[0:00:00] Hey, team. Oh, gee. Are we late? Guys. Good to be here. Go, America. Go, America. Team America. Go. Yes, indeed. So, America, Wright Brothers, first man on the moon, the internet. The two of us, Wright Brothers. I like that. Indeed. We've...
[0:00:26] We've essentially led every major technological revolution, which has been to the enormous benefit of America, ordinary Americans, where I guess we're now calling this superintelligence, which I think is actually good branding. And so let's talk about how America wins this, and we make that winning good for ordinary Americans. Well, excuse me, Diet Coke. Sorry. So first of all,
[0:00:54] Artificial superintelligence is a complete reinvention of the computing stack. Everything from the way the processors, the chips are designed and the way they operate to the way software is now trained and developed to the way that the applications are able to perform what they're able to do and how they perform. Every single layer of the computing stack has been reinvented. This way of doing computing,
[0:01:23] Unlike previous ways of retrieving information, this is a generative approach to computing. And therefore, in order for us to succeed, the first thing we need is energy. Because we need the energies to run the compute, to compute, run the algorithms, the algorithms then feed the applications. And so at the foundational layer is energy. And from that energy, we build the land power and shell, and that hosts the infrastructure, so on and so forth. And so the way we win,
[0:01:52] Of course, we are world class in chip design. We're world class in algorithm development. And so one of the things that we need to do is just be vigilant about enabling power delivery and power generation to happen. And President Trump has been visionary in being pro-energy. Without the pro-energy vision of President Trump, it would have been impossible for us to build the data centers on top of it. Now, of course, these data centers are coming online so quickly.
[0:02:22] We're coming into a lot of neighborhoods and a lot of communities. We have to do a better job in partnering with the communities and bring them along and make them part of the prosperity and the success of this industrial revolution.
[0:02:38] That's how we win. We've got to build the world's best chips. We have to have lots and lots of energy, create incredible algorithms. And then lastly, make sure that America is enthusiastic about the technology so that we diffuse it into all of these various industries, from manufacturing to healthcare, space exploration, so on and so forth. Awesome. Elon, how do we win? How do we keep winning? Well, we need a lot of power, a lot of energy generation.
[0:03:12] So scaling energy, I think, is one of the biggest things. You've got to scale energy. You've got to scale chip production. And I think we're winning on software, on anything intellectual or digital. But in the long run, we need to be able to generate enough power to compete with, I think, essentially China is the biggest competitor. And China has about...
[0:03:42] Three times the electricity production of the United States. And I think we'll figure out how to make shifts. So really the central challenge I think is long term is going to be how do you generate enough power and how do you ensure that we have enough logic and memory fabrication capacity in the United States or in a safe region. We need the Wright brothers on the job. Yes. Well, it turns out the Wright brothers are on the job. Elon, if you bring on...
[0:04:11] Thank you very much.
[0:04:41] And I think actually will materialize to be approximately 1% increase in GDP. Because the intelligence per watt keeps increasing. So if you just plot on a curve, how is the useful intelligence per watt increasing? Both at the hardware level, at Jensen's, GPUs keep getting better in terms of the compute per watt.
[0:05:11] And the algorithms also get better in getting more intelligence per watt. So that's maybe an interesting note. I would bet anyone that 1% increase in power usage corresponds to roughly 1% increase in GDP. And so 10 gigawatts would be a 2% increase in GDP. With the Rubens, I think. Well, let's do the simple math. So each gigawatt can probably generate about, call it,
[0:05:41] $40, $50 billion of economic output each year. Yeah, I mean, 1% would be, yeah, I mean, yeah. 1% of the United States would be... $300 billion. Yeah, $300 billion. And so 50, one-sixth. 1% of the, 1% increase, so 5 gigawatts would be, yeah, well, 5 gigawatts would be $300 billion. Yeah, right. Exactly, $60 billion a gigawatt. Yeah, that's right. The math, math. Yeah, yeah.
[0:06:11] That's what we're good at. Yes, yes. Not good at too many other things, but man. Yeah, I mean, it's going to increase quite dramatically year over year. So, yeah, I mean, from a SpaceX standpoint, we're looking at long-term, how do we access a lot more energy than could be accessed on the ground, and that's with space. So orbital compute is going to be a very big deal.
