WEBVTT - Inside Stargate with OpenAI CEO Sam Altman 0:00:03.680 --> 0:00:06.840 I'm here on the streets of Abilene, Texas, asking people 0:00:06.880 --> 0:00:10.440 if they've heard of something called Stargate. So have you 0:00:10.480 --> 0:00:12.240 heard of Stargate? I have? 0:00:12.240 --> 0:00:13.200 Have you heard of Stargate? 0:00:13.240 --> 0:00:13.320 Oh? 0:00:13.760 --> 0:00:15.400 Have you heard of Project Stargate? 0:00:15.600 --> 0:00:16.200 A little bit? 0:00:16.320 --> 0:00:16.959 Not very much. 0:00:17.280 --> 0:00:18.799 Until recently, I had not. 0:00:19.160 --> 0:00:21.119 Yeah, I've heard just so just a smitch about it, 0:00:21.160 --> 0:00:24.279 that it was announced by Trump, and that's kind of 0:00:24.360 --> 0:00:25.480 really all that I've heard. 0:00:25.920 --> 0:00:28.080 A year ago, if you asked me what stargate is, 0:00:28.440 --> 0:00:31.200 I might have said something like, oh, yeah, that's an 0:00:31.240 --> 0:00:35.040 extraterrestrial device that allows interplanetary travel by creating a wormhole 0:00:35.080 --> 0:00:37.600 between two ring shaped portals, as depicted in the nineteen 0:00:37.680 --> 0:00:40.720 ninety four film Stargate and the various television spinoffs. 0:00:41.159 --> 0:00:42.640 How does she old of the Stargate? 0:00:44.360 --> 0:00:45.440 But in twenty twenty. 0:00:45.200 --> 0:00:48.760 Five, Stargate isn't just something you'll hear about from Jaffa Warriors, 0:00:48.760 --> 0:00:52.479 from Chulac. Now it's something you'll hear from a different 0:00:52.479 --> 0:00:55.480 type of TV personality, Stargate. 0:00:55.880 --> 0:00:58.560 So put that name down in your books, because I 0:00:58.600 --> 0:00:59.920 think you're going to hear a lot about. 0:01:01.240 --> 0:01:05.000 On January twenty first, President Trump flanked by oracles Larry Ellison, 0:01:05.120 --> 0:01:09.360 Open Ai, Sam Altman, and SoftBank's Macioshi Sun announced Stargate, 0:01:09.680 --> 0:01:13.280 an AI infrastructure project investing one hundred billion dollars in 0:01:13.360 --> 0:01:16.640 data center development and a stated intention of investing five 0:01:16.720 --> 0:01:18.959 hundred billion dollars over the next four years. 0:01:19.520 --> 0:01:23.080 That's great that you're coming in together. That's a massive 0:01:23.080 --> 0:01:25.080 group of talent and money. 0:01:25.319 --> 0:01:28.600 On Fox News, White House Press Secretary Caroline Levitt pitched 0:01:28.600 --> 0:01:31.480 Stargate as a triumph for America and the Trump administration. 0:01:31.920 --> 0:01:35.960 President Trump is very excited about this infrastructure announcement in 0:01:36.080 --> 0:01:42.119 the field of AI. From the basic optics of the announcement, 0:01:42.319 --> 0:01:44.960 one might assume that this is some kind of ambitious 0:01:45.000 --> 0:01:48.360 federal moonshot, perhaps the result of a historic high tech 0:01:48.440 --> 0:01:52.200 infrastructure bill. Maybe it's part of something called the Stargate Act, 0:01:53.280 --> 0:01:55.960 but none of that is the case. Stargate is a 0:01:56.000 --> 0:02:00.760 private initiative. Stargate LLC, a joint venture fund by American 0:02:00.800 --> 0:02:04.320 companies open Ai and Oracle, as well as Japanese company 0:02:04.360 --> 0:02:07.800 SoftBank and MGX, an AI investment firm owned by the 0:02:07.840 --> 0:02:12.280 government of Abu Dhabi. At its core, Stargate is a 0:02:12.320 --> 0:02:16.280 tech industry powerplay by the hottest company in AI. Stargate 0:02:16.320 --> 0:02:19.959 will serve chatchipt maker open AI as it builds increasingly 0:02:20.080 --> 0:02:23.600 powerful AI products. And while Stargate is currently investing in 0:02:23.639 --> 0:02:26.639 American facilities, the ambitions are global. 