Aaron Ginn: America should flood China with NVIDIA chips to win the AI trade war

intelligent model yet. State-of-the-art reasoning, next level vibe coding, and deep multimodal understanding. We have Aaron Gin, the CEO and co-founder of HydroHost. Good to see you. It's been too long. How you doing?

Welcome back.

Doing all right. Yeah. Yeah. I just got back from Southeast Asia.

Oh, yeah.

Well, yeah. And uh despite what some of the critics were were saying, I wasn't I wasn't smuggling anything I shouldn't have been smuggling.

[laughter]

Are you getting accused of smuggling? Um, that is wild. Well, uh, give us the, uh, give us the update on, uh, your interpretation of the H200 deal, where things landed, where you think the next, uh, debate points will be. I'd love to hear that.

Yeah. Like the I mean, as you know, I've said a number of times that the even though I am a China hawk, that we have to realize that Trump fundamentally is a pragmatist.

Sure. and he is not interested in the ideological warfare and I think that him taking over the Republican party proves that and the orientation of restoring to America which we all know like I wrote one of the first arguments for a new Monroe doctrine so now you saw that the department of war has been putting that out the white house has been very clear that it's trying to reshore its emphasis back to uh left in South America and to build up the ring of fire from you know South Korea down Singapore essentially uh to be more sovereign essentially NATO as we've approached ourselves with uh having them uh spend their own money to support their own military that the expectation is that Asia would do the same thing. Um, so a deal with China on trade was always on the table and was very likely because both parties actually want this to happen and and I I do believe that with uh like with Huawei that that yes, it is the national champion and they do want to proliferate it, but they are but they're more interested in selling it to other people and promoting themselves as a means of trade like they did with the Belt Road initiative

than they are for internal uses like that. That's just kind of my uh estimation talking to people in the region and I mean just came back from there. So So Nvidia being in in that region and being sold particularly China I think is important for American interests whether it's an H20 or B30 or H200 I'm kind of less opinionated on. I just think that we need to be there because half of all AI engineers are Chinese and and we can go through any list of anyone scrolling at the bottom of this uh you know this this broadcast and you can see that the probably the majority are Asian descent you know minimally and and as America we just have to realize that like we're good at certain things and we're not good at certain things. So, we're very good at commerce. We're very good at free markets. We're very good at enterprise. We're very good at creating the default framework, the rules of the road for the world to transact. I I think when it comes to trying to do these other strategies, I think we're we're not very good at it. And we just kind of have to realize that also that the president of the United States is interested in making deals. He's not interested in trying to get involved in other people's business.

Yeah. Uh let's I mean there's like a million different pieces here, but uh let's start with uh uh the fact that there are so many fantastic genius AI researchers in China. Uh why is it in the why is it in the interest of the United States to help them? Is it that if they start developing on Nvidia chips, it gets the e the Kudo ecosystem stronger and we will actually get more open- source software basically for free from China that will come back to America and make our AI efforts stronger or is it we want them to be familiar with Nvidia when ultimately they come over here on 01 visas and start working for American companies?

I I think that that's too narrow focused. It's sort of like Boeing. Do you want Boeing in China?

I mean, yes, because I want Boeing to be a thriving American business. I want Boeing to maximize profits, and I don't see the the 747 as a piece of I don't I don't see this the 747 as a nuclear bomb. the Yeah, I mean I mean I think that that that's the key brier on the second one is like do you think that

an H2 H200 is a is a I don't know like a like a some form of missile technology.

Yes. Yes. Yes. Increasingly I increasingly I I I have I have backed off the idea that that uh these chips should be viewed as uh as weapons

like a defense technology.

Yes. I backed off and

yeah because I view it much more like telco technology. I agree.

And and and like and we did not win at that. We lost that. And and that was because we didn't offer the world an alternative. We just were like, "Hey, don't use don't use Huawei."

Yeah.

And then it's like and they were like, "Okay, well, give us something else." And we were like, "Well, I don't have anything to give you."

Right. Well, so what are they?

We're we're sort of separating things out here because on on the one hand like there's give the rest of the world something and then there's other which is like give China something which is like there is a difference there, right?

Uh

yeah. No, no, no, that's true. I I'm totally down to go belt for belt and road for road all over the all over the globe, right? Uh but the question of like can like is there any world where we could get a different cell phone tower deployed in Beijing that's not a Huawei tower? Like no way, right? It's just impossible.

