Open source AI is essential for startups, traditional industries, higher education, and research — damaging it with policy would break the innovation flywheel.

(00:00:06) Jensen, thanks so much for joining us today. (00:00:08) So great to have you guys. (00:00:10) What an amazing year. (00:00:11) What a year. (00:00:11) Happy Hanukkah. (00:00:12) Merry Christmas. (00:00:14) Happy New Year coming up. (00:00:15) Happy holidays. (00:00:16) So with everything that's happened in 2025 and being in the middle of the vortex with it, what do you reflect on and say like this surprised you most or this is the biggest change? (00:00:28) Let's see, there's some things that didn't surprise me. (00:00:30) Like, for example, the scaling laws didn't surprise me because we already knew about that. (00:00:34) The technology advancement didn't surprise me. (00:00:37) I was pleased with the improvements of grounding. (00:00:40) I was pleased with the improvements of reasoning. (00:00:42) I was pleased with the connection of all of the models to search. (00:00:49) I'm pleased that it (00:00:50) that there are now routers that are in front of these models so that it could, depending on the confidence of the answers, go off and do necessary research and just generally improve the quality and the accuracy of answers. (00:01:05) I'm hugely proud of that. (00:01:07) I think the whole industry addressed one of the biggest skeptical (00:01:12) responses of AI, which is hallucination and generating gibberish and all of that stuff. (00:01:18) I thought that this year, the whole industry, everything from every field, from language to vision to robotics to self-driving cars, the application of reasoning and the grounding of the answers, (00:01:36) Big, big leaps, would you guys say, this year? (00:01:40) I mean, things like open evidence do for medical information where doctors are not really using that as a trusted resource. (00:01:45) Like Harvey for legal, you're really starting to see AI emerge as one of these things that's become a trusted tool or counterparty for experts to actually be able to do what they do much better. (00:01:55) That's right. (00:01:56) And so in a lot of ways, I was expecting it, but I'm still pleased by it. (00:02:00) I'm proud of it. (00:02:01) I'm proud of all of the industry's work in this area. (00:02:03) I'm really pleased. (00:02:04) And (00:02:06) and probably a little bit surprised, in fact, that token generation rate for inference, especially reasoning tokens, are growing so fast, several exponentials at the same times, it seems. (00:02:20) And I'm so pleased that these tokens are now profitable. (00:02:25) that people are generating, I heard somebody, heard today that OpenEvidence, speaking of them, 90% gross margins. (00:02:33) I mean, those are very profitable tokens. (00:02:35) And so they're obviously doing very profitable or very valuable work. (00:02:39) Cursor, their margins are great. (00:02:41) Claude's margins are great. (00:02:43) For the enterprise use of OpenAI, their margins are great. (00:02:46) So anyways, it's really terrific to see that we're now generating tokens. (00:02:52) that are sufficiently good, so good in value that people are willing to pay good money for. (00:02:57) And so I think these are really great grounding for the year. (00:03:00) I mean, some of the things that, the narrative that, of course, the conversation with China really, really occupied a lot of my time this year, geopolitics. (00:03:12) the importance of technology in each one of the countries. (00:03:14) I spent more time traveling around the world this year than just by any time in the history, all of my life combined. (00:03:21) My average elevation this year is probably about 17,000 feet. (00:03:24) So it's nice to be here on the ground with you guys. (00:03:28) And so I think geopolitics, the importance of AI to all the nations, all worth talking about later. (00:03:35) Of course, I spent a lot of time on expert control and making sure that our strategy is nuanced. (00:03:41) really grounded and promotes national security, but recognizing the importance of various facets of national security. (00:03:51) A lot of conversations around that. (00:03:54) of course, of course, lots of conversation about jobs, the impact of AI, energy, labor shortage. (00:04:03) I mean, boy, we covered everything, did we not? (00:04:04) That's a lot, yeah. (00:04:05) Everything was AI. (00:04:06) Everything was AI. (00:04:08) Yeah, it was incredible. (00:04:09) Yeah, AI was definitely the center of the storm for like every one of those themes. (00:04:12) Maybe one we can start with actually is jobs, because they're jobs and employment. (00:04:16) Because when I look at the traditional AI community, even before things were scaling and even before AI was really working, there was a strong sort of doomsday component. (00:04:26) in the people working on AI, oddly enough. (00:04:28) The people who are most trying to push the field forward were often the people who are most pessimistic, which is very odd. (00:04:32) Why would you do both at once? (00:04:34) And I feel like that narrative has taken over some subset of media or some set