0:02:26.880 --> 0:02:27.840 Want to say this missed that. 0:02:28.080 --> 0:02:31.440 On February third, Sam Altman joined Masiyoshi San on stage 0:02:31.480 --> 0:02:34.240 in Tokyo, where they discussed building out Stargate in Japan. 0:02:34.760 --> 0:02:38.800 The Stargate have to also expand into Japan because of 0:02:38.960 --> 0:02:40.280 the regulation. 0:02:40.560 --> 0:02:42.120 We have to respect. 0:02:41.600 --> 0:02:45.960 The national security, the privacy. 0:02:46.120 --> 0:02:49.720 Loo soft Bank is building a big data center here. Yes, yes, 0:02:49.919 --> 0:02:52.080 so we're going to expand. We're excited to surround on 0:02:52.120 --> 0:02:55.399 Stargate into Japanese infrastructure also right. 0:02:55.919 --> 0:02:58.560 Four days later, Altmann told an audience in Berlin that 0:02:58.600 --> 0:03:01.040 he hopes to find a way to overcome regulatory hurdles 0:03:01.080 --> 0:03:03.240 in Europe in order to expand Stargate there. 0:03:03.320 --> 0:03:07.320 Well, I'm very convinced that most Europeans want AI to 0:03:07.320 --> 0:03:09.600 be used here to happen here. They wanted to revitalize 0:03:09.600 --> 0:03:13.119 economic growth, they want to drive science, they want European infrastructure, 0:03:13.160 --> 0:03:15.760 and so do. We would love to do like a 0:03:15.800 --> 0:03:18.880 Stargate Europe. We need help, but we'd love to build 0:03:18.880 --> 0:03:20.559 something in here. For Europe that government. 0:03:20.280 --> 0:03:21.480 Operate beyond that. 0:03:21.760 --> 0:03:25.480 Amarati Staate owned investment firm MGX, which was not represented 0:03:25.520 --> 0:03:28.160 on stage at the press conference with President Trump, has 0:03:28.200 --> 0:03:31.120 a stated goal of quote enabling the AI fabric of 0:03:31.160 --> 0:03:35.600 the global economy, and on May seventh, OpenAI announced Open 0:03:35.640 --> 0:03:38.680 AI for Countries, described as a new initiative within the 0:03:38.680 --> 0:03:41.800 Stargate project to quote partner with countries to help build 0:03:41.880 --> 0:03:45.200 in country data center capacity. But even if the American 0:03:45.200 --> 0:03:48.440 public has no direct stake in Stargate, the ventures alliance 0:03:48.480 --> 0:03:52.040 with President Trump isn't superficial. The President has promised to 0:03:52.120 --> 0:03:54.520 use the power of the Executive Office to help Stargate 0:03:54.520 --> 0:03:56.640 cut any red tape that may stand in their way, 0:03:57.080 --> 0:03:59.960 an approach towards rapid AI development that marks a star 0:04:00.280 --> 0:04:02.000 change from the previous administration. 0:04:04.240 --> 0:04:05.160 One thing is clear. 0:04:06.000 --> 0:04:08.520 To realize the promise of AI and avoid the risk, 0:04:08.600 --> 0:04:10.200 we need to govern this technology. 0:04:10.560 --> 0:04:14.400 President Biden that after signing an Executive Order on ntificial Intelligence, 0:04:14.440 --> 0:04:18.880 the most significant step today in regulating the technology. On 0:04:18.960 --> 0:04:21.640 the day he was inaugurated, President Trump sitting at a 0:04:21.680 --> 0:04:23.960 desk on stage at a rally like event in the 0:04:23.960 --> 0:04:27.640 Capitol One Arena in Washington, d c repealed President Biden's 0:04:27.680 --> 0:04:31.360 AI protections as part of a procession of new executive orders. 0:04:31.520 --> 0:04:33.720 Could you imagine Biden doing this? 0:04:33.839 --> 0:04:34.520 I don't think so. 0:04:35.040 --> 0:04:37.400 And the very next day, President Trump sought to pave 0:04:37.440 --> 0:04:40.760 the way for rapid AI infrastructure expansion by declaring a 0:04:40.880 --> 0:04:45.279 national energy emergency. President Trump is declaring a national emergency 0:04:45.360 --> 0:04:46.040 in the US. 0:04:46.400 --> 0:04:48.760 Well, this is a first for a president to declare 0:04:48.839 --> 0:04:52.960 a national energy emergency, and in short, we're not in one. 0:04:53.040 --> 0:04:55.599 In fact, most analysts say we have we're very close 0:04:55.640 --> 0:04:58.480 to having a short term oversupply of oil and gas 0:04:58.560 --> 0:04:59.960 in this country. 