But but but then you go back to the fact of like you're right like is you know all metaphors fail. That's why they're metaphors. So, so the the the other exogenous factors you got to think through that's different from Telos's is the fact of like the level of AI engineering talent and fact like it's funding in as in like you know propelling forward American AI. That's just the reality of of where we are. The the second is the fact of like they do not have the foundry capacity or the foundry skill set to produce similarly level at scale. Uh scale absolutely is true. they could be at some point uh be approaching some form of like five nanome process and there's like rumors around that. So that's a little bit of an obsession because it's it is a barrier of technology that like we uh isolate the west has been skilled at not particularly America, Japan and the Dutch but but that's there we don't want to fund that enterprise.

Y

and I think that that's where the pragmatism comes into play. It's like if we don't offer them something, they'll just do it themselves. And I'd rather take money from them and build our own sort of re resuring in America.

But but to be clear but to be clear, they're still going to do it everything themselves,

right? It's just going to take them

how how does

I don't think so.

You think that China will happily be dependent on foreign product.

I mean, last time there were rumors of this sale, they they were putting out press releases saying like, "Hey, if you're gonna buy Nvidia, you got to prove it. you you got to prove that you need it. You got to prove that Huawei won't work. It felt like there was a lot of push back from the CC. I don't believe buying NVIDIA.

I don't believe at all that they're just going to be like, "Thank you for the H200's. We're going to stop investment in our domestic uh chip uh manufacturing."

Yeah. So, but so but but the the problem like in that flow one, you assume a level of CCB control that's just not realistic. Sure. that that that it's it's a country of like 1.3 billion people with very successful entrepreneurs.

Yeah. But they've been ripping five year they've been ripping fiveyear plans for over and over and over and over.

I mean like like Yeah. But but that stuff doesn't matter. Like you have to understand that like that there's lots of propaganda that think that they say that they actually never happens in real world. Go look at Africa. Go look at all the airports and see actually how what actually happened. Go look at the ports. Yes, there are certain example of ports actually being transferred like the Sri Lanka port, but go look at the rest of ones like they didn't happen. So, so they say lots of things, but the question is like what matters in terms of the national security orientation of America. And I would much rather take money from their economy and put it back into our economy than trying to give them both the demand side and the supply side argument. You're you're making a supply side argument, which I think is very much how CCB thinks about the world, which is like, oh, we need to build our own chip supply. But in the orientation, I think that Jordy, you're more biased towards, which is like, well, let's also give them the demand argument. I'm like, no, I don't want to give them the demand argument. like like I I I I would rather mitigate that as much as possible to say that yes like they have uh you know probably like five or six multi-billion dollar operations that like want to have these types of chips that are non-military oriented and by applying these export controls that are overly extreme you're basically giving them a demand argument as well and I'm saying I don't want the demand argument like they're still gonna have a supply argument that's fine like go buy a BYD you know car and hopefully it doesn't catch on fire like so like you know like like just Because by

Are you Are you Are you bearish on BYD? You're You're They're catching strays right now.

Uh I saw one of them on fire in Hong Kong. No joke. [laughter]

Yeah, I have video of it. I texted you.

Wait, yeah, let's go let's go to the cars and talk about what we were debating earlier. Um, if if you're if we are just talking about cars instead of chips, of course there's going to be a metaphor here, but if we're just talking about cars and we want and and the frame is like we're we're all China hawks and we're putting on like rah rah, we want America to win the electric car race. Should Tesla have sold Teslas into China? because it feels like they did and then China was able to make the a whole bunch of cars that compete directly with Tesla and are it feels like maybe better maybe cheaper maybe better and cheaper on some sort of paro frontier. Um, if if we run that experiment, we try and think through the counterfactual. If Tesla had never sold a car in China, where would the the the Sous7 be, the jumping Huawei car, the Xiaomi car, the BYD Han, the one that has the karaoke machine in it? Like there's some crazy cars. Would those exist? Would they be delayed? Would they be would they happen faster? How how do you think about that?

Yeah. So, so I I think this is where there is a reality, right, which is that they they're going to copy and but I don't think that that is something we can change. Yeah. Speaking as a half Chinese American. So, so it's just the nature of what what is the way they think about

competition. It's just competition. You're going to want

Yeah. They view competition as copying. And and importantly like like like the there there's one world where you know they need the industrial power of like 1 million H200's and there's another one where if you give them a single H200 they can tear it down and like try and reverse engineer it as much as possible and then they can also do all the learning curve stuff on their on their companies on their national champions on Huawei. I I think it's it is reasonable in terms of an argument against expert controls if you view this as a means of trying to slow down their own model level progress

like like I can at least cogently understand what's being presented. I do not agree with this from the perspective of hardware engineering

because again if you don't have the ability to make a car

Yeah. then then what's the risk, right? like like like just mitigate their ability to make a car

