] Okay. So now that AI can do this, what [05:58] happens to all the industries like [05:59] healthcare and financial services and [06:01] life sciences, manufacturing, logistics, [06:03] transportation, [06:05] on and on and on, retail, advertising, [06:09] future entertainment, the the list the [06:11] list of the list of conversations you [06:14] can have about now that AI can do this, [06:16] what can it do as a result subsequently? [06:18] That's an interesting conversation, but [06:19] you should go upstream meaning [06:21] industrially what does that mean? And so [06:23] the first thing you realize is this. Go [06:25] back to first principles. [06:27] I had told you just now that AI is [06:30] software and it's being produced by a [06:32] computer. Now what happened to the [06:35] computer that is made it possible to do [06:37] this? Well, the big idea is about if you [06:42] think about the computer as we know it [06:43] today really emerged about 64 years ago. [06:46] IBM system 360 was the biggest [06:49] announcement of computing and 64 years [06:52] ago IBM was the most valuable company in [06:54] the world. Okay. And they created the [06:56] modern understanding of computers. [06:58] Everything that we can describe about [07:00] computer was really described in 1964. [07:02] For 60 for 40 years largely has remained [07:05] the same. And what happened was what [07:09] happened was that in that form of [07:11] computing is called pre-recording. [07:14] You write down your story. you save it [07:16] to a file. You write a program by hand, [07:18] you save it to a file. You uh take a [07:21] picture, you save it to a file. You [07:22] record music, you save it to a file. You [07:24] make a video. Right now, we're in [07:25] streaming right now, but somebody's [07:26] going to record it. You're going to save [07:27] it to a file. And when you want to use [07:29] it later, you retrieve it from the disc [07:32] drive. Does that make sense? And the [07:34] retrieval process is done intelligently. [07:37] So that's why everybody's retrieval of a [07:40] news story is a little bit different. [07:41] It's called a recommener system. But [07:43] basically computers as we know it today [07:45] is a retriever retrievalbased system [07:48] which is the reason why these buildings [07:50] are called data centers. They store [07:53] data. [07:54] Notice they don't call them computer [07:56] centers because you're not doing much [07:58] computing. They just store data that you [08:00] retrieve based on what you touch on your [08:02] phone. Well, what happened now? If you [08:06] look at what I just described, in order [08:07] for this AI to work as I described it, [08:11] every time I say something to it, [08:14] I have to give it new information, we [08:16] call it context, I give it a new prompt [08:20] that's called a query. [08:22] Between the context and the query, it [08:25] will understand it first, reason about [08:28] it, and it will produce an output based [08:30] on that context in that query, based on [08:33] the circumstance. Does that make sense [08:34] so far? Okay, give me one nod. Okay. And [08:37] so if that is the case, every time you [08:41] use the AI, [08:44] the content is produced originally every [08:47] single time. Everything I'm saying to [08:49] you right now is being produced in real [08:52] time. And it's because my explanation is [08:55] based on the fact that I realize all of [08:57] you come from 60 different countries, [09:00] 128 different families. You all have [09:02] many different backgrounds. Some of you [09:05] probably came from the computer [09:06] industry. Most of you probably did not. [09:08] And so I'm explaining the information to [09:10] you in a way that is sufficiently deep. [09:13] But ultimately my goal is this. So that [09:16] you know how to make your next [09:19] investment. That's what I'm leading to. [09:21] And so I'm going to give you enough [09:23] sufficient information that you can [09:25] reason about it for yourself so that [09:27] when you see something in the next time [09:28] you go, that's worth investing in. [09:30] That's going to be a big industry. [09:32] That's a hundred billion dollars right [09:33] now and it looks really big, but that's [09:35] nothing compared to how big it's going [09:37] to be. I'm going to give you the [09:39] intuition to solve that problem. Okay? [09:41] And so, so here we are went from a [09:43] computer industry that was largely based [09:46] on retrieval for 60 years and all of a [09:47] sudden one day it's completely generated [09:50] in real time. We call it intelligence. [09:53] This this is what I'm doing right now [09:54] for you. I'm demonstrating intelligence, [09:57] contextual awareness. He gave me a [09:59] prompt and here comes my answer. Does it [10:01] make sense? [10:02] >> Extreme intelligence. Extremely [10:07] artificial intelligence. Okay. And so [10:11] and so here I so here here what's going [10:13] on now. I just gave you one word [10:15] earlier. It's called generative AI. The [10:17] the computer of today has been [10:19] completely reinvented. It's now [10:21] generative. [10:23] And every single toe, every single [10:25] letter you see, every single word you [10:26] see in your in the future, every video, [10:28] every image, every ad, every TV [10:31] commercial, every single time you read a [10:33] story, [10:34] >> a news story, every one of them will be [10:37] different. What Constantine sees, what I [10:39] see, what you see will be completely [10:41] different because it'll be generated for [10:43] you. because your your interest, your [10:47] context, who you are, for what reason [10:49] you asked, how you asked is completely [10:52] different. Does it make sense? And so [10:54] therefore therefore [10:57] every single pixel that you see, every [10:59] single sound that you hear in the [11:01] future, every video you see in the [11:02] future will be originally generated, not [11:05] retrieved. Which means in the future we [11:08] need a lot more generators. [11:11] >> And these generators is what we built. [11:14] That's what we build for a living. These [11:16] are large computers and they're [11:18] generating intelligence. Now, the next [11:20] question is this. [11:23] Well, how big could it be? How big can [11:25] it be? And so, it turns out it turns out [11:29] the amount of information, the amount of [11:31] generation of intelligence is we do it [11:35] for about a billion people in the world [11:36] today. [11:39] Now that I told you that AI has become [11:42] agentic, meaning that it can actually do [11:44] work by itself. Well, if it can do work [11:47] by itself, then one agent can [11:49] communicate with another agent and say, [11:51] I have some work to do. Let's team up [11:52] together. Let's do some work. And now [11:54] you have all these different agents and [11:56] they're all working together to solve [11:58] problems inside your company. Say, so [12:00] inside our company, Constantine knows [12:02] that we we're huge users of agentic AI. [12:05] We have hundreds of thousands of agents [12:08] probably running around right now that [12:09] are doing work and they're talking to [12:11] each other and they're solving problems. [12:13] All guardrailed, all sandboxed, [12:16] all guardrailed and sandboxed, but [12:18] they're all working with each other. [12:20] Which means in the future it is very [12:21] likely that the internet that we use [12:24] today for a billion people will likely [12:28] mostly be several billion call it a [12:32] hundred billion agents working around [12:34] the clock and they're using the internet [12:36] and talking to each other and what are [12:38] they saying? So for example, there'll be [12:40] uh companies working with companies uh [12:42] employees agents working with other [12:44] employees agents. Uh there'll be [12:46] self-driving cars which are agentic. [12:49] There will be robots which are agentic. [12:51] All the manufacturing systems, every [12:53] building will be agentic. There'll be [12:55] agents all over the place and they'll be [12:57] using the internet and all of those [12:59] commands that they generate to each [13:01] other will be generated. all of the [13:04] thoughts that they have to understand [13:07] all generated. Does that make sense? And [13:09] so the basically the world is going to [13:11] be this layer of computing [13:14] that's going to cocoon the earth and [13:16] it's going to be generating intelligence [13:18] all the time. Now I just said something [13:20] that sounds ridiculous [13:23] except in fact it's already happened [13:26] >> twice. [13:28] So 300 years ago [13:31] a company in Germany called Seammens [13:34] produced a machine and this machine is [13:36] really interesting machine. You go up to [13:38] this machine you light it on fire [13:42] and then this incredible invisible force [13:44] comes out the other end. [13:46] >> Nobody understood what it was. We [13:48] understand it now as electricity. [13:52] How many power generators are there in [13:53] the world? We call it a grid. [13:58] Power generation cocoons the planet. We [14:01] call it the grid. And then of course 20 [14:04] some odd years ago, [14:06] earlier than that, 35 years ago, this [14:09] networking scheme, networking [14:12] matrix fabric was created here in the [14:14] United States eventually became the [14:16] internet. And where is it? It cocoons [14:18] the world. And so now you have energy, [14:20] communications, intelligence, and it [14:23] will cocoon the world. and we'll use it [14:24] for, you