0:05:00.160 --> 0:05:02.360 With a key Trump campaign promise. 0:05:02.360 --> 0:05:05.080 Drill baby drill in a drill, baby drill. 0:05:05.800 --> 0:05:08.599 Throughout his campaign, President Trump linked the needs of the 0:05:08.640 --> 0:05:11.640 AI industry with a call to increase energy production and 0:05:11.720 --> 0:05:13.760 deregulate the oil and gas industry. 0:05:13.839 --> 0:05:17.719 AI is a big deal for us to compete successfully 0:05:17.760 --> 0:05:20.280 with China and others on AI, We're going to need 0:05:20.360 --> 0:05:23.040 twice the electricity that we currently produce. 0:05:23.080 --> 0:05:26.679 Can you believe that the declaration of a National energy 0:05:26.680 --> 0:05:29.480 emergency is one of several tools that President Trump is 0:05:29.560 --> 0:05:32.960 using to try and help stargate bypass environmental protections and 0:05:33.080 --> 0:05:35.719 other regulations that may impede its rapid growth. 0:05:36.800 --> 0:05:40.479 I will approve new drilling, new pipelines, new refineries, new 0:05:40.520 --> 0:05:44.279 power plants, new reactors, and we will slash the red tape. 0:05:44.520 --> 0:05:45.960 We will get the job done. 0:05:46.160 --> 0:05:51.240 We will create more electricity also for these new industries 0:05:51.279 --> 0:05:55.760 that can only function with massive electricity. And we'll get 0:05:55.800 --> 0:05:57.760 it done. And nobody else is even going to come 0:05:57.760 --> 0:05:58.880 to us, They won't even try. 0:05:59.279 --> 0:06:02.839 Partnering with the Trump administration provides a crucial strategic advantage 0:06:02.839 --> 0:06:05.400 for open ai and its allies in the AI race, 0:06:05.839 --> 0:06:09.039 a green light to aggressively pilot their global ambitions on 0:06:09.080 --> 0:06:13.200 American soil, regardless of the consequences, and the first place 0:06:13.200 --> 0:06:18.839 they're breaking ground is in Abilene, Texas. The immediate downsides 0:06:18.839 --> 0:06:22.800 of this are measurable but arguably boring. Mainly, there will 0:06:22.800 --> 0:06:26.360 be a climate impact. Microsoft, which was open AI's exclusive 0:06:26.400 --> 0:06:30.640 cloud computing provider prietor Stargate, has admitted to significantly backsliding 0:06:30.680 --> 0:06:33.240 on its climate goals because of the AI boom. Here's 0:06:33.279 --> 0:06:36.320 what Microsoft president Brad Smith told Bloomberg about the company's 0:06:36.360 --> 0:06:37.000 climate goals. 0:06:37.520 --> 0:06:41.480 In twenty twenty, we unveiled what we called our carbon 0:06:41.560 --> 0:06:46.200 moonshot our goal of being carbon negative by twenty thirty. 0:06:47.120 --> 0:06:51.960 That was before the explosion and artificial intelligence. So in 0:06:52.040 --> 0:06:56.520 many ways, as I say across Microsoft, the moon has moved. 0:06:57.000 --> 0:07:00.560 It's more than five times as far away as it 0:07:00.680 --> 0:07:03.360 was in twenty twenty. If you just think about our 0:07:03.440 --> 0:07:08.720 own forecast for the expansion of AI and its electrical needs. 0:07:09.000 --> 0:07:12.400 And President Trump seems to take every opportunity to dissuade 0:07:12.440 --> 0:07:14.720 businesses from relying on renewable energy. 0:07:14.960 --> 0:07:18.040 They can fuel it with anything they want, and they 0:07:18.080 --> 0:07:21.160 may have call as a backup, good clean call. 0:07:21.240 --> 0:07:25.160 You know, the wind is wonderful if you like high costs, 0:07:25.640 --> 0:07:28.400 if you like having to replace the windmills every nine years. 0:07:28.400 --> 0:07:31.080 You know, they wear out, and they're very intermittent, and 0:07:31.120 --> 0:07:32.600 they do kill a hell of a lot of birds. 