WEBVTT - Nvidia, Siemens CEOs Talk Building Industrial AI Operating System 0:00:00.040 --> 0:00:07.000 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:07.840 --> 0:00:09.920 One of the biggest stories that we have been following 0:00:10.000 --> 0:00:12.000 in the last twenty four hours, though, of course, has 0:00:12.000 --> 0:00:15.080 been the important keynote address from Jensen Wong at the 0:00:15.120 --> 0:00:18.560 Consumer Electronics Show. It has been feeding a number of 0:00:18.560 --> 0:00:21.480 storylines and moving the stock ever since that took place 0:00:21.520 --> 0:00:24.639 in Las Vegas yesterday. And joining us right now for 0:00:24.760 --> 0:00:27.320 a very special conversation, We've got good news for you. 0:00:27.400 --> 0:00:31.600 It's Bloomberg's Ed Ludlow with the CEO of Nvidia, Jensen Wong, 0:00:31.600 --> 0:00:34.200 as well as the president's CEO of Siemens, Roland Bush. 0:00:34.200 --> 0:00:37.320 They're at the cees right now and we want to 0:00:37.360 --> 0:00:39.600 hand things out to Las Vegas. Ed Ludlow, you have 0:00:39.680 --> 0:00:40.360 the con. 0:00:42.840 --> 0:00:43.520 Thank you, Joe. 0:00:43.560 --> 0:00:46.120 Over one hundred and seventy five years, Siemens has been 0:00:46.159 --> 0:00:49.400 at the forefront, if i'd say, several industrial revolutions. In 0:00:49.520 --> 0:00:54.160 Vidia is at the forefront of the latest AI industrial revolution, 0:00:54.280 --> 0:00:57.600 and it's a pleasure to have you both here with us. Gentlemen, Roland, 0:00:57.720 --> 0:00:59.959 I want to try and understand how real this is. 0:01:00.320 --> 0:01:03.800 You call it an industrial AI operating system. We're going 0:01:03.840 --> 0:01:06.080 to talk about how you're joining forces on the software 0:01:06.080 --> 0:01:08.760 side and the hardware side. But I think the most 0:01:08.800 --> 0:01:11.400 useful place to start would be, could you outline a 0:01:11.520 --> 0:01:15.479 timeline for when Semens goes beyond its own footprint around 0:01:15.520 --> 0:01:19.000 the world using this operating system to customers actually doing 0:01:19.520 --> 0:01:21.679 things at scale, because I think when you're on stage, 0:01:21.720 --> 0:01:23.119 that's the kind of sense I was getting. 0:01:23.280 --> 0:01:24.880 You want to accelerate to scale. 0:01:25.400 --> 0:01:28.720 Well, I mean this technology is already in place and 0:01:28.800 --> 0:01:29.360 it's working. 0:01:29.600 --> 0:01:31.280 I mean we see there are many examples, and it 0:01:31.360 --> 0:01:35.480 brought some on stage today where you have customers starting 0:01:35.480 --> 0:01:38.160 with a digital twin of a product. They want to 0:01:38.200 --> 0:01:41.280 manufacture a digital train of the manufacturing side and bring 0:01:41.319 --> 0:01:43.759 that all together and not before that you. 0:01:43.720 --> 0:01:45.720 Have it optimized them and build in the real world. 0:01:46.280 --> 0:01:49.240 And AI is already working on the shop for they 0:01:49.240 --> 0:01:52.280 call it machine learning. But the point is with the 0:01:52.360 --> 0:01:54.920 new models, you can bring that to the next level. 0:01:55.120 --> 0:01:58.200 That means you go away from giving an advice to 0:01:58.240 --> 0:02:01.560 somebody with technology, but really act on your behalf. This 0:02:01.720 --> 0:02:05.640 is when things go more autonomous or adaptive, and you 0:02:05.720 --> 0:02:06.960 see that already starting. 0:02:07.400 --> 0:02:09.760 The big thing is how can we scale it? 0:02:09.960 --> 0:02:12.800 Because it requires a lot of skills from our customers, 0:02:12.840 --> 0:02:15.359 a lot of technology, and it's still not that easy 0:02:15.400 --> 0:02:16.920 to implement the working on it to. 0:02:16.919 --> 0:02:19.760 Make it easy to deploy and easy to use, but 0:02:19.960 --> 0:02:23.280 you see picking up momentum in each stuff we do, 0:02:23.560 --> 0:02:26.239 and it's also great examples as a shipbuilding is a 0:02:26.360 --> 0:02:29.680 short everlenge, but even start ups using our technology like 0:02:29.800 --> 0:02:30.840 commentalst Fusion. 