And Twitter. I forgot Twitter. You were at Twitter.
There's a bunch. All right. So, let's start here. Every conversation about AI in Washington DC is about the quote race with China. How do you most simply define what that race is?
Okay. Simply put, the way me, David Sacks, the White House see this is we are in an existential race with China. And it is the American AI stack versus the Chinese AI stack. So to be very specific, it means our American AI chips. Uh you heard from Dr. Lisa recently or Jensen, our American models and our applications on top versus Huawei and Cambricon versus uh DeepSeek and their models on top. And look, if there is a race, I see this as a business strategy. We need to have a metric for winning. And for me, the metric is market share for American AI. If there is a token inference around the world, I want it to be on an American chip using an American model. I wish I could get a big TV screen in my office which just shows me how many tokens are inferenced and how many of them are on our chips and our models.
Is there a percentage that you think means we are winning? It's obviously not going to be a hundred. So what's what's that what's that inflection point?
Well, uh it's very hard to get an actual number right now for tokens, but right now we are dominating. And uh you know, for me, I I'm a student of tech history like you are. I think about Windows and Intel. A lot of people here in the 30s and 40s, if you look at the 90s, Windows and Intel was the dominant platform of software and hardware working together. So I want us to be the dominant platform when any token is inferenced all over the world.
When you think of this this race with China, is this for the United States, is this primarily an economic issue or a national security issue?
It's both. Um, and they're highly linked because obviously, look, it's obvious to everybody here the economic impact of AI and the flywheel that happens when AI leads to productivity and it makes your company's innovation better. But it also goes into national security because as you can tell these models and applications are at the heart of so much of defense tech and military applications and they wind up influencing each other. There's also a national security issue if you have models from our adversaries running inside our country or our allies countries, models which don't share our values. So for us it's a combination.
Is there any, you know, we talk the US China with this very binary. Is there any other country you see out there and think yeah they're they're the dark horse but they they could show up?
Uh I think look uh the way I see it is there could be but look I think a lot of countries want to work with America. Uh the conversations I've had with multiple heads of state with multiple ambassadors is look we want to use your GPUs. We want to use Nvidia. We want to use AMD. We want to use other companies. We want to use your model. So I think a lot of countries want to work with our companies and one of the things that we have done over the last eight months is to try and find ways to make it easier to do so.
Uh Dario from Anthropic was up here and he said he thought it was nuts that the United States allows chips, Nvidia chips for example to be sold into China, albeit older models of them. You're obviously part of White House that is allowing that at least until today when China stopped it. What's the counterargument to Dario?
Let me take a couple of minutes on this. So, so I think there were three mistaken beliefs that the Biden administration and a lot of these folks who don't want us to export chips believe. Number one is they thought America has a chip shortage. That's supply constraint. Second, they believe China can't manufacture chips. Third, they believe that China can't do open source models or models at large. Fourth, they believe that America is on a path to building AGI or ASI. AI models can kill us all. So, we need to retain those abilities. Working backwards, I would say that we are nowhere on the path of you know terminator or self-improving AI regardless of what Eliezer Yudkowsky says. Uh number two, if you see the news today, it is very clear that China has chip capability and in fact they're trying to stop our American companies from selling chips to China. So they are building out their Chinese AI stack. Third is you know there is no supply constraint like if you heard like Dr. Lisa or if you talk to Jensen, you know, on their earnings or other places, they will tell you that, you know, we have all the GPUs that we need for American companies. And so, uh, and of course, when it comes to Chinese models, my first day on the job, DeepSeek comes out and we are in a heated race. So all this leads us now to believe the way to compete with Huawei and Cambricon, the way to make sure we don't have a repeat of what happened with 5G is to make sure we give China these older GPUs in limited quantities. America has the latest and greatest in enough quantities to build these superclusters. But by giving them these older GPUs limited quantities, we make sure that Huawei and DeepSeek don't have a flywheel that's in group.
I mean, did Beijing call that bluff today though by the reports that they are no longer allowing Nvidia chips to be sold to their big tech companies?
Well, I think the way they see it is kind of it sort of affirms what Sacks and I have been saying, which is like they have enough chips. They want their stack to win. Um, there have been news reports of uh Huawei trying to export GPUs to uh other countries. So I see this as in a way of them doing an AI belt and road initiative and it is on us to make sure when it's our allies or uh it's the world over they are using our stuff and not their stuff.
It's interesting. I mean, if if Beijing kind of agrees with your point of view on this, which is why they're doing that as you say, does that give you kind of concern because they are indeed going to be keep plugging more res, you know, Huawei is going to be able to sell to their companies. They get more money, they get to build better chips. In other words, is is what China did today not allowing US chips in, is that actually bad for the United States in this race?
Well, I think if we can't, our companies can't get developers and applications, if there is not GitHub repos which are writing code on American chips, that's obviously bad for us. But I think, you know, I want to go back to some of the people who are opposed to us exporting GPUs of any kind. I always ask them what does Huawei want us to do, right? Like they probably want uh us to not ship them AMD, not ship them Nvidia. If we if people are advocating the same policies that Huawei wants us to do, I probably have some questions about that.
Uh let me so if if one of the goals of you and Sacks and and the administration is to build up the as you say the US AI industry the stack etc. How does it help US chip makers to have to pay a cut to the US government when they're exporting to China?
Well the way we see it is uh look first uh this there's been kind of a lot of this fake news about what what has happened. Number one is what's happened right? So the Biden uh administration never actually export controlled the H20s or the MI380s from AMD. Uh we brought this under a licensing regime, we being you know the Department of uh Commerce, where we said we want to know exactly where the chips are going and in what quantity.
And when you sell them, we get a we get off the top.
And I think that you know um I saw uh the Secretary of Commerce uh talk about this the other day. Look, the president does what he does best which is like look we are letting you do this, you know, let's make sure the.
How does that but how does that help the industry if the industry like in theory take your Huawei example right? The the reason you don't want Huawei to be able to sell as much because then they get to it becomes a virtuous cycle for them right? They get more money to develop. If Nvidia could get that more money to develop, that would be good for Nvidia. That would be good for the US industry. But they can't because they're giving some of it to the IRS.