0:07:32.600 --> 0:07:34.120 If you want to see a lot of dead birds, 0:07:34.160 --> 0:07:36.440 go to a windmill and just look underneath the seat. 0:07:36.440 --> 0:07:37.640 Birds all over the place. 0:07:37.840 --> 0:07:42.280 Nothing can destroy coal, not the weather, not a bomb, nothing. 0:07:42.320 --> 0:07:44.920 It might make it a little smaller, might make it 0:07:44.920 --> 0:07:46.920 a little different shape. 0:07:47.320 --> 0:07:49.520 On the other hand, the long term upsides of the 0:07:49.520 --> 0:07:54.200 stargate facilities may be harder to quantify, but sound fantastically utopian. 0:07:54.840 --> 0:07:57.960 Here's Altman, echoing Larry Ellison's hope to use Stargate to 0:07:58.040 --> 0:07:58.680 cure cancer. 0:07:59.040 --> 0:08:01.680 We will be amazed at how quickly we're carrying dis cancer, 0:08:01.720 --> 0:08:04.600 and that one in heart disease, and what this will 0:08:04.600 --> 0:08:08.720 do for the ability to deliver very high quality healthcare, 0:08:08.800 --> 0:08:11.960 the costs, but really to cure the diseases at a 0:08:12.040 --> 0:08:12.840 rapid rapid rate. 0:08:12.960 --> 0:08:15.960 These kinds of monumental outcomes are of course yet to 0:08:15.960 --> 0:08:19.240 be proven, but one thing is certain. The build out 0:08:19.280 --> 0:08:20.960 of Stargate is happening. 0:08:23.720 --> 0:08:24.680 On this week's. 0:08:24.360 --> 0:08:26.360 Episode of The Circuit, I had a chance to tour 0:08:26.440 --> 0:08:29.400 the construction site of the new Stargate facilities being built 0:08:29.400 --> 0:08:32.679 in Abilene, Texas, and here on the podcast, we're sharing 0:08:32.760 --> 0:08:35.720 our extended interview with the man who arguably stands to 0:08:35.760 --> 0:08:39.360 gain the most from this venture, open Ai CEO Sam Alman. 0:08:39.880 --> 0:08:42.280 I sat down with Altman at open AI's headquarters in 0:08:42.320 --> 0:08:45.079 San Francisco a week after the launch of GPT four 0:08:45.120 --> 0:08:49.320 row's new image generation capabilities, a feature that went totally viral. 0:08:49.640 --> 0:08:52.000 I think our conversation revealed a lot about why he 0:08:52.080 --> 0:08:55.840 sees the success of Stargate as essential to continuing open 0:08:55.880 --> 0:08:58.360 AI's status as the leader in the field of AI. 0:09:19.720 --> 0:09:22.760 So I saw what's happening in Appling with my own eyes. 0:09:23.280 --> 0:09:25.800 It's happening. It's happening. People look like they were working 0:09:25.880 --> 0:09:26.840 very hard and fast. 0:09:27.120 --> 0:09:30.360 People never love what I show up for data center business. No, 0:09:30.480 --> 0:09:35.280 why A well, because I like have opinions about how 0:09:35.800 --> 0:09:38.080 you know, like it's not my area expertise, and yet 0:09:38.120 --> 0:09:39.720 I have opinions. And so people are like, Okay, you 0:09:39.720 --> 0:09:42.480 know we've thought about that before, Like is this Yes, 0:09:42.480 --> 0:09:44.160 we're going as fast as we can. Yes, we have 0:09:44.240 --> 0:09:46.080 tried this other kind of cool thing. Yes, we realized 0:09:46.080 --> 0:09:47.400 this is like a stupid way of doing it. We're 0:09:47.440 --> 0:09:48.440 trying to fix something exactly. 0:09:48.480 --> 0:09:51.480 Whistlelone, take me back to the beginning before stargate. What 0:09:51.640 --> 0:09:54.040 did you start seeing that made you realize we need 0:09:54.200 --> 0:09:57.320 more compute, more power, more connectivity, that you were hitting 0:09:57.320 --> 0:09:58.520 a limit and needed to scale. 