of other things, despite all the things that we think are very positive about what AI has done. (00:04:41) That's going to help with healthcare, with education, with productivity, with all these other areas. (00:04:46) And in general, whenever we have a technology shift, you have a shift in terms of the jobs that are important, but you still have more jobs. (00:04:52) That's right. (00:04:52) Could you talk about how you think about employment and jobs and sort of what people are saying and what you think the real narrative is there? (00:04:57) Maybe what I'll do is I'll ground it on three points in space, three points in time. (00:05:03) Now, maybe very near future, and then some point out in the distance. (00:05:12) And maybe some counter narratives. (00:05:15) something else to think about with respect to jobs. (00:05:18) In the near term, one of the most important things is that AI is not just, AI is software, but it's not pre-recorded software, as you know. (00:05:28) For example, Excel was written by several 100 engineers. (00:05:32) They compiled it. (00:05:33) It's pre-recorded, and then they distribute it as is for several years. (00:05:37) In the case of AI, because it takes into the context what you asked of it, (00:05:43) what's happening in the world, right? (00:05:44) Contextual information. (00:05:46) It generates every single token for the first time, every time, which means every time you use the software in everything that we do, AI is being generated for the first time ever. (00:05:58) Just like intelligence. (00:05:59) Our conversation today relies on some ground truth and some knowledge, but it's every single word is being generated for the first time here. (00:06:08) The thing that's really (00:06:09) really quite unique about AI is that it needs these computers to generate these tokens every single time. (00:06:16) I call them AI factories because it's producing tokens that will be used all over the world. (00:06:22) Now, some people would say it's also part of infrastructure. (00:06:25) The reason why it's infrastructure is because obviously it affects every single application. (00:06:29) It's used in every single company. (00:06:32) It's used in every single industry. (00:06:33) It's used in every single country. (00:06:34) Therefore, it's part infrastructure like energy and internet. (00:06:38) Now, because of that, (00:06:39) And the amount of computers that's necessary to generate these tokens, and it's never happened before, and because we need these factories, three new industries have emerged. (00:06:47) Number one, well, three new type of plants have to be created. (00:06:50) Number one, we have to build a lot more chip plants. (00:06:54) TSMC is building, right, SK Hynix, building a lot more plants. (00:06:58) And so we need more chip plants. (00:07:00) We need more (00:07:01) Computer plants. (00:07:02) These computers are very different. (00:07:04) These are supercomputers that the world's never seen before, right? (00:07:07) Grace Blackwell looks like a very different type of computer than anything that's ever been made. (00:07:12) And entire rack is 1 GPU. (00:07:15) And so we need new supercomputer plants. (00:07:17) And then we need new AI factories. (00:07:20) These 3 plants are currently being built in the United States (00:07:24) at very large scale, quite broadly all over the United States for the very first time. (00:07:29) The number of construction workers, plumbers, electricians, technicians, network engineers, you know, right, the number of the skilled labor that's necessary to support this new industry in the near term, it'll be enormous. (00:07:43) Let's just face it. (00:07:45) I'm so excited to hear that electricians are seeing their (00:07:48) Paychecks double. (00:07:49) They're being paid to travels. (00:07:52) Like us, we go on business trips. (00:07:53) They're going on business trips. (00:07:55) And so it's really terrific to see that these three industries are now three types of plants, factories, are just creating so much jobs. (00:08:05) The next part is the near-term impact of AI on jobs. (00:08:12) And one of my favorites is, I love Jeff Hinton. (00:08:16) He said, (00:08:18) some five, six, seven years ago, that in five years' time, AI will completely revolutionize radiology, that every single radiology application will be powered by AI, and that radiologists will no longer be needed. (00:08:37) And that he would advise the first profession not to go into is radiology. (00:08:42) And he's absolutely right. (00:08:45) 100% of radiology applications are now AI powered. (00:08:48) That's completely true. (00:08:49) And in some eight years time, it is now completely pervaded radiology. (00:08:57) However, what's interesting is that the number of radiologists increased. (00:09:01) And so now the question is why? (00:09:03) And this is where the difference between task versus purpose of a job. (00:09:09) A job has tasks and has purpose. (00:09:12) And in the case of a radiologist, the task is to study scans, but the purpose is to diagnose disease. (00:09:21) And to do research. (00:09:22) And that, exactly, and they're doing research. (00:09:24) And so