0:02:30.600 --> 0:02:35.120 Systems Jensen, we've discussed the five layer cake often. You 0:02:35.160 --> 0:02:37.680 know within video it started with the GPUs, but it's 0:02:37.720 --> 0:02:42.480 now software, the element of simulation. On stage, you talked 0:02:42.480 --> 0:02:46.760 about integrating the software side, in particular into DA Again, 0:02:46.800 --> 0:02:49.000 a similar question for you, more of a timeline. What 0:02:49.160 --> 0:02:51.040 is it you think you'll do first in how you 0:02:51.040 --> 0:02:52.080 guys work closer together. 0:02:52.240 --> 0:02:55.000 The first thing, let me just say very quickly, we're 0:02:55.000 --> 0:02:57.919 announcing a big partnership between us. We've known each other 0:02:57.919 --> 0:03:00.120 for a long time, but this is a partnership were 0:03:00.120 --> 0:03:01.480 announcing is really a big deal. 0:03:02.240 --> 0:03:02.480 One. 0:03:02.800 --> 0:03:08.560 We're accelerating their EDA software, We're accelerating their simulation software. 0:03:08.919 --> 0:03:14.280 We're integrating AI technology, physical AI and agentic AI into 0:03:14.400 --> 0:03:19.040 their team center and their factory automation operating system, and 0:03:19.080 --> 0:03:22.920 so we're working together across this entire spectrum. When we 0:03:22.960 --> 0:03:25.480 accelerate the software, then we'll get to use it to 0:03:25.520 --> 0:03:29.600 design our chips and systems. When we accelerate their simulation software, 0:03:29.720 --> 0:03:32.560 we'll use it in our AI factories to simulate the 0:03:32.560 --> 0:03:38.760 thermal properties of our AI factories. When we integrate our 0:03:38.800 --> 0:03:44.400 automation and agentic systems into their AI industrial operating system, 0:03:44.680 --> 0:03:47.320 we can then use it in our factory floors with 0:03:47.440 --> 0:03:51.040 our partners, for example, fox Con And so we're working 0:03:51.120 --> 0:03:53.280 across this entire spectrum together and we're going to put 0:03:53.280 --> 0:03:56.200 the technology to use basically as soon as we can. 0:03:56.320 --> 0:03:57.760 What's the net effect for you, Jensen? 0:03:57.920 --> 0:04:02.400 Is it improves efficient capital allocation. I know that might 0:04:02.440 --> 0:04:04.600 sound a bit dry, but actually, right now, that's the 0:04:04.680 --> 0:04:08.080 answer everyone's searching for. How is this investment in AI 0:04:08.560 --> 0:04:11.080 and use of the technology actually changed things in the 0:04:11.120 --> 0:04:11.840 real world for me? 0:04:12.120 --> 0:04:13.840 Announced yesterday Vera Rubin. 0:04:14.080 --> 0:04:18.400 It takes six different chips to integrate into this incredible 0:04:18.400 --> 0:04:22.240 system called Vera Rubin. And when you're done, each one 0:04:22.279 --> 0:04:26.400 of these Vera Rubin GPUs is two hundred and forty 0:04:26.760 --> 0:04:31.120 thousand wants and it is ten times more energy efficient 0:04:31.120 --> 0:04:34.080 than a last generation. It is ten times more cost 0:04:34.080 --> 0:04:37.200 efficient than a last generation but it's still the technology 0:04:37.240 --> 0:04:42.640 is insanely complicated. One hundred fifteen thousand engineering years came 0:04:42.680 --> 0:04:45.920 together to build this system. And so when we accelerate 0:04:46.080 --> 0:04:49.200 EDA tools, when we accelerate simulation tools, and when we 0:04:49.400 --> 0:04:53.920 can eventually and I'm hoping very soon design entire vera 0:04:54.000 --> 0:04:58.960 Ruben systems inside a Semen's digital twin, the chance the 0:04:59.080 --> 0:05:02.880 ability for us to create much much more complex systems 0:05:02.920 --> 0:05:06.039 will scale. We'll do it much more efficiently. And so 0:05:06.120 --> 0:05:08.520 this is really about being able to do the impossible, 0:05:08.920 --> 0:05:12.120 and being able to do it impossible the impossible right 0:05:12.320 --> 0:05:13.120 the first time. 0:05:13.640 --> 0:05:18.839 And and and and once they then realize that AI 0:05:19.080 --> 0:05:22.760 creates real world impact, this is where it really deploys 0:05:22.800 --> 0:05:23.880 the full power. 