0:09:59.800 --> 0:10:01.360 We we used to think a lot about the compute 0:10:01.360 --> 0:10:05.520 we would need to train the models, and what we 0:10:05.559 --> 0:10:07.360 didn't used to think about as much as how much 0:10:07.360 --> 0:10:09.960 people were going to use these models. I mean, we 0:10:09.960 --> 0:10:11.520 didn't think about it. Just turned out people want to 0:10:11.559 --> 0:10:15.760 use the models much more than we imagined. And a 0:10:15.800 --> 0:10:18.520 couple of years ago, maybe after the launch of GPT 0:10:18.679 --> 0:10:22.560 four in chatwipt, it really started to hit like, oh man, 0:10:22.760 --> 0:10:24.840 this is this is like gone from a lot of 0:10:24.840 --> 0:10:28.040 compute to like the biggest infrastructure project in history. And 0:10:28.080 --> 0:10:29.520 we started thinking about ways to do it, and we 0:10:29.520 --> 0:10:31.439 started trying to understand where the limits in the supply 0:10:31.520 --> 0:10:37.559 chain were, and out of that emerged Stargate. There were 0:10:37.559 --> 0:10:39.920 other smaller things that we did. First. We have, you know, 0:10:40.080 --> 0:10:44.040 worked with other partners. We've worked with Microsoft to build 0:10:44.040 --> 0:10:48.120 that like very gigantic amounts of compute. But you know, 0:10:48.200 --> 0:10:50.200 this is like the next step in that evolution, right. 0:10:50.280 --> 0:10:52.439 I mean, you traveled around the world talking to people 0:10:52.440 --> 0:10:52.800 about this. 0:10:53.280 --> 0:10:55.840 How did you get to talking with Masiyoshi's son and 0:10:55.920 --> 0:10:58.480 Larry Ellison and how did you all come together? 0:10:59.840 --> 0:11:03.240 I'm well, I've known Massa for a long time, but 0:11:03.480 --> 0:11:08.839 I I was on in twenty twenty three. I did 0:11:08.880 --> 0:11:10.679 sort of two long trips. I think that was the year. 0:11:10.720 --> 0:11:14.760 That's all it's all about. I did two kind of 0:11:14.760 --> 0:11:16.560 long trips around the world, and a lot of it 0:11:16.600 --> 0:11:20.199 was to talk to developers and governments, but a lot 0:11:20.240 --> 0:11:22.160 of it was also to just really try to like 0:11:22.440 --> 0:11:25.360 get my head around the supply chain. This is not 0:11:25.440 --> 0:11:27.160 meaning I've done before, Like I had not before thought 0:11:27.160 --> 0:11:28.600 about like what it was going to take to get 0:11:28.600 --> 0:11:31.280 compute up and running at this scale. And there are 0:11:31.320 --> 0:11:36.160 a lot of like hard pieces. And on one of 0:11:36.160 --> 0:11:40.120 those trips I met with Masa. Masa has like thought 0:11:40.160 --> 0:11:44.360 for a long time about chip fabs in particular, but 0:11:44.640 --> 0:11:46.320 really kind of the whole thing that it takes. And 0:11:46.360 --> 0:11:49.640 we got to speaking about what it would you know, 0:11:49.679 --> 0:11:51.400 what it would take to do compute at this kind 0:11:51.440 --> 0:11:53.319 of scale. And then it took us a while to 0:11:53.760 --> 0:11:56.240 figure out. I mean it is a is a complex 0:11:56.280 --> 0:11:58.040 supply chain, a lot of partners and obviously a lot 0:11:58.040 --> 0:11:59.000 of capital, you. 0:11:58.960 --> 0:11:59.520 Know, soft things. 0:11:59.520 --> 0:12:02.480 Obviously your financial partner oracles a technical partner. 0:12:02.679 --> 0:12:05.280 Why couldn't you get what you needed from Microsoft? 0:12:05.520 --> 0:12:07.040 I mean, we do get a lot of great stuff 0:12:07.040 --> 0:12:08.880 from Microsoft, but I think this is more than anyone 0:12:08.920 --> 0:12:12.200 company can deliver. We have. Microsoft will do a lot 0:12:12.240 --> 0:12:14.559 of compute for us a lot long. We We're very 0:12:14.559 --> 0:12:15.160 happy about that. 0:12:15.320 --> 0:12:16.280 Why call it stargate? 0:12:18.520 --> 0:12:21.320 It began as a codename, and sometimes code names stick. 0:12:22.800 --> 0:12:24.079 But it means something, right. 0:12:26.360 --> 0:12:30.040 The design of one of the very early layouts of 0:12:30.080 --> 0:12:33.880 a data center looked a little bit like Stargate from. 0:12:33.120 --> 0:12:35.080 The show A futuristic wormhole. 