[0:06:44] In fact, and Google agrees. They just actually sent up or are either about to or soon about to send up some TPUs. I think there was an H100 that was put into orbit like last year, I think. So it does work in space. And SpaceX is aiming together with Tesla to do 200 gigawatts of solar production per year. And if brought to space,
[0:07:13] You get the nameplate or better on the solar power, whereas if it's on the ground, you're going to get somewhere between a fifth and an eighth of the solar power generation, and you need enormous batteries. Whereas it's essentially always sunny in space. And right now is a great time to do it because the economic returns of these computing systems are so good that the launch cost, although it's still quite high, in terms of the overall economic
[0:07:43] Benefit is quite affordable. Absolutely. This is the perfect time to do it. Putting NVIDIA Compute in space is the best way to get there. Yeah, climbing the Kardashev scale with American power and American chips on American rockets, I love it. So, Jensen, you said something that I thought was important about how when we bring the energy to all these small towns, we're re-industrializing America, we're revitalizing small towns, we're modernizing the grid. These are things that...
[0:08:12] Essentially, every American politician has wanted for the last 35 years. What would you say, what are some specifics you can give about how data centers are good for ordinary Americans in the communities and how they're getting better for those communities? Yeah, we used to call them data centers. We call them data centers because that's where we hold data.
[0:08:34] But the things that we're building now, these are really SI factories, super intelligence factories. And the reason why it's a factory is because you're producing something of economic value. You're just not storing it. And so these SI factories are real factories. It's like a big, giant power generator in real factories. It's got a lot of computing and networking in there, power generation, cooling systems. It's complicated things. And so we're building out, just as the internet today allows you to...
[0:09:02] Find anything, anywhere. In the future, with these AI, excuse me, SI infrastructure, SI factories, you'll be able to know anything and do anything from anywhere. And so, now we're building this infrastructure out, and call it across all of the United States, somewhere between 10, 20 gigawatts a year, let's just estimate. That creates just probably about a million jobs.
[0:09:29] In building out these data centers, don't forget you've got to build the power generation plants, you've got to build the data centers, you've got construction, you've got power cooling, you've got pipe fitters. And so this is the first time in probably, as you mentioned, 35, I was going to say 50 years, first time in 50 years, we are re-industrializing the United States, creating a whole bunch of blue-collar jobs. And I think we want an economy that has a shape.
[0:09:58] That includes white-collar jobs and blue-collar jobs, not just white-collar jobs. And so I think that this is our great chance of reshaping our economy and bringing that type of builders and craft back into the United States. Absolutely. And, I mean, it feels like what Colossus and MacroHard and MacroHarder have done have been amazing, the communities around them, the jobs they've created. How do you think about that?
[0:10:26] Yeah, the positive economic impact has been tremendous. I mean, there's actually a job shortage in the region, whereas there used to be quite a bit of unemployment, and now there's over-employment, if you will. There's more jobs than there are people. And the tax benefits for the community have been tremendous. I think we might have doubled their tax budget or something.
[0:10:55] Quite tremendous. We do small things that are helpful, like provide half-price Starlink access. We're building a water recycling plant that's a $250 million water recycling plant. I think you do need to do things for the community, and if you do so, they will reciprocate. Awesome. Now, one of the things that people forget is that for the first time in probably forever, that we have the market forces.
[0:11:24] to support the investment of sustainable energy. You give me an example of a fusion, fission, small nuclear reactor, large reactor. Almost every form of sustainable energy is being funded today. And so we have the opportunity to fund the future of energy without the government subsidizing. You give me an example of a great energy battery storage idea. It's going to get funded. Awesome.
[0:11:52] By the way, on behalf of all Americans, I want to thank both of you. I think combined, you have created more than $5 trillion of wealth for ordinary Americans in their 401ks, tens of thousands, maybe hundreds of thousands of jobs directly, many tens of billions of wages. So thank you. I think Jens has done $5 trillion by himself, actually.
[0:12:17] I'm sorry? I think Jensen's done 5 trillion by himself. I think I've done around 3.5 or something. I was trying to be conservative. So you guys both had a great Monday. I was watching the Starship launch, first orbital Starship launch, while I was reading about the open agent safety framework from NVIDIA, which was an exciting morning.
[0:12:44] So maybe you could each talk about each of those because we need Starship to get those 200 gigawatts of American solder with American chips, American superintelligence into orbit. I'll talk about his and you talk about mine. No, I'm just kidding. So let's see. As you know, we want to advance AI, advance SI. I'm going to get good at this. We're going to advance SI in a responsible and optimistic way.