0:05:23.720 --> 0:05:25.239 And also the economic power. 0:05:25.800 --> 0:05:27.560 It's not only in the data centers are any eye 0:05:27.600 --> 0:05:29.880 factories which we see, but also on the edge because 0:05:29.960 --> 0:05:32.640 once you start influencing with low latency, you bring this 0:05:32.760 --> 0:05:35.599 technology to the edge. This is a huge potential for 0:05:35.680 --> 0:05:39.760 our customers to to deploy this technology. And this includes 0:05:39.800 --> 0:05:42.600 of course in pluts hardware where we come from from 0:05:42.640 --> 0:05:43.080 the JIPS. 0:05:43.839 --> 0:05:45.240 It goes into the controllers. 0:05:45.400 --> 0:05:49.400 Some of our controllers run on GPUs and then it goes. 0:05:49.320 --> 0:05:51.920 All the way to the industrial PC exactly. 0:05:52.240 --> 0:05:57.480 And these are aids. We super charge it and they 0:05:57.520 --> 0:06:00.599 can now run algorithms train than the cloud. 0:06:00.640 --> 0:06:02.600 They can run it on the shop floor and. 0:06:02.600 --> 0:06:05.680 Do all that trick what we talked about it in 0:06:05.760 --> 0:06:09.400 retime optimization and running in a planned and that makes 0:06:09.400 --> 0:06:10.040 a huge difference. 0:06:10.040 --> 0:06:12.839 So it de drives economy. The question that's being searched 0:06:12.839 --> 0:06:14.640 for is how is this going to manifest in. 0:06:14.560 --> 0:06:15.039 The real world. 0:06:15.080 --> 0:06:17.400 You know, the emphasis that this is cs I think 0:06:17.520 --> 0:06:21.320 is physical AI is not the manifestation of the final 0:06:21.360 --> 0:06:23.839 stage of physical AI, just one giant robot that you 0:06:23.880 --> 0:06:27.120 guys call a factory In the manufacturing context, is that 0:06:27.160 --> 0:06:29.920 where you're seeing demand from actual customers, you know that 0:06:30.080 --> 0:06:32.880 they need a factory that is automated one hundred percent 0:06:33.320 --> 0:06:35.159 and genuinely autonomous in some degree. 0:06:35.480 --> 0:06:37.880 Number one is if you want to build a factory, 0:06:38.080 --> 0:06:41.800 and very often you're missing out on labor, and we 0:06:41.920 --> 0:06:44.880 talk about on skilled labor, it's hard to find Number one. 0:06:44.920 --> 0:06:49.320 Number two is once you build it autonomous and automated, 0:06:49.600 --> 0:06:51.880 then you have a much much higher yield, but you 0:06:52.160 --> 0:06:54.560 can generate you use lesser energy. By the way, at 0:06:54.600 --> 0:06:56.840 the same time, before you optimize it in a way. 0:06:57.160 --> 0:06:59.760 So therefore there's a lot of benefits. And if you 0:06:59.800 --> 0:07:03.080 want to talking about the United States up manufacturing the 0:07:03.120 --> 0:07:06.440 United States, you need to go as digital and as 0:07:06.560 --> 0:07:09.400 automated and AI super changed as possible. 0:07:10.120 --> 0:07:16.320 A factory is robotic and it is orchestrating robots that 0:07:16.360 --> 0:07:20.000 are building systems that are also robotic, like for example, 0:07:20.040 --> 0:07:21.520 self draging cars a robotic system. 0:07:22.240 --> 0:07:24.480 And the reason why it's. 0:07:24.320 --> 0:07:27.160 So hard to deploy robots today is because it's hard 0:07:27.160 --> 0:07:31.960 to program these robotic systems. The software expertise that necessary 0:07:32.000 --> 0:07:34.440 to customization necessary is really intense. 