0:12:35.520 --> 0:12:36.280 Not like that at all. 0:12:37.280 --> 0:12:40.560 Okay, you announced this at the White House the day 0:12:40.559 --> 0:12:44.200 after the inauguration. How did this get to President Trump's 0:12:44.240 --> 0:12:46.600 desk and what was that moment? 0:12:46.920 --> 0:12:50.120 Like, Well, the President is super interested in infrastructure, and 0:12:50.160 --> 0:12:54.040 he's made a big priority which I think is wonderful 0:12:54.559 --> 0:13:00.599 about permitting the energy production and the facilities for the 0:13:00.679 --> 0:13:03.320 data centers. So that was kind of how the conversation started. 0:13:03.440 --> 0:13:08.000 Okay, we're in the middle of a step change for AI, 0:13:08.160 --> 0:13:09.960 and clearly what we saw just in the last week 0:13:10.000 --> 0:13:13.760 from you alone, you're seeing what's next on the roadmap. 0:13:14.240 --> 0:13:18.120 How does that inform what the design of the data 0:13:18.160 --> 0:13:20.200 centers of the future need to be. 0:13:22.960 --> 0:13:26.559 I mean, I think there's like a lot of details 0:13:26.600 --> 0:13:29.120 we're learning about how to design these data centers, what 0:13:29.160 --> 0:13:32.839 you want really at all levels, from like the chip 0:13:32.880 --> 0:13:36.079 to the sort of architecture of whole data center. But 0:13:36.559 --> 0:13:39.400 the main thing that's been on my mind, and I 0:13:39.400 --> 0:13:42.160 think on many people's mind, is just how much inference 0:13:42.240 --> 0:13:46.760 demand there is. We are crazily constrained and we have 0:13:48.160 --> 0:13:52.600 a gigantic computer fleet, like gigantic, giantic and yet still 0:13:53.120 --> 0:13:54.800 if we had twice as much, we would be able 0:13:54.800 --> 0:13:56.880 to offer much better products and services, which just no 0:13:56.880 --> 0:14:02.240 shortage of demand. So for me, there are all the 0:14:02.240 --> 0:14:05.520 technical lessons about what we've learned and how we want 0:14:05.559 --> 0:14:07.559 to build this, but mostly we just want a lot. 0:14:07.480 --> 0:14:08.640 Explain the math to us. 0:14:08.760 --> 0:14:10.559 How does it all add up to needing five hundred 0:14:10.600 --> 0:14:13.520 billion dollars because folks out there don't get the accounting, 0:14:14.480 --> 0:14:17.480 who's putting in what, how much funding is really secured? 0:14:18.559 --> 0:14:19.960 How does that add up in terms of like what 0:14:20.000 --> 0:14:22.400 do we need that for? How does the funding come together? 0:14:22.480 --> 0:14:24.480 Like, explain that number? Five hundred billion? 0:14:24.520 --> 0:14:28.480 Oh, well, that covers the capacity we think we need 0:14:28.480 --> 0:14:31.880 for the next few years given our growth projections. You know, 0:14:32.000 --> 0:14:35.800 we we The interesting question is if we had if 0:14:35.840 --> 0:14:37.400 we knew how to get a trillion dollars right now, 0:14:37.440 --> 0:14:41.520 which we don't, would we be able to deploy that 0:14:41.560 --> 0:14:45.120 profit in the next few years? And I'm not sure 0:14:45.200 --> 0:14:47.160 about that, but I feel confident we can like make 0:14:47.160 --> 0:14:48.440 five hundred billion of value back. 0:14:48.560 --> 0:14:51.440 Let's talk about that, because you just raised even more money, 0:14:51.640 --> 0:14:54.240 you know, or right into open AI. 0:14:55.320 --> 0:14:57.720 How confident? It's a lot of money, tens of billions 0:14:57.760 --> 0:14:58.200 of dollars. 0:14:58.360 --> 0:15:01.240 How confident are you that open ai is going to 0:15:01.240 --> 0:15:05.360 be a financially sustainable and profitable enough company to justify 0:15:05.720 --> 0:15:06.720 all of this investment. 