know, it'll just be a commodity [14:26] and we'll use it all over the place. And [14:28] so what Nvidia does for a living is this [14:30] new machine. The machine that was [14:32] invented three years, 300 years ago is [14:34] called the dynamo. [14:36] That dynamo, [14:40] anything that moves comes in. Could be, [14:42] you know, waterfalls, it could be wind, [14:44] fire, steam. [14:47] transfer it from motion atoms [14:51] right to electrons [14:55] atoms to electrons. We then take the [14:58] electrons into our machine called [15:00] Nvidia. [15:01] Electrons now comes into our machine [15:04] comes in this factory and what comes out [15:07] are numbers. These numbers [15:10] depending on how you combine them turns [15:13] into language [15:16] math. It can also turn into a new [15:19] language. We've learned proteins. We [15:22] learned the the language of human [15:24] biology. We learned the language of the [15:27] physical world, physics, climate, [15:30] weather. We learned the language of the [15:34] 3D world, robotics, self-driving cars. [15:38] We learned the language of all kinds of [15:40] different forms of intelligence. [15:42] But the point being now these two [15:46] machines 300 years apart, [15:49] atoms in, electrons out, electrons in, [15:52] numbers out. And those numbers could be [15:54] re rejiggered, reformulated into all [15:56] kinds of different intelligence. That's [15:58] what we built. That's what we do for a [15:59] living. And that's why I call it a [16:01] factory because it's producing. We call [16:04] them tokens, but they're just numbers, [16:06] tokens. [16:08] And these tokens are intelligence. [16:10] That's it. That's what we do. It's not [16:14] that hard. Brilliant. [16:24] Now you know what AI is for. And now you [16:26] know what AI's how AI is built and how [16:29] how big it's going to be. [16:31] >> Thank you all for joining today. [16:32] >> Thank you. [16:34] >> Good job. Excellent question. [16:39] >> I I've really felt that carried it. You [16:41] know the prompt. Um [16:43] >> you laid it up for me. [16:45] >> Um okay. So this is a massive [16:47] revolution. Yeah. [16:48] >> Um and you you laid out three [16:50] transformations. uh energy [16:52] transformation which touches everyone [16:54] today and a lot of the people in the [16:56] audience are part of these manufacturing [16:58] and energy producers globally, [17:00] telecommunications which connects all of [17:02] us and now intelligence and in energy [17:05] you talked about the generator [17:07] telecommunications I guess the [17:08] comparable would be the switch or [17:10] something along those lines for routing [17:12] communications globally and now in the [17:14] intelligence revolution it's the GPU at [17:16] the core and the AI factory like the [17:19] H100 or any of the the new systems that [17:23] bring everything you need under the same [17:25] hood. Vera Rubin, what have you [17:26] >> and these these factories, so you know, [17:28] these generators, it's each one of our [17:32] each one of our units, we call it a [17:35] rack. There's 72 chips inside. We we we [17:40] um we manufacture uh call it 8 million [17:43] of them this year, but but uh 72 of them [17:47] go into a rack. That rack weighs two [17:49] tons. [17:51] Uh it is uh $4 million, [17:54] has one a.5 million parts [17:57] and um it's the most expensive piece of [18:00] equipment in the world. And we [18:02] manufacture them like like um I guess it [18:05] would be like phones. I mean we crank [18:07] them out and they go into you know data [18:10] centers all over the world and yeah we [18:12] build these machines in volume. [18:16] They they are big devices. This is this [18:19] is how you do your weightlifting. [18:21] >> Yes, [18:21] >> I understand. [18:22] >> No volume discounts. [18:27] >> Okay. So, you laid out this picture of a [18:29] very exciting world [18:31] >> that we are in. We're in the middle of [18:33] this revolution. We're [18:35] >> you could say decades in. You could say [18:36] years in. [18:38] >> Certainly now in the the mainstream of [18:41] the intelligence revolution. [18:43] >> How do we participate? I'm sure [18:45] everybody here wants to participate in [18:47] this revolution. [18:48] >> Yeah, excellent question. [18:49] >> Let's start with big enterprises and [18:51] then let's get to individuals as well. [18:54] How do people join this movement? [18:56] >> And so now I gave you two mental models. [18:58] I'm going to give you one more mental [18:59] model. So one mental model there really, [19:02] you know, we could talk about this story [19:04] and hopefully we cover all four phases. [19:06] I just talked about what AI can do. I [19:09] talked about how AI is made and these [19:11] things are in these factories. These [19:12] factories