in the case, in their case, the fact that they're able to study more scans more deeply, they're able to request more scans, do a better job diagnosing disease, the hospital's more productive, they can have more patients, (00:09:41) which allows them to make more money, which allows them to want to hire more radiologists. (00:09:46) And so the question is, what is the purpose of the job versus what is the task that you do in your job? (00:09:52) And as I spend most of my day typing. (00:09:57) That's my task, but my purpose is obviously not typing. (00:10:00) And so the fact that somebody could use AI to automate a lot of my typing, and I really appreciate that, and it helps a lot, it hasn't really made me, if you will, (00:10:10) less busy in a lot of ways, I've become more busy because I'm able to do more work. (00:10:14) So I think that that's the second part to consider is the task versus the purpose of the job. (00:10:19) This example really strikes home because my sister-in-law, Erin, actually leads nuclear medicine at Stanford. (00:10:25) So she's in radiology. (00:10:27) And with all the technology advancements that are coming, (00:10:30) These doctors really welcome it and they are working 20 hours a day trying to do more research and serve more patients. (00:10:36) Exactly. (00:10:37) And I think one thing that is often missed beyond the sort of diversity of jobs being created by this investment in infrastructure is actually how much latent demand there is for different goods. (00:10:51) that we need in society, like better healthcare. (00:10:53) I don't think anybody feels like, you know what, we have reached the tip top, mountain top of like what American healthcare or global healthcare could be. (00:11:03) And the more we can make these people productive, the more demand there will be. (00:11:07) That's exactly right. (00:11:08) If Nvidia was more productive, it doesn't result in layoffs. (00:11:13) It results in us doing more (00:11:15) more things. (00:11:16) I'm at your new hire class today. (00:11:17) You seem to be hiring every week anyway. (00:11:20) That's exactly right. (00:11:22) The more productive we are, the more ideas we can explore, the more growth as a result, the more profitable we become, which allows us to pursue more ideas. (00:11:33) And so I think you're absolutely right that if (00:11:35) If the job, if your life, if the world, the problems, is literally already specified and there's no other problem to solve, then productivity would actually reduce the economy. (00:11:48) But it's clearly going to increase the economy. (00:11:51) I think the next part that I would consider is, people say, gosh, all of these robots that we're talking about, it's going to take away jobs. (00:12:00) As we (00:12:00) We know very clearly. (00:12:02) We don't have enough factory workers. (00:12:04) Our economy is actually limited by the number of factory workers we have. (00:12:07) Most people are having a very hard time retaining their workers. (00:12:14) We also know that the number of truck drivers in the world is severely short. (00:12:19) And the reason for that is people don't want those jobs where you have to travel across the country and live in different parts of the world, different parts of the country, you know, every single night. (00:12:27) And so people want to stay in their town, stay with their families. (00:12:30) So I think the first part is that having robotic systems is going to allow us to cover the labor shortage gap, which is really, really severe and getting worse because of aging population. (00:12:44) This is not only in the United States, it's all over the world, as you guys know. (00:12:48) And so we're going to cover the labor shortage. (00:12:52) But the second part that people forget, (00:12:54) And there are shortages as well in other places that people talk about AI being relevant. (00:12:59) Accounting would be an example where there's shortages there. (00:13:01) Nursing is another example. (00:13:03) So you can go through multiple other industries and say, okay, there's gaps. (00:13:06) And AI is trying to help fill those gaps. (00:13:09) That's exactly right. (00:13:10) And so automation is going to help us increase and solve the labor gap. (00:13:16) Now, people also don't remember that when we have cars, (00:13:22) we need mechanics to take care of our cars. (00:13:25) And if you look at the robo taxis that are even on the streets today, it's taken 10 years for that to happen. (00:13:32) Look at all the maintenance crews and all of the various hubs that they're in where you have to take care of these robo taxis. (00:13:40) And just imagine we have a billion robots. (00:13:44) It's going to be the largest repair industry on the planet. (00:13:47) So I think a lot of people don't (00:13:48) They just have to think through. (00:13:50) And this is the part where you said, when we create this type of automation, we create this other job. (00:13:55) Right now, look at AI, it's creating so many jobs. (00:13:59) The AI industry is creating a boom of jobs. (00:14:02) I think one of the core challenges here is it's very easy to draw a straight line of extrapolation from