[0:13:14] And we know that the future, we strongly believe that the future that SI is going to enable is the future that we want to be part of. And so we're optimistic about that. And that's the reason why we're driving so hard. On the other hand, we have to make sure that we develop this technology safely and securely. There are several different parts of it. While you're training the model, that's generally a fairly safe environment. But while you're evaluating the model, while you're testing the model, you have to make sure.
[0:13:41] That the agent, agentic system, the model itself, is highly contained. That containment system has to be really secure. And the second thing is that while you're testing it, you want to monitor that just because, even though you have it contained, you should never trust that it's contained, and so you want to monitor it in real time. And you want to ideally monitor it with another chip so that it is outside, if you will, called out of band.
[0:14:09] Outside the environment, outside the sandbox of the agent itself. And so you want to contain and you want to monitor. Once you're done evaluating in this way, when you deploy, you want to use a very similar system. And this is no different than just about any digital, any human employee that you hire. You give them a containment of rights and tools and access. And then you monitor on a regular basis, on a continuous basis. And so in the case of agents, we just have to make sure containment.
[0:14:39] And monitoring has to be done in the most secure of ways. And we introduced two technologies. One's called Open Shell. It's a containment system. It's basically kind of like a browser for websites. It's now essentially a browser of agents. And that browser keeps the agent isolated from the rest of the computer. And the rights that you give it are only the rights it should have. And then the monitoring system.
[0:15:06] is a new chip called Bluefield, and it keeps an eye on everything that the agent is doing at all times. And if it does anything that's out of bounds and against the policy that has been given, then it escalates as far as you need it to escalate. And I thought it was a great analogy in the technical blog where you said in the early days of Internet, everybody was worried about your credit card could get stolen. We made the Internet safe, and that's why we have Amazon, we have Netflix, we have Google.
[0:15:36] We have X. We have so many great things because we made the internet safe. Now we're doing the same with AI. And safety is not inconsistent with advancing the technology. Safety and advancing the technology is one and the same thing. It's no different than when you have autopilot. Not only is it advanced technology, it's also safer technology. And so you advance the technology in that way. The two are not in conflict. The idea that...
[0:16:04] If you advance capability, it somehow compromises safety. If you advance safety, somehow you fall behind. Those are false choices. Yeah, safety lets us accelerate. But Elon, if the Starship's going to be a big part of those 200 gigawatts, what can you tell us about Monday's flight, what you learned, how you're feeling about the program? I think the Starship is going great. We took Starship to orbit for the first time, and we deployed Starlink B3 satellites.
[0:16:33] Which are enormous satellites. I mean, these are satellites with a wingspan of 737. So, really enormous. In fact, it's the largest payload delivered to orbit since Skylab. So, now we expect Starship to ultimately, next year I think, probably reach a weekly or twice weekly cadence. We've got not Tom Brown here. I know him by his X handle. Welcome.
[0:17:02] Good to see you guys again. It was a long run. Yes, it was. My car took a wrong turn, so you guys got here faster than me. Impressive. I'm glad you made it. Elon, please continue. We're talking about how Starship is going to let us get to 200 gigs of American solar with American chips with American superintelligence. Starship is going to do a lot of things. It's intended to carry enough
[0:17:30] Payload to the moon and Mars to build a self-growing city on the moon and Mars. I think as a rough order of magnitude, you need about a million tons to deliver to the surface of the moon to create a self-growing city or a similar number for Mars. In addition, it will launch at least a few hundred gigawatts a year of AI compute. 200 gigawatts a year would be...
[0:17:57] A 40% increase annually in the United States total energy consumption. So it's quite a big number. And it might end up being a terawatt point A. So this is serious tonnage. Awesome. And Tom, Anthropic, your company, your co-founder of Anthropic, you announced Opus 5.5 in the last several days. And I thought maybe we'd go to the exponential that we are on. And I think it's helpful to ground.
[0:18:25] What can Opus 5.5, and maybe it's for everybody, what can Opus 5.5 do to make the lives of ordinary Americans better today? And where are we going to be in a year? Yeah, so Opus 5.5 is a model that I think just continues the normal progress that we've had of getting kind of more IQ points for each of these super intelligences with each model release.
[0:18:53] And so the main things that people will see if they use it now is that it can do a task that takes a longer time than they'd be able to delegate to a previous model. And so I think the things that people should see is just like as a normal American, if you want a task, if you want to actually delegate a task to it, it will be able to do that better is one thing. And then two is the thing that I think people notice less, which is just that there's going to be a huge amount of software being written by these models. And so all of the products that people use,
[0:19:23] are going to be getting higher quality, cheaper over time. And it'll have this effect of just like all of the tools that we use now were written by people doing engineering work. And now it's going to be written by both people and super intelligences doing much, much more engineering work than was possible before. Absolutely. But I think it's doing so much more than this. There's already multiple stories of AI.