0:15:07.560 --> 0:15:09.840 It's looking like we're doing really well. I mean, like 0:15:09.880 --> 0:15:14.160 we have to. We are definitely doing something unprecedented. But 0:15:15.880 --> 0:15:19.920 you know, it seems like good. Like I feel confident 0:15:19.960 --> 0:15:22.280 in the bed doesn't mean something can't go wrong. 0:15:23.040 --> 0:15:26.720 You tweeted that the GPUs were melting you. 0:15:26.880 --> 0:15:28.680 Yeah, I didn't realize we're gonna take that literally. 0:15:29.240 --> 0:15:31.280 I mean, I get that it was a joke, but 0:15:31.960 --> 0:15:33.240 it ran very hot. 0:15:33.040 --> 0:15:34.800 But like the metal is not actually melting. 0:15:35.000 --> 0:15:39.960 GPU shortage, bro that the team can't sleep, you're so busy. 0:15:40.440 --> 0:15:43.240 What happens when a launch goes viral? And how does 0:15:43.280 --> 0:15:44.400 stargate solve for this? 0:15:45.480 --> 0:15:48.440 Well, first of all, this level of my route, this 0:15:48.480 --> 0:15:51.000 is an unusual thing this last week. I don't think 0:15:51.000 --> 0:15:52.480 this has happened in the history of tech to any 0:15:52.480 --> 0:15:56.880 company before, like the I don't know of any I've 0:15:56.880 --> 0:15:59.080 seen viral moments, but I have never seen anyone have 0:15:59.120 --> 0:16:00.800 to deal with an influx usage like this. 0:16:01.080 --> 0:16:03.560 You added a million users in an hour or something. 0:16:03.280 --> 0:16:07.120 With yeah, I mean more some hours. But yeah, it 0:16:07.240 --> 0:16:10.760 was like a it was like unprecedented wild and also 0:16:10.880 --> 0:16:13.160 like making an image is not a it's not exactly 0:16:13.200 --> 0:16:15.000 like a low compute task the way we do it 0:16:15.000 --> 0:16:18.200 with the new image then, so we had to do 0:16:18.240 --> 0:16:20.080 a lot of very unnatural things. We had to you know, 0:16:20.080 --> 0:16:22.480 borrow compute capacity for research. We had to slow down 0:16:22.520 --> 0:16:24.880 some other features because it's not like we have like 0:16:25.000 --> 0:16:27.840 hundreds of thousands of GPUs sitting around just like spinning idly. 0:16:28.320 --> 0:16:32.600 So with if we had more GPUs, we would be 0:16:32.760 --> 0:16:38.680 more able to handle demandswer just like this, and also 0:16:38.720 --> 0:16:42.240 we wouldn't have to put such restrictions on I was 0:16:42.400 --> 0:16:44.200 just this morning making the list of what we have 0:16:44.400 --> 0:16:47.600 what we'd like to launch in the next few months, 0:16:48.120 --> 0:16:50.400 and I was like, you know, I had the list 0:16:50.440 --> 0:16:52.360 over here, and I had like my rough math of 0:16:52.400 --> 0:16:54.160 when we're getting GPUs here and where I thought we 0:16:54.160 --> 0:16:55.760 could get the efficiency games, and I was like, this 0:16:55.800 --> 0:16:58.440 is not gonna work. Like we're gonna have to like 0:16:58.560 --> 0:17:00.640 pull some of these things, make some trail of limits, 0:17:00.640 --> 0:17:02.600 some things I don't like, but this is not all 0:17:02.640 --> 0:17:03.920 of this is not going to fit here. 0:17:04.520 --> 0:17:05.720 So it's that directly linked. 0:17:06.040 --> 0:17:08.760 Yeah. I mean you can find efficiency gains and we 0:17:08.840 --> 0:17:10.600 chart to do. You can limit features and usage and 0:17:10.640 --> 0:17:13.160 we do that too. But you know, more compute means 0:17:13.240 --> 0:17:15.240 we can give you more AI and you can use 0:17:15.280 --> 0:17:17.120 it for images or write softwareor whatever you want. 0:17:17.600 --> 0:17:21.160 Elon's got his own data centers and XAI just bought 0:17:21.359 --> 0:17:23.520 x which means he's got access to all of that 0:17:23.560 --> 0:17:24.359 additional data. 0:17:25.160 --> 0:17:28.040 He claims Grock is the smartest AI on Earth. How 0:17:28.040 --> 0:17:29.080 do you think it compares? 0:17:29.880 --> 0:17:32.440 I don't use it much. I think it's probably a 0:17:32.560 --> 0:17:33.119 very good model. 