0:07:34.480 --> 0:07:36.119 It's just too much. 0:07:36.760 --> 0:07:39.200 And so the fact that we could now apply artificial 0:07:39.240 --> 0:07:44.520 intelligence physical AI technology to these robotic systems make them 0:07:44.680 --> 0:07:49.080 easier to teach. You show it a few demonstrations, and 0:07:49.080 --> 0:07:50.560 the AI learned it by itself. 0:07:50.960 --> 0:07:52.880 You think, would you guys are, say Jenson, that you've 0:07:52.880 --> 0:07:54.920 solved for that software limitation. 0:07:55.440 --> 0:07:58.920 That software limitation is the chat GPT moment of it 0:07:58.960 --> 0:08:01.400 is now here. I think over the course of the 0:08:01.480 --> 0:08:03.480 next couple two three years, we're going to make some 0:08:03.640 --> 0:08:04.640 really big breakthroughs. 0:08:05.160 --> 0:08:08.480 Let's please talk about energy, electricity, power supply, call it 0:08:08.520 --> 0:08:11.400 what you will in turn. For each of you, how 0:08:11.400 --> 0:08:14.080 worried are you about it as a bottleneck and what 0:08:14.160 --> 0:08:16.560 is your experience day stay in running both companies. 0:08:17.160 --> 0:08:19.760 In that respect, energy should always be a bottleneck for 0:08:19.800 --> 0:08:23.080 any industry, and this is a new industry that's growing 0:08:23.200 --> 0:08:24.280 incredibly fast. 0:08:24.640 --> 0:08:25.120 As you know. 0:08:25.760 --> 0:08:28.960 AI is both the technology that's going to revolutionize many applications, 0:08:29.000 --> 0:08:31.840 and we're talking about some of them here, but the 0:08:31.880 --> 0:08:37.079 AI industry itself, the manufacturing of the artificial intelligence takes energy. 0:08:37.400 --> 0:08:40.480 It takes energy, it takes AI factories. It's exactly the 0:08:40.520 --> 0:08:45.280 reason why from Hopper to Blackwell we increased energy efficiency 0:08:46.040 --> 0:08:50.960 by tenx. From Blackwell to Reubin, we increased energy efficiency 0:08:51.040 --> 0:08:55.720 again by tenx. And that translates directly to our customer's 0:08:55.800 --> 0:09:00.760 revenues because in the case of an AI factory, whatever 0:09:01.440 --> 0:09:04.600 factory size you have, you're limited by the power, and 0:09:04.640 --> 0:09:07.480 that power within that power constrained, you want to have 0:09:07.520 --> 0:09:11.160 the most tokens or most AI per want that you 0:09:11.200 --> 0:09:15.319 can possibly generate. And so every time we improve energy efficiency, 0:09:15.520 --> 0:09:20.040 we're effectively improving both the AI capabilities for our customers 0:09:20.440 --> 0:09:23.839 and their revenues because they're always constrained by power. 0:09:24.000 --> 0:09:26.280 The response to your keinot and forgive me rowan one minute. 0:09:26.440 --> 0:09:29.440 Was you know vera rubin ten x through put really 0:09:29.480 --> 0:09:34.720 important tokens generated at one tenth. But they'll still say, hey, Jensen, 0:09:34.920 --> 0:09:37.920 what if the electricity is just not there, it's not available. 0:09:38.120 --> 0:09:41.040 Well, there's electricity, they're never enough electricity. 0:09:41.640 --> 0:09:43.920 That's actually what is the anxiety level for you both 0:09:43.960 --> 0:09:45.080 about that issue. 0:09:45.640 --> 0:09:49.280 There's always energy, they're never enough energy. And this is 0:09:49.600 --> 0:09:53.040 every and every industrial revolution will be energy constrained. 0:09:53.080 --> 0:09:55.920 And this industrial revolution is also energy constrained. 0:09:56.559 --> 0:09:59.200 Now one of the most important things here speaking about 0:09:59.200 --> 0:10:03.000 the United States, if not for President Trump's pro energy 0:10:03.040 --> 0:10:06.040 growth agenda, we would have a very hard time growing 0:10:06.040 --> 0:10:08.920 at all. In order for our new industry to emerge, 0:10:08.960 --> 0:10:11.920 you need energy, and so I think it's safe to 0:10:11.960 --> 0:10:14.600 say that we wish we had more energy in United States. 