are, you know, each gigawatt [19:15] is about $50 billion. So if you've ever [19:17] seen, it's the most expensive factory in [19:19] the world, but it also that one $50 [19:22] billion factory also generates 300400 [19:26] billion dollars in intelligence. And so [19:29] the the production value is incredible. [19:32] Okay, the return on investment is [19:34] extremely fast. And so these so that's [19:37] the factory part. The part that I'm [19:38] going to tell you now and this is very [19:40] important when you think about [19:41] investment is what does the industrial [19:43] layout look like for for AI and the the [19:46] the way to think about the industrial [19:48] view is think of it as a five layer [19:50] cake. Now I told you on the bottom is [19:53] energy. The bottom is remember I said [19:56] the dynamo. Okay. Now we of course [19:58] different different AC generators and [19:59] things like that power generation. And [20:01] so on the lowest layer is energy. This [20:03] is the single greatest opportunity in [20:07] several generations for the energy [20:09] industries to grow. The very first time [20:12] in probably I don't know a hundred years [20:15] since the energy grid in many countries [20:18] could be invested in. This is the best [20:20] opportunity to invest in sustainable [20:22] energy. If you care about sustainable [20:23] energy, nuclear, air, you know, or wind, [20:27] solar, you name it, whatever form, [20:29] hydrogen, whatever form, so long as it [20:31] produces energy, it's going to get [20:33] funded. And so that tells you something [20:35] about how great of a time this is [20:37] because we have a trillion dollars. Just [20:39] think this one year, this year alone, [20:41] we're going to put a trillion dollars [20:43] from the market into this entire five [20:45] layer cake I'm I'm about to describe to [20:47] you. So the first layer is energy. [20:48] That's the reason why Semens is doing so [20:50] well. That's Mitsubishi is doing [20:52] fantastically. GE Vernova, I mean [20:55] everybody. The first layer of the cake [20:57] is energy. The second layer of the cake [21:00] is chips and computers and networking [21:02] and switches and silicon photonics. Does [21:04] that make sense? It's all the computers. [21:06] The third layer of the cake, we call it [21:09] infrastructure. [21:10] land, power, shell, money, [21:16] data center operations, [21:18] every one of them in scarce supply [21:20] today. And so that's the that's the next [21:23] layer is infrastructure layer. And then [21:25] the layer that everybody sees that [21:27] everybody thinks is AI is the model [21:28] layer. Does it make sense? That's the [21:30] next layer. It sits on top of the [21:32] computers, the the cloud infrastructure. [21:34] And this is the greatest opportunity in [21:36] recent human in human history that I [21:38] know that so much marketdriven [21:41] investment is naturally coming into the [21:43] ecosystem. This is a great time to build [21:45] and so so now that's the model layer. [21:47] The model layer is uh open AI it's [21:50] anthropic but this is the part that you [21:53] can't overlook. This is very important. [21:55] So you have two companies that you know [21:57] of that you hear about. Um uh however [22:00] don't forget AI as I was explaining [22:02] earlier has learned the language [22:05] it learns the language and the meaning [22:07] the mean the language the meaning of [22:10] anything that is structural. So that [22:14] layer that what's really important is so [22:16] we hear we we talk about all the [22:17] language but don't forget um we learn [22:20] any you can learn anything with [22:21] structure and so let me give you an [22:23] example of something with structure. Uh [22:24] today when I walked into room I was [22:26] expecting a lot of people and it wasn't [22:29] unexpected the way you appeared. [22:31] >> Now if some of you were hanging off the [22:33] ceiling and floating in midair and some [22:36] of you you know the one human but body [22:39] parts in 17 different places um and and [22:42] I could see through some of you then [22:44] then it's hard to learn that because [22:46] every time it's different okay because [22:48] it's hard to learn quantum things. [22:51] However, things with structure we can [22:52] learn, right? People have eyes and so [22:55] you can learn these things. Okay? And so [22:57] 3D I I learned the laws of physics. Uh I [23:00] I sat down and notice I sat down with [23:03] confidence. I wasn't ex I wasn't 50% 53% [23:08] of the time I landed safely on the [23:10] chair. The other 47% of the time I went [23:12] right through it. [23:13] >> And so I can't trust it, but 100% time. [23:16] Do you guys understand? And so if if you [23:19] if things are predictable and are [23:22] predictable then there's structure you [23:23] can learn from it and you can learn the [23:24] meaning of it. Okay? And so we learned [23:26] the meaning of protein. We learned the [23:27] meaning. We're learning the meaning of [23:29] genes. Not just sequencing it, not not [23:31] just crisper editing it, but what is the [23:33] meaning of that gene? What is the [23:35] meaning of a cell? [23:37] >> Why does a cell do what the cell does? [23:39] What happens in two cells coming [23:40] together? And so this is no different [23:42] than imagine I learned the meaning of a [23:44] cell the way I learned the meaning of a [23:46] word. And what what is what happens when [23:48] I took take two words, put them [23:49] together? These two words activate each [23:51] other, turn into something else of [23:53] another meaning. Okay? And so from a [23:55] computer's perspective, it doesn't care [23:57] if it's a cell, a protein, a word, an [24:00] image, a car. Does that make sense? It's [24:03] just tokens. [24:05] >> And so we have to figure out as computer [24:07] scientists, we have to figure out how to [24:09] represent the world's information in all [24:12] these different ways so that the [24:14] computer can understand it. Understand [24:16] it, understand it, reason about it, come [24:18] up with a plan, generate an action. the [24:21] intelligence loop proteins the same way [24:23] cells the same way the human anatomy is [24:26] the same way it must be predictable it's [24:28] predictable because tomorrow morning I'm [24:30] largely the same it must be predictable [24:33] okay and so um we're learning all these [24:36] different my point is there are two [24:38] language models that you guys know about [24:39] but AI is a giant industry [24:43] the industry of everything else physical [24:45] is about $80 trillion [24:48] it is actually you know the most [24:50] important frontier, the one the parts [24:52] that we're not talking about. And then [24:54] on top of that, this model, this [24:56] technology then feeds into all the stuff [24:58] that Constantine gets to see these days, [25:00] which is all of these startups that are [25:03] coming up with revolutionary ideas in [25:05] financial services, in legal, in [25:08] accounting, in transportation, [25:11] logistics. Isn't that right? And so what [25:14] that layer above this last year, a [25:18] hundred billion dollars of venture [25:20] capital investment, the single largest [25:23] year of VC investment in the history of [25:25] humanity. [25:27] All of that money is going into that [25:29] fifth layer, the top layer, which is the [25:32] the layer that apply applications to [25:35] enhance human condition. And so there [25:38] are five layers when you think about AI [25:41] and you want to invest in this future [25:43] and I promise you this future is going [25:45] to be gigantic because two years ago [25:47] zero it's approximately we're about to [25:50] put $1 trillion in but that's 1 trillion [25:53] out of the [25:55] you know we're going to be putting in [25:56] probably the AI industry I'm going to [25:58] guess for a second probably something [26:00] along the lines of 20 trillion a year. [26:02] We're one trillion dollars in of a 20 [26:05] trillion dollars a year ecosystem [26:08] because the production of intelligence, [26:10] you just got to ask yourself how [26:12] important is intelligence and who needs [26:15] it and how much of it do you want? And [26:18] so those are kind of the basic questions [26:20] and all of that intelligence whether [26:22] it's for proteins or cars or robots or [26:25] language or math, science, whatever it [26:27] is, has to be generated by these [26:28] machines. And so this five layer cake [26:31] >> is the industrial version and I think [26:34] that's a good way to think about where [26:35] to invest [26:37] >> hugely. So so you've described what is a [26:40] multi-t trillion dollar opportunity to [26:43] become part of this revolution and that [26:45] includes the hardware and the [26:47] facilities. If it's $50 billion for a [26:50] gigawatt and there's 100 plus coming [26:52] online in the next several years, that [26:55] is trillions plus the application layer [26:57] where that is many many more trillions [26:59] plus plus plus and that means real jobs [27:03] for people doing the the hands in [27:06] building and then also [27:07] >> right now exactly and it's we have to we [27:10] have to really emphasize this and this [27:11] is very important to you. Um [27:14] there every country has a different [27:16] attitude about AI today. Would you guys [27:19] agree with that? Every country because [27:21] everybody's culture is a little [27:22] different. Okay. And and here's here's [27:25] my recommendation. [27:27] Um be careful with with the analogies [27:30] and the science fiction stories that [27:32] this is Terminator and and words like [27:35] singularity and and um ideas that that [27:38] somebody say in 20% chance this will be [27:41] the end of humanity as we know it. Okay, [27:44] those kind of those kind of [27:45] articulations of AI is just nonsense. [27:49] >> It is complete nonsense. Oh, we have no [27:51] idea how it works. This is so [27:54] mysterious. We don't even know how it [27:55] works. It might just get up out of its [27:57] seat and walk out tomorrow morning. [28:00] There's no question in my mind it's [28:01] computer and software. And there's no [28:04] question in my mind they know how it [28:06] works. And do you know how I know they [28:07] know how to works? Because every single [28:09] year apparently it's getting better. If [28:11] you don't know how something works, how [28:12] do you make it better? [28:14] I have no idea how it works. But I know [28:16] how to make it better. [28:19] That's nonsense. So why are they saying [28:22] these things? That's an interesting [28:23] question. However, don't let it scare [28:25] you. You must engage it. You may or may [28:28] not lose a job to an AI, but you will [28:31] absolutely lose a job to someone who [28:33] uses AI. Would you agree with that? [28:35] >> Yes. [28:35] >> Okay. So, let's not worry about the [28:37] things you're not sure about and focus [28:39] on the things that you are sure about. [28:42] I am absolutely certain I will lose my [28:45] job to someone who uses AI. So, before I [28:48] worry about AI, let's just go make sure [28:50] I use AI. [28:53] And so the part that now why is that why [28:55] is that common sense so important for me [28:58] to tell you? Because some of you have [29:00] children. [29:02] What are you advising them? Run away [29:06] or make sure whatever this technology is [29:08] that gives people superpowers that you [29:10] go make sure you use it. So I'm hoping [29:14] that we do two things. one, [29:18] we are and we're doing everything we can [29:21] to build this technology safely for the [29:23] world. I promise you so much computer [29:26] science, so much investment, so much [29:29] passion dedicated to making this [29:31] technology safe for everybody to use. [29:33] And I can prove it. Use Chad GB2 years [29:36] ago and use it again. [29:39] The amount of hallucination [29:41] completely reduced to almost nothing to [29:44] the point where it's producing knowledge [29:46] not only accurately [29:49] contextually relevant [29:52] relevant to the moment. And if it [29:54] doesn't know the answer, it does [29:55] research. And when it comes up with an [29:57] answer, it even questions itself before [29:59] it tells you an answer, it reflects on [30:01] it. and it comes up with two or three qu [30:04] two or three different answers and it [30:06] reflects on those before it produces the [30:08] answer for you. The amount of safety and [30:11] guard railing grounding of truth the [30:14] technology had advanced so fast to make [30:17] it safe. I am certain I can tell you [30:19] this completely [30:22] with fact. I prefer my car today than [30:26] the car that was a 100 years ago. [30:29] The technology is a lot better, but it's [30:31] a lot safer. [30:33] >> And it takes a lot of technology to be [30:34] invented in order for it to be safe. And [30:37] so I can tell you that it is our job, it [30:40] is the responsibility of the technology [30:41] industry is the responsibility [30:43] scientists and engineers for us to build [30:45] AI safely. Two, it is your [30:47] responsibility to make sure that you [30:50] tell the people that you love, whether [30:53] it's your family, your kids, your [30:55] grandkids, or the company you work for, [30:57] or the country that you're in, whatever [31:00] we do, [31:02] engage AI. If we think it's a [31:05] superpower, engage it. Because if we [31:08] don't engage it, somebody else will. [31:12] We're not going to lose our lives to AI. [31:13] We're going to lose our lives to [31:14] somebody who uses AI. [31:17] And so that's my Well, that that's too [31:21] serious. [31:22] >> That's the uh [31:23] >> that's too ser that was because he said [31:24] the word job. [31:27] >> So I've I've got a trigger and the [31:29] trigger is a bunch of people making [31:31] stuff up about jobs. [31:33] >> We put a trillion dollars into the [31:35] world's ecosystem this year, did we not? [31:37] What's it doing? Making jobs. [31:41] Right now, the energy sector, more jobs [31:44] than ever. The chip sector, more jobs [31:46] than ever. Infrastructure layer, more [31:48] jobs than ever. Everything from land, [31:50] power, shell, finances, AI model layer, [31:53] more jobs than ever. And we just said a [31:55] hundred billion dollars last year went [31:57] into the upper layer, more jobs than [31:59] ever. We're creating so many more jobs. [32:01] Now, somebody might say, well, what [32:03] about the traditional jobs? So, let me [32:05] give you the example. [32:07] You know that everybody's job and their [32:09] task is related not the same. [32:14] A job and the task you do in the job is [32:17] related not the same. So for example my [32:20] job is to be the CEO to lead the [32:22] company. Most of most of the time and I [32:24] spend a lot of it today most of my time [32:27] my task is [32:29] typing and talking. [32:31] >> And so you could say CEO equals typing [32:34] and talking. [32:37] Both of them AI does in a superhuman way [32:41] and I'm busier than ever. Then give you [32:44] that of course that's a really cute [32:45] example but let me give you the a more a [32:48] deep example and you can now apply it. [32:51] So 10 years ago more a slightly more [32:53] than that one of the world's leading [32:55] computer scientists wanted to warn [32:58] everybody about the power of AI. And so [33:01] he said and Constantine probably knows [33:03] who it is. He said the first job that AI [33:07] will destroy and eliminate and I advise [33:09] nobody goes into this field because this [33:12] field will be wiped out [33:14] >> radiology [33:14] >> is radiology. Computer vision is [33:17] superhuman [33:18] already 12 years ago. [33:20] >> Computer vision. A computer can [33:22] recognize images, detect anomalies with [33:26] superhuman capability. Never gets tired, [33:28] never miss a detail. 12 years ago it was [33:30] able to do that. and he predicted as a [33:32] result of that radiology is going to be [33:34] wiped out. Well, he was absolutely [33:37] right. [33:39] Radiology was completely [33:42] penetrated by computer vision. Computer [33:44] vision [33:46] proliferated through every single form [33:48] of radiology and every radiology stack [33:50] and every radiologist today is augmented [33:53] by computer vision. However, the [33:55] interesting thing is this. Radiology [33:57] demand went up. The number of [33:59] radiologists in the world went up. Why? [34:04] Audience participation please. Why? [34:11] >> I heard some of the things. It's all [34:13] true. It turns out radiology spends a [34:16] lot of time studying scans. [34:20] But the purpose of the radiology, the [34:22] purpose of the radiologist is to work [34:25] with doctors to diagnose disease. [34:29] >> To work with doctors to diagnose disease [34:32] and because it's now automated, they are [34:35] more productive. So two things happen. [34:39] More patients are admitted into the [34:42] hospital. [34:43] They do more scans. [34:46] The radiology department became more [34:48] profitable. [34:50] When they realized they were more [34:51] profitable and they were admitting more [34:54] patients, they hired more radiologists [34:58] >> so that they could admit more patients [35:01] so that they could make more money, take [35:03] care of more people because as it turns [35:05] out, there are a lot of people who are [35:07] suffering and they're waiting to get [35:09] into the hospital. So now [35:13] let's pretend for a second. [35:16] Do you appreciate the computer [35:18] scientists tell you it's going to be the [35:20] end of the world for radiologists? [35:25] My point is we have to be responsible [35:28] about what we make up because we could [35:30] have done harm. And it turns out the [35:33] number of people who want to be [35:34] radiologists after his speech because it [35:36] permeated through everything the number [35:38] of radiologists started to decline. [35:41] >> But we need more radiologists. [35:43] >> Now somebody recently said 90% of [35:45] software coding will be gone. And [35:47] therefore we don't need software [35:48] engineers. [35:49] >> Meanwhile, we're hiring more software [35:50] engineers than ever. [35:51] >> Y [35:52] >> and the reason for that is because a [35:53] software engineer's job is to solve [35:55] problems and dream up problems to solve [35:58] innovate. [35:59] I never hired somebody and said, "Hey, [36:01] guess what? You're a software engineer. [36:03] Listen, [36:05] here's a keyboard. Show me how many [36:08] words a second you can type." [36:11] Typing is not the job of a software [36:13] engineer. Coding is not their job. [36:16] Solving problems is their job. And so, I [36:18] just gave you two examples. Task versus [36:21] purpose. Does that make sense? [36:24] >> It turns out this example happens all [36:27] over the place. [36:28] But because we have such a contrived [36:31] such a naive understanding [36:34] that computer scientists could say [36:36] things like 50% of the jobs will be [36:38] gone. [36:41] Software coding is completely [36:43] irrelevant. Radiology is