0:17:33.359 --> 0:17:35.600 I mean, obviously there are lots of competitors out. 0:17:35.400 --> 0:17:42.879 There, there's a lot of good models. I think I 0:17:42.880 --> 0:17:47.160 think good models will become very bountiful, very plentiful. 0:17:47.840 --> 0:17:50.680 So what's your edge going to be the best. 0:17:50.520 --> 0:17:52.959 Infrastructural air and the best top of the stack. You know, 0:17:53.000 --> 0:17:56.960 we have way more people use chatpt than use any 0:17:56.960 --> 0:18:01.960 other AI service, way more, and I think you'll see 0:18:01.960 --> 0:18:03.480 with a lot of the new futures we roll out 0:18:03.520 --> 0:18:06.760 that will become more of an advantage over time. 0:18:06.920 --> 0:18:09.200 So, once this is all built, what's the grand vision? 0:18:11.160 --> 0:18:13.520 Give people tools to let them do whatever they're going 0:18:13.560 --> 0:18:16.800 to do better. You know this is in some sense 0:18:16.840 --> 0:18:18.879 this is like it is different this time, But in 0:18:19.040 --> 0:18:21.800 many other senses, this is just like another tool and 0:18:21.840 --> 0:18:25.960 another piece of technological history. People will use it and 0:18:26.040 --> 0:18:28.760 unuse their creative energy and make all sorts of stuff 0:18:28.760 --> 0:18:30.960 that you and I love, or some stuff that we don't, 0:18:31.440 --> 0:18:36.720 or stuff that just entertain themselves personally. The area I'm 0:18:36.760 --> 0:18:40.040 most excited about is air for scientific discovery. I think 0:18:40.080 --> 0:18:42.480 that will you know, we're not there yet, but we're 0:18:42.480 --> 0:18:44.760 not far away. I think twenty twenty five will be 0:18:45.359 --> 0:18:48.400 a world where we have agents do a lot of work, 0:18:48.440 --> 0:18:50.199 but work of the kind of work and things we 0:18:50.240 --> 0:18:52.240 already know how to do. I'm hopeful that twenty twenty 0:18:52.280 --> 0:18:56.000 six will be a big year of like really new 0:18:56.000 --> 0:18:56.880 scientific progress. 0:18:57.320 --> 0:19:00.840 Do you envision multiple stargets on every continent? Like where 0:19:00.880 --> 0:19:02.240 else are we going to see you break ground? 0:19:02.280 --> 0:19:03.480 You will see them on in the comments? 0:19:03.560 --> 0:19:05.800 Yes everywhere, I don't know. 0:19:05.760 --> 0:19:09.240 Mind everywhere, Like this is like you want to visit 0:19:09.240 --> 0:19:11.680 on this These are hard things to do. 0:19:12.280 --> 0:19:14.359 Well, Look, this is an audacious bet on the future. 0:19:14.800 --> 0:19:15.639 What are the risks? 0:19:18.800 --> 0:19:21.439 I mean, maybe you know people stop wanting to like 0:19:21.520 --> 0:19:23.919 pay for AI services and then we have a difficult 0:19:23.960 --> 0:19:25.240 financial position. Mm hmm. 0:19:26.600 --> 0:19:29.000 How do you feel about one chip maker in Nvidia 0:19:29.440 --> 0:19:32.480 having so much power over the industry, the future of 0:19:32.520 --> 0:19:33.000 the industry. 0:19:33.080 --> 0:19:35.960 They make an incredible product, and if you make an 0:19:36.000 --> 0:19:38.119 incredible product, then people want it. This is what happens 0:19:38.119 --> 0:19:38.280 to you. 0:19:39.560 --> 0:19:40.960 I do want to talk about the jobs then, because 0:19:40.960 --> 0:19:43.360 obviously we saw so many people working there. There's this 0:19:43.640 --> 0:19:46.000 lofty promise that AI data centers are going to create 0:19:46.000 --> 0:19:50.480 thousands and thousands of jobs. Meantime, AI is destroying jobs elsewhere, 0:19:50.560 --> 0:19:53.320 and I feel like even you know, there is serious 0:19:53.359 --> 0:19:57.480 anxiety out there. People are scared totally, even among the 0:19:57.480 --> 0:19:59.879 best engineers and technologists, people are scared. 0:20:00.920 --> 0:20:02.119 What do you say to those people? 0:20:03.400 --> 0:20:06.720