0:10:14.840 --> 0:10:17.439 You wi should have more energy. I think the world 0:10:17.480 --> 0:10:19.680 all wish we had more energy, and so we have 0:10:19.720 --> 0:10:22.480 to invest in all sorts of different forms of energy. 0:10:22.720 --> 0:10:26.120 But whatever energy you have, you have to make it 0:10:26.160 --> 0:10:28.360 as energy efficient as possible. And that's one of the 0:10:28.400 --> 0:10:32.000 reasons why both of us drive our technology roadmap so hard, 0:10:32.240 --> 0:10:35.360 because every single new generation of technology is more energy 0:10:35.360 --> 0:10:36.320 efficient than the last. 0:10:36.840 --> 0:10:37.959 That's kind of the nature of it. 0:10:38.040 --> 0:10:40.120 You are in the AI factory. I won't say data 0:10:40.120 --> 0:10:42.480 center is data centers where you store data. That you 0:10:42.559 --> 0:10:44.360 are in the AI factory supply chain. 0:10:44.760 --> 0:10:45.439 You must see it. 0:10:45.520 --> 0:10:48.080 So what we see is number one is you see 0:10:48.120 --> 0:10:52.840 that the energy demands is roughly scaling with the GDP growth. 0:10:52.960 --> 0:10:56.400 It's the coupling because executive of this effect that we 0:10:56.440 --> 0:10:59.400 are providing more and more efficient technology, So that means 0:10:59.720 --> 0:11:03.280 the link. But still it grows along with economic growth. 0:11:03.880 --> 0:11:06.400 And then on top comes a demand for data centers. 0:11:06.480 --> 0:11:09.240 They demand high quality of energy at the same time, 0:11:10.320 --> 0:11:13.679 which in some cases creates bottlenecks. Is it not from 0:11:13.720 --> 0:11:16.599 power generation? And the talk is it either renewables or 0:11:16.679 --> 0:11:19.600 gas turbines. There's a huge bottleneck for gas turbines as 0:11:19.600 --> 0:11:22.400 we speak. It goes all the way to high voltage transformers, 0:11:22.880 --> 0:11:26.480 to medium voltage into switch technologies. So therefore you see 0:11:26.520 --> 0:11:29.400 along the whole supply chain, obviously there's a huge demand 0:11:29.720 --> 0:11:32.360 we are serving that. There's a reason why this business 0:11:32.360 --> 0:11:35.440 of ours is growing extremely fast. We can keep up 0:11:35.440 --> 0:11:37.640 with the demand of our customers. But in some cases 0:11:37.679 --> 0:11:39.559 you might end up in bottlenecks if you keep on 0:11:39.679 --> 0:11:41.160 going as fast as possible. 0:11:41.280 --> 0:11:42.520 It's a regional dependence. 0:11:43.120 --> 0:11:46.280 In some cases, where you have good programs, good policies, 0:11:46.640 --> 0:11:48.920 you see that it's catching up along with the demand, 0:11:48.960 --> 0:11:51.560 and others you might end up in a gap. But 0:11:51.640 --> 0:11:56.240 you see the demand is higher than supplypply chain and we. 0:11:56.200 --> 0:11:59.800 Are ramping up standing we're invested, which is investing ultimately 0:11:59.840 --> 0:12:00.880 the condition we want. 0:12:01.440 --> 0:12:03.720 We want a condition where the demand is very strong 0:12:03.760 --> 0:12:06.720 because there's great demand for new technology. 0:12:07.080 --> 0:12:10.800 Specific to you, Jensen, how severe is the memory bottleneck 0:12:10.920 --> 0:12:11.360 right now? 0:12:12.480 --> 0:12:14.400 Well, the memory bottleneck is severe. 0:12:14.920 --> 0:12:17.720 But we're fortunate to have worked with all of you know, 0:12:17.760 --> 0:12:19.800 first of all, and Video is the only company that 0:12:19.880 --> 0:12:23.400 works with all three of the HBM suppliers, and all 0:12:23.520 --> 0:12:26.480 three of them are major customers and major suppliers of us, 0:12:26.520 --> 0:12:30.560 and so we had the good fortune of working with 0:12:30.559 --> 0:12:33.319 them for a very long period time, and we've got 0:12:33.320 --> 0:12:34.160 things all planned out. 0:12:34.440 --> 0:12:35.000 It's gonna be ak. 0:12:35.240 --> 0:12:36.840 You've been asked a lot about China in the last 0:12:36.840 --> 0:12:38.319 twenty four hours. I'm not going to ask you a 0:12:38.360 --> 0:12:41.520 slightly different question. What is the attitude of the Chinese 0:12:41.520 --> 0:12:44.800 government in allowing h two hundred to go into the country. 