going to be [36:44] wiped out because we think about it from [36:46] the task perspective. We forgot the [36:48] purpose of the job. [36:50] >> There were radiologists before there [36:52] were workstations. There are going to be [36:54] radiologists after AI. There were soft [36:57] there were engineers before software [36:59] coding. I promise you that there will be [37:01] engineers after. Does that make sense? [37:03] And so that's the the way to think about [37:06] jobs. And um I've now covered two [37:09] things. One, if your country is not [37:12] investing in AI, there's a massive boom [37:15] of jobs you're missing out on. [37:18] If your country or your company is not [37:20] investing in AI, there's a level of [37:23] elevation of your people that you're [37:26] missing out on. AI is not going to [37:29] eliminate jobs. AI is going to elevate [37:32] your job. If I were a plumber today, [37:35] largely I get a job task sheet or a [37:39] schematic. [37:42] However, if I'm a plumber tomorrow, it [37:44] is very likely I'm a designer as well. M [37:47] >> does that make sense? Because you and I [37:49] both know [37:50] >> Yeah. [37:50] >> that we could just use AI to generate [37:53] these incredible designs of a kitchen. [37:56] If I'm a carpenter, [37:59] >> I can turn if I were a saleserson of [38:02] furniture. I'm going to be an interior [38:03] interior designer for sure. [38:06] >> And so I've elevated my craft. Went from [38:09] somebody who sells furniture to somebody [38:11] who could advise you on how beautiful [38:13] your home could be. I went from somebody [38:15] who's a carpenter, you expected me to [38:17] come and just, you know, put some wood [38:19] together and now I'm your home designer. [38:22] >> You I've elevated my craft. I've given [38:25] you so many examples, but that's my [38:27] point. I think the narrative about AI is [38:30] absolutely wrong. [38:32] >> And the goal is to scare everybody out [38:36] of it [38:38] >> so that some people could benefit from [38:39] it. But AI, as you know, is the greatest [38:44] force for eliminating the technology [38:46] divide in my entire career. [38:49] >> I spent 40 some odd years, my entire [38:51] life has been in computer design. I [38:53] spent 40 some odd years and this entire [38:56] time the technology we created became [38:59] more and more and more and more complex [39:01] and the number of people who could who [39:03] could program these computers as a [39:05] percentage of population declined. [39:08] >> Who in this room knows C++? [39:12] Come on, cut it out, you weirdos. [39:16] This row is just is a startup company. [39:18] Okay. And so so so this row Okay, so [39:22] we're we're looking at 2%. Okay, 2%. And [39:26] this is a very strange room. This is a [39:28] very strange room. And so 2% of society [39:30] knows C++. How many people know human? [39:35] Okay, more than 2%. And so and so [39:38] everybody now know everybody now can [39:40] program a computer. And yet in the past [39:42] only 2% can. We have closed the [39:44] technology divide. [39:46] >> We got to bring everybody with us. Does [39:47] that make sense? Okay. Anyways, that's [39:50] it on a Friday night. That's too [39:52] serious. That that that is extremely [39:54] optimistic and I agree. Uh and it's [39:57] great to hear from someone who is closer [39:59] to actual building of the actual [40:02] technology that powers everything than [40:04] anyone else in the world. Um, so Jensen, [40:07] you talked about a future where we move [40:09] from retrieval, this paradigm that we've [40:12] had our entire lives in to generation, [40:14] where everything is customized, [40:16] knowledge is customized for the [40:18] individual. A world where we have the [40:21] generation of intelligence paralleling [40:24] from the energy to the [40:26] telecommunications revolution to now the [40:28] intelligence revolution. You talked [40:30] about these languages that the computer [40:31] can speak, not just English or German, [40:35] but even protein. You talked about five [40:38] layers of participation, a abundant [40:40] opportunity to participate in this [40:42] revolution for everyone in this room and [40:44] everyone listening. And you talked about [40:46] how this transformation is going to be [40:48] something that has real consequences. [40:50] Real consequences that allow people to [40:53] move from just doing the task to [40:55] dreaming the problems and the solutions. [40:58] maybe even a life of purpose and a life [41:00] where we move from carpenters to [41:02] architects. Thank you, Jensen. Please [41:05] join me in thanking Jensen Swang, the [41:07] man who made this all happen. Thank you. [41:11] Thank you so much. You're awesome. [41:13] Really appreciate it. Thank you. That [41:14] was great.