[0:19:51] I'm sorry, superintelligence. Superintelligence finding a life-saving drug that can be used for a disease that it wasn't understood that it could safely treat. So it's saving lives. I think it's having profound impacts on education. But maybe each of you could talk. For ordinary Americans, what's superintelligence doing today and what's it going to be able to do in a year? I'm an ordinary American. And I use Claude to...
[0:20:21] Help me do research and instead of giving me just answers in most of my research, maybe there's comparative research that I'm doing and there's two or three different approaches to something and I wanted to answer a list of different questions and compare and contrast rates of growth or rates of change or whatever it is. And so I describe literally every single question I have.
[0:20:46] And nothing gives me more joy than... And I ask it to create a nice report and plot it out. Sometimes it's a pie chart, sometimes it's a bar graph. And I just hit enter. And nothing gives me more joy than just watching it go... And then it comes back and it creates a beautiful report. And then instead of researching all these different topics for four, six, eight hours, which back in the old days I would on a weekend, learn something new, it goes off and creates me this really, really beautiful...
[0:21:15] Final report in about three minutes. And, you know, it's all condensed and now you're getting smarter faster. Love it. What about for you? How are you seeing super intelligence helping ordinary Americans? What do you think it'll do in a year or two? Maybe a year or two, take a year or two from now. Yeah, well, I think it's helping people organize their lives and do better at their jobs. So I think it really will become...
[0:21:45] It's essential, if it isn't already, for people to use AI in their professional lives, certainly, and in their private lives. So AI is just an incredibly helpful, useful thing. I've heard many anecdotes of AI helping people with medical issues, and in many cases where the doctor was wrong and they submitted their...
[0:22:13] X-rays or MRI to AI, and AI got it right. But for AI, they would have died. And I've heard many such cases. Awesome. And hopefully Opus 56 tells Tom, don't take that left. Take the right. Hopefully. I think it's important these...
[0:22:36] These superintelligences, they're imbued with the values of their creators. And we do have, there are American values that we all share. The First Amendment, free speech, commitment to truth. Can you talk about what each of you are doing to imbue, help make sure that these superintelligences are imbued with the right values?
[0:23:02] Are we allowed to say anything controversial? Say something controversial. Okay, so the SIs are tools. And the challenge with an Si with values, meaning it has judgment, depending on how you use that Si, it may judge you, and its effectiveness would be affected by that judgment.
[0:23:31] And so it's a little bit like a steak knife where if you were served medium well, it cuts perfectly. But if you were served rare steak and it looks awfully like live meat, somehow it doesn't cut. And so I think there's a question in my mind yet about how much judgment you really want these tools to have. And so the intentions of the user.
[0:24:00] Obviously, I use the knife as an example, but you can think of a whole bunch of other examples. The intentions are a little bit different. Whether you're a defender or an attacker, for example, in cybersecurity, the defender's actions could resemble an attacker. And you want the defender, you want this SI to activate on behalf of the defender, even though it appears to be an aggressive action.
[0:24:30] Great. I think we're running up on time, but maybe if you guys could share, you spent a lot of time at the White House. Is there anything that you can share with the world about what came out of what was evidently a long and productive lunch? We did good. Yeah. We signed a joint declaration regarding SI safety. And I think that'll be published by the White House soon, if it hasn't been already.
[0:25:00] But we agreed to a number of things that include joint monitoring, board special committees, and just generally grading each other's homework, which is a lot better than if people just grade their own homework for obvious reasons. And it includes everything from rigorous internal controls.
[0:25:23] Well-established methods, well-established processes, very clearly documented intentions of the products that you're building, internal controls to make sure that you're delivering on those intentions, internal audit that escalates, external audits from a third party, and then industry sharing best practices. I thought the declaration has a lot of teeth. I thought it was really well written.
[0:25:53] And I think this is a practice that, quite frankly, as standards for our industry going forward, every single company ought to embrace. And so POTUS pulled it all together. He watched over the meeting and was able to bring everybody out. And I thought it was a really well run. Awesome. Well, on behalf of all Americans, thank all of you for doing everything you can to make the future as awesome and optimistic for American Americans.
[0:26:23] As you can and you are, so thank you.