0:12:44.880 --> 0:12:48.199 I understand the position on licenses, the administration here in 0:12:48.240 --> 0:12:51.439 the United States position, but what does China's government say 0:12:51.480 --> 0:12:53.120 to you about how they want to approach it. 0:12:53.760 --> 0:12:56.360 I haven't spoken directly to them, but. 0:12:57.880 --> 0:13:01.640 Ultimately the way they'll commune ok through us to us, 0:13:01.720 --> 0:13:05.000 we'll be through the companies. If the companies are allowed 0:13:05.040 --> 0:13:08.080 to buy in video products in China, then they'll be 0:13:08.120 --> 0:13:10.599 strong demand. And we're seeing strong demand, and so I 0:13:10.640 --> 0:13:14.440 think indirectly they've communicated with their companies. 0:13:15.520 --> 0:13:19.679 I've learned a lot about Semens's work in software, and 0:13:19.760 --> 0:13:23.400 I've always thought the word Semens to be snall us 0:13:23.440 --> 0:13:27.800 with industrial manufacturing, physical things. It's the same lesson we've 0:13:27.800 --> 0:13:30.480 been through recently with in video, trying to understand. 0:13:30.160 --> 0:13:32.559 Actually the software competencies is. 0:13:32.480 --> 0:13:35.000 That an area where we might see you be active 0:13:35.040 --> 0:13:37.440 in m and a think about what you else you 0:13:37.520 --> 0:13:39.520 might need that you might not get from the Jensen 0:13:40.280 --> 0:13:41.319 in video partnership. 0:13:41.840 --> 0:13:46.280 So you might because we will increase our software competence 0:13:46.360 --> 0:13:50.360 our software Tolio actually it's alibit short of thirty billion, 0:13:50.400 --> 0:13:53.800 which Semens invested in MNA to below our software competence, 0:13:54.080 --> 0:13:57.960 which brings us to the point that's missing today exactly 0:13:58.000 --> 0:14:02.040 today we can build the most comprehensive physics based digital. 0:14:01.720 --> 0:14:03.480 Twin of whatever you want to build. 0:14:04.559 --> 0:14:07.839 Still, there are pieces and operational software software which runs 0:14:07.880 --> 0:14:11.280 plans where you can look into that I gives new 0:14:11.320 --> 0:14:14.359 spaces obviously be invested in dogmatics. 0:14:14.440 --> 0:14:18.240 It's life science, so it's about molecules. We can imagine 0:14:18.280 --> 0:14:21.680 doing more there also in simulation. But as we're building. 0:14:22.920 --> 0:14:26.000 A data backbound for this life science industry super elevent. 0:14:26.040 --> 0:14:29.240 We did it for other industries with Team Center. We 0:14:29.360 --> 0:14:31.640 repeat it now with Luma for life science, which is 0:14:32.040 --> 0:14:35.080 again shortening cightful times and reducing costs for life science 0:14:35.080 --> 0:14:38.880 and drugs and medicines. So and I mean the beauty 0:14:38.880 --> 0:14:41.160 of it is that we live in both worlds. We 0:14:41.320 --> 0:14:43.480 have to domain know how we know how to run things, 0:14:43.520 --> 0:14:48.200 how to operate, how to automate plans, buildings, grades, trains, 0:14:49.200 --> 0:14:52.280 and we have their software competence too, and that makes 0:14:52.320 --> 0:14:54.840 us in a unique position. Then and then we need 0:14:54.840 --> 0:14:59.440 strong partners which have complimentary technology, which is so amazing. 0:15:00.040 --> 0:15:02.400 If that comes together, you'll see matching things happen well. 0:15:02.440 --> 0:15:04.760 In video will have access to new technology on the 0:15:04.800 --> 0:15:08.400 infant side through GROC. I'm going to do my classic DNSEN. 