like, oh, there are tools that help lawyers be more productive. (00:14:13) It's going to replace the lawyers. (00:14:15) But it's actually, it takes like a step of incremental reasoning to say, (00:14:20) there's a sucking sound in the economy for everything in AI infrastructure. (00:14:23) There's actually a sucking sound toward all of this demand that is latent in the places where we have gaps, where I think a lot of policymakers have focused on, you know, we can't replace or reduce what we have, when it's really, there's far more demand in what we actually are not. (00:14:39) And in the case of a lawyer, what's the purpose of the lawyer versus the task of the lawyer? (00:14:45) Reading a contract, writing a contract is not the purpose of the lawyer. (00:14:50) The purpose of the lawyer is to help you resolve conflict. (00:14:54) And that's more than reading a contract. (00:14:56) It's more than writing a contract. (00:14:58) The purpose is to protect you. (00:15:00) That's more than reading a contract. (00:15:02) It's more than writing a contract. (00:15:04) And so I think just, it's really, really important to go back to what is the purpose of the job versus the task that we use. (00:15:12) to perform that job. (00:15:13) That changes over time. (00:15:14) the other big theme of the year that you mentioned that I think is really important to touch upon is both China is sort of in the rise of Chinese open source in particular, where some of the highest scoring models against benchmarks now are Chinese models on the open source side. (00:15:27) On the closer side, it's still a lot of the US models, but things like Quinn, Deepseek, et cetera, are doing very well. (00:15:33) You've long been a proponent for open source in general. (00:15:35) Could you share your views about both China emerging (00:15:39) for AI, for open source, and what the US should be doing in terms of both open source as well as its own industries. (00:15:44) When you think about these complicated, interconnected, dependent networks of problems, this big goop of a mesh of problems, it's always good to go back and find a framework for what it is that we're talking about. (00:16:01) In the case of AI, what is AI? (00:16:06) Well, of course, (00:16:07) The technology of AI and the capabilities of AI is about automation. (00:16:14) It's about automation of intelligence for the very first time. (00:16:16) And you could combine it with mechatronics technology to embody that mechatronics and make it perform tasks. (00:16:27) So that's what's AI, automation. (00:16:30) But what (00:16:31) What is the stack that makes AI possible? (00:16:33) What's the technology stack, the functional stack? (00:16:35) And of course, the easiest way to think about that is it's kind of like a five-layer cake, which is at the lowest level is energy. (00:16:45) It transforms energy to the output that I just described. (00:16:48) The next layer is chips. (00:16:50) The next layer is infrastructure. (00:16:51) And that infrastructure is both hardware, software, right? (00:16:54) This is where land-powered shell, this is where construction is. (00:16:58) data centers are, the software stack, orchestrating. (00:17:02) So it's software and hardware. (00:17:04) The layer above that is where everybody thinks about, which is AI, which is the models. (00:17:10) We know this, but it's really helpful to understand that AI is a system of models. (00:17:15) And AI is a technology that understands information. (00:17:22) And there's human information. (00:17:24) And so we oftentimes (00:17:26) think about AI as a chatbot. (00:17:28) But remember, there's biological information, there's chemical information, there's physical information, information of all kinds. (00:17:35) There's financial information, there's healthcare information, there's information of all modalities, all kinds. (00:17:42) AI is really, really broad. (00:17:44) And of course, human language is at the foundation of many things, but it's not the essence of everything because as you know, biology, (00:17:53) Molecules don't understand English. (00:17:55) They understand something else, right? (00:17:57) Proteins don't understand English. (00:17:58) They understand something else. (00:17:59) I think the next layer, the important thing is, that's where the AI models are, but there's a whole, AI is very, very diverse. (00:18:07) And then the layer above that is applications. (00:18:09) And it depends on the industry. (00:18:11) And you already mentioned Open Evidence, you mentioned Harvey, there's Cursor, there's all kinds of, right? (00:18:16) There's all kinds of applications. (00:18:17) Full self-driving is really an application, an AI application that is embodied into a mechanical car. (00:18:23) And Figure is a AI application that has been embodied into a mechanical human. (00:18:28) And so you got all these different applications. (00:18:32) Well, this five-layer stack is one way of thinking about it. (00:18:36) And then the next way of thinking about it, I just mentioned, is AI is really diverse. (00:18:39) When you now have this framework of what the technology capabilities are, how to build the technology