0:15:08.440 --> 0:15:11.680 Could you please clarify for me question, which is is 0:15:11.720 --> 0:15:14.720 this an acquisition or is it a licensing deal spread 0:15:14.760 --> 0:15:17.880 over time. Because Jonathan and Sunny joined. 0:15:18.120 --> 0:15:21.240 We hired four hundred or something like that a little 0:15:21.240 --> 0:15:28.320 bit less incredible engineers, and we also license their technology. 0:15:27.680 --> 0:15:33.200 They designed an architecture that's very, very different than what 0:15:33.240 --> 0:15:38.000 we've done, and it's focused on low latency token generations 0:15:39.200 --> 0:15:42.440 and video is incredibly good at inference, and we're great 0:15:42.480 --> 0:15:45.880 at training post training as well as the inference phase 0:15:46.400 --> 0:15:50.680 and test time scaling of AI. And so we've got 0:15:50.400 --> 0:15:54.720 that space covered. I'm excited about some of the work 0:15:54.720 --> 0:15:56.400 that we might be able to do together to invent 0:15:56.480 --> 0:16:00.000 a new segment that might be able to address future use. 0:16:01.240 --> 0:16:03.600 I haven't described it to people, but the time will come. 0:16:03.760 --> 0:16:07.240 So you know, Jonathan is kind of behind the TPU 0:16:07.360 --> 0:16:10.560 and the LPU, and you know, before of this happened, 0:16:10.560 --> 0:16:14.360 there was Rock V two and V three and you 0:16:14.440 --> 0:16:16.280 know all of that, and it's leading to me to ask, 0:16:16.480 --> 0:16:18.480 what would that new platform or segment be. 0:16:18.840 --> 0:16:20.920 What is it that you're trying to build on. 0:16:20.920 --> 0:16:23.680 With those I haven't told anybody yet, but I'm here, Jens. 0:16:23.840 --> 0:16:24.480 You can't tell me. 0:16:24.760 --> 0:16:26.080 You know, I'm going to look in the camera and 0:16:26.120 --> 0:16:29.840 tell you that that come to a GtC conference in 0:16:29.840 --> 0:16:32.600 the future, and I will tell you all the secrets. 0:16:33.480 --> 0:16:35.760 I'm going to make a hard pivot because I have 0:16:35.840 --> 0:16:42.920 to data centers in space I reported just before the holiday. 0:16:43.000 --> 0:16:46.840 Is that the reason SpaceX wants to go public, Bear 0:16:46.880 --> 0:16:49.160 with me, Bear with me. The reason that SpaceX wants 0:16:49.160 --> 0:16:51.880 to raise thirty forty billion dollars whatever that is needs 0:16:51.880 --> 0:16:55.920 to buy the GPUs. Have you discussed data centers in 0:16:55.960 --> 0:16:58.560 space with Elon and SpaceX, Jensen. 0:16:58.520 --> 0:17:01.520 I can't discuss what I have told everybody. I've discussed 0:17:01.600 --> 0:17:02.280 with anybody. 0:17:03.280 --> 0:17:06.320 You think it's a viable technology platform, sure? 0:17:06.680 --> 0:17:09.360 Sure, I mean there's lots of energy in space, right 0:17:09.440 --> 0:17:14.639 and and the cooling is abundant in space, and so 0:17:14.760 --> 0:17:21.760 that the challenges of AI factories are different out in space. 0:17:21.760 --> 0:17:24.639 It would be AI factories in space. Yeah, forgive the 0:17:24.720 --> 0:17:27.280 ignorance almost, but are we literally talking about the same 0:17:28.040 --> 0:17:30.880 architecture for the GPU that goes into an AI factory 0:17:31.359 --> 0:17:32.919 still just being able to go into kind of a 0:17:32.960 --> 0:17:34.160 satellite form factor? 0:17:34.440 --> 0:17:34.640 Yeah? 0:17:34.720 --> 0:17:37.200 Sure, but the way that you would cool it empower 0:17:37.240 --> 0:17:40.320 it would be very different, and so the system design 0:17:40.359 --> 0:17:41.439 will be radically different. 0:17:41.840 --> 0:17:42.920 The chips will be the same. 0:17:43.200 --> 0:17:45.800 You're pretty concerned with data centers or WAI factories here 0:17:45.840 --> 0:17:49.000 on Earth to seem and see the need for or 0:17:49.119 --> 0:17:50.879 viability of such a technology. 0:17:51.480 --> 0:17:54.240 I don't know whether we explored or potentials on Earth yet, 0:17:54.600 --> 0:17:58.240 but there's one beauty in this idea. Think about any 0:17:58.320 --> 0:18:01.760 kind of manufacturing, which want you want to bring to space. 0:18:02.560 --> 0:18:05.440 Everything but you produce you want to bring down to Earth. 0:18:05.760 --> 0:18:08.960 It's hard if you do energy, how to bring it 0:18:09.000 --> 0:18:11.000 down to Earth. It's hard if you produce any kind 0:18:11.000 --> 0:18:14.600 of hardware how to bring it down. But tokens intelligence, 0:18:14.760 --> 0:18:17.280 I mean, you can transfer easily to Earth. So therefore, 0:18:17.520 --> 0:18:20.880 if I would starting produce something in space, I would 0:18:20.880 --> 0:18:21.360 start there. 0:18:22.640 --> 0:18:24.359 We should talk about what's on Homas driving. 0:18:24.840 --> 0:18:28.720 Have you seen Elon Musk's response to your keynote yesterday? 0:18:29.520 --> 0:18:32.240 What do you say part one in short ways, Well, 0:18:32.520 --> 0:18:33.080 we were. 0:18:32.920 --> 0:18:35.199 Already doing that smiley emoji on. 0:18:35.440 --> 0:18:38.159 X the the other part of it, by. 0:18:38.000 --> 0:18:39.200 The way, I would be surprised. 0:18:40.760 --> 0:18:43.080 First of all, I think I think the Tesla stack 0:18:43.240 --> 0:18:46.200 is the most advanced AV stack in the world, and 0:18:48.520 --> 0:18:51.120 I think the Tesla AV operations is the most advanced 0:18:51.160 --> 0:18:57.720 in the world. And I'm fairly certain that that they 0:18:57.880 --> 0:19:01.880 were already using n ai yes, and whether whether their 0:19:02.160 --> 0:19:05.800 their AI also did reasoning or not, it is somewhat 0:19:05.840 --> 0:19:07.840 secondary to that first part. 0:19:08.440 --> 0:19:11.000 His point was that the first ninety nine is hard enough, 0:19:11.040 --> 0:19:13.200 but the long tail thereafter is important. 0:19:13.200 --> 0:19:14.600 You know, it's very difficult. 0:19:14.760 --> 0:19:16.679 The question I got from the audience that was at 0:19:16.680 --> 0:19:19.840 your keynote was on a dollar per mile basis, what 0:19:19.960 --> 0:19:22.520 is the fundamental difference on your stack and your software 0:19:22.560 --> 0:19:25.120 approach to Tesla's vision based approach? 0:19:28.359 --> 0:19:29.760 Ours is also vision based. 0:19:30.200 --> 0:19:32.280 You know, of course we have we have in addition 0:19:32.400 --> 0:19:34.600 to to vision, we also have radar and light r 0:19:35.880 --> 0:19:39.920 but but otherwise otherwise the approach is rather similar. 0:19:40.640 --> 0:19:41.000 I think. 0:19:41.160 --> 0:19:44.280 I think Elon's approach is about as state of the 0:19:44.440 --> 0:19:50.119 art as anybody knows of a tunnel of striving robotics, 0:19:50.200 --> 0:19:52.880 and so it's it's a it's a stack that's hard 0:19:52.880 --> 0:19:53.440 to criticize. 0:19:53.600 --> 0:19:54.520 I wouldn't criticize it. 0:19:54.520 --> 0:19:57.720 I would just encourage them to continue to do what 0:19:57.760 --> 0:19:58.160 they're doing. 0:19:58.200 --> 0:19:58.960 They're doing a great. 0:19:58.880 --> 0:20:02.199 Job physically for you is and forgive my pronunciation, but 0:20:02.359 --> 0:20:07.159 er langan manufacturing site. When's that real? It's the same question. 0:20:07.760 --> 0:20:11.040 I mean, in contrast to the side you was talking 0:20:11.040 --> 0:20:13.520 about before this one exists. I mean, we are we 0:20:13.560 --> 0:20:15.760 are ramping up, we are automating it as we speak, 0:20:15.920 --> 0:20:19.399 using technology. I showcase some of that, and we are 0:20:19.440 --> 0:20:21.520 we are really deploying step by step now on that 0:20:21.680 --> 0:20:25.520 very same side, this technology we was talking about, the iebrain, 0:20:25.640 --> 0:20:30.879 the digital trim composer, real worlds and really time connectivity 0:20:31.000 --> 0:20:33.640 to what happens on the shop floor, so you can 0:20:33.720 --> 0:20:37.720