and how diverse it is, then you can come back and think about, okay, (00:18:52) Let's ask the question, how important is open source? (00:18:55) Well, without open source, today, of course, the Frontier models, the leading labs have chosen to use a closed source application approach, which is just fine. (00:19:09) what people decide to do with their business models is really in the final analysis. (00:19:13) It's their business. (00:19:14) And they have to calculate what is the best way for them to get the return on investment so that they could scale up and make better advances. (00:19:22) However they made that calculus is fantastic. (00:19:25) On the other hand, without open source, as you know, startups would be challenged. (00:19:31) Companies that are in different industries, whether it's manufacturing or transportation or (00:19:39) It could be in healthcare. (00:19:40) Without open source today, all of that AI work would be suffocated. (00:19:46) And so they just need to have something that's pre-trained. (00:19:48) They need to have some fundamental technology about reasoning. (00:19:52) From that, they could all adapt, fine-tune, train their AI models into exactly the domain and application they want. (00:20:01) And so what people really, really miss is just the incredible pervasiveness and (00:20:08) the importance of open source to all of these industries, large companies without open source, some of the 100 year old companies that I work with in industrial spaces and healthcare spaces, they would be suffocated. (00:20:22) They wouldn't be able to do that. (00:20:23) Open source at this point is driving all of our data centers. (00:20:25) It's driving a big chunk of telephony in the world in terms of Android or other devices. (00:20:30) It's driving, I know you're putting a lot of the industrial applications. (00:20:32) So it's already pervasive. (00:20:34) And I think the big question is... (00:20:35) Open source without open source, higher ed. (00:20:38) Education research startups. (00:20:40) I mean, the list goes on. (00:20:41) And so we talk all day long about the tip, the most visible part of that, the part that's most newsworthy maybe. (00:20:52) But underneath that is such an important space of open source AI. (00:20:58) And whatever we decide to do with policies, do not damage that innovation flywheel. (00:21:05) So I spent a lot of time (00:21:07) educating policy makers to help them understand whatever you decide, whatever you do, don't forget open source. (00:21:15) Whatever you decide, whatever you do, don't forget biology. (00:21:20) I think the counter narrative here that is worth addressing is that essentially like (00:21:27) there should be a monolithic vertical player and monolithic asset in the one model that does it all, and that we can't give away that crown jewel to other countries or non-American companies. (00:21:39) And your argument is like, we actually need this huge diversity of AI applications and the American advantage is actually, or any sovereign advantages in the whole stack, right? (00:21:50) The capability to deliver any piece of it. (00:21:52) I guess someday we will have got AI. (00:21:56) When is that day? (00:21:57) But that someday, that someday is probably on biblical scales, I think galactic scales. (00:22:04) I think it's not helpful to go from where we are today to God AI. (00:22:10) And I don't think any company practically believes they're anywhere near God AI. (00:22:17) And nor do I see any researchers having any reasonable ability to create God AI. (00:22:25) the ability to understand human language and genome language and molecular language and protein language and amino acid language and physics language, all supremely well. (00:22:36) That God, AI just doesn't exist. (00:22:39) And yet we have a lot of industries that need AI. (00:22:44) AI is, if you will, at the simplistic level, it's just the next computer industry. (00:22:51) And give me an example of a company, an industry, (00:22:55) a nation who doesn't need computers. (00:22:58) And we all don't have to wait around for God AI for us to advance, right? (00:23:02) So God AI is not showing up next week. (00:23:04) I'm fairly certain of that. (00:23:06) Okay, that's great. (00:23:06) And God AI is not going to show up next year, but the whole world needs to move forward next week, next year, next decade. (00:23:13) I think that the idea of a monolithic, gigantic company, country, nation state that has God AI, (00:23:24) It's just- It's unhelpful. (00:23:26) It's unhelpful. (00:23:27) It's too extreme. (00:23:28) Then in fact, if you want to take it to that level, then we ought to just all stop everything. (00:23:34) What's the point of having even governments? (00:23:36) I mean, why are they doing policies? (00:23:39) God AI is going to be smart enough to avert, work around any policy. (00:23:43) And so what's the point? (00:23:44) And so I think that we ought to bring things back to the ground, ground level, and start thinking about things practically. (00:23:52) and use common sense. (00:23:55) This seems to be like a big theme in general in terms of this conversation where there's been a lot.