which has generally been my my my position is like I I want to mitigate their ability to make chips and and and so in a world where there's a demand and supply right they in SECP because they're they're communists so they loves focusing on supply solutions to everything which by the way doesn't work as I mean we're all free marketers here we're all a biased towards Austrian like we know that their philosophy of the world doesn't work so so they they can talk about supply side engineering all they want but again they they don't have the ability right now to aggressively address their levaker problems and so in that world I want to just win because if they can't if if we remove the demand stuff because all other companies want to use Nvidia

if we remove the demand stuff we're removing one of the biggest parts the uh that would give them justification to accelerate on the SMIC side but if they're just going to go and play around and be like hey we're going to dump money into SMIC because we want you local but we're everywhere It's it's like the fast food thing of like sure they have local fast food but McDonald's and Starbucks crush and and so like that that that's more how I think about it. They're still going to try to copy, but if you don't give them the ability to actually have the toolkit, then it's a much safer transaction. And I think I think I think John what what you said is is important is that the the debate is highly uh mixed up between you think an H200 is an F-35. If you remove that as an a prior then the world becomes significantly simpler to understand. But if you view what you know the the the H2 under black wall or whatever as a as a patriot missile then then like there's no real place to like have a conversation to be rational.

Yeah.

Yeah. Because my view is like you know I don't have as much of an opinion on you know GB or B

that kind of like debate. I I just I think American companies need to be in China because and because we want the ability to not only address the second uh largest economy in the world and like have access to like that is the Trump position and that's my position like we should sell to the world but we should be an export nation

but but the other is like China is going to coffee and it's just part of the equation. Flip flip around for me and give me the sorry

uh I I wanted to hear flipping I I wanted to hear more shifting to domestic semiconductor supply chains. Uh what is your theory of the current horse race between uh tranium TPU and Nvidia? It felt like Nvidia was dominant last year the year before. Then we were hearing rumblings of like hey maybe AMD is getting it together on the software side. Hey, maybe this TPU thing could be used to train a decent model. The Deep Mind team seems to have been able to do that. Uh, are we moving towards more of an oligopoly? What are you thinking about how the the you know American semiconductor AI accelerator horse race is playing out?

Uh, my my view from the the on the TPU side, I don't really know who would buy that as like just like a product. I I think Meta is the closest one that kind of makes sense, but why would any of the cloud companies do it? Because they are in a position right now trying to build their own and they're trying to dele why they're building. They're trying to delever from Nvidia. So why would they want to omingle themselves with an actual direct comp, you know, like a a direct competitor and sell their stuff and and then there was [laughter]

Yeah. Like like there's there's also this like weird zealatry around TPUs. I may cuz we had Thanksgiving there. It's not going to happen. But to say that like there's like minimal switching costs is absolutely absurd and insane. There's significant switching costs and and so if you're of the mindset of like hey well we have all this capex and it kind of makes sense and I that's justifiable like I can understand that but that's like the same reasoning that you could apply to uh the multiloud argument and we didn't really see that happen at all

uh basically until GPUs. Now, now you're seeing much more multicloud products because GPUs made everyone multicloud. So, and if you were multicloud before GPUs,

you were as a function of an acquisition, it was not generally intentional.

So, so I think TPUs makes sense for Google just as a like vertically integrated strategy because that's what they like to do. Y

and and like there probably some other companies that maybe like maybe Apple I could maybe see that because they love lowc cost chips just in general. Um, but I don't really see it breaking beyond Google because I just don't know who is really incentivized to buy that. Uh, and then I agree with you on AMD side. Their software is getting better. I think that's just the natural um orientation of of just they they like Lisu is the most reasonable you could say competitor to Nvidia's general purpose computing dominance. Um, I'm pretty bearish on tranium uh and uh TPUs just from a from a chip perspective I'm bearish on. But as an integrated strategy, it like it makes sense for them as a company. Nvidia, that's not really it as

we're running out of time, but I wanted to ask you about one other thing. Uh there uh seems to be a number of people trying to manu manufacture a space data center pump.

Oh yeah.

As somebody uh and even Nvidia was posting from their corporate account uh uh uh some renders of a company called StarCloud, some space data centers that they're working on. As somebody that has spent a lot of time uh in data centers or around data centers all over the world, uh what are your timelines? Are we talking uh Gavin Baker was saying uh the most important thing in the next 3 to four years is data centers in space? Had plenty of people on the show that believe that it's an exciting possibility or opportunity, but maybe their timelines are more like 10 to 20 years. How do you think about it?

Uh most of the stuff that has been announced will not happen. I there'll be

uh no like then think of it as like a 50%. So if they announce like 5 gawatt probably going to ding like two and and because this stuff's really building this stuff is actually quite hard and America has not really behaved in a manner to me that believes it's that serious of a of a thing like this is a very niche issue and and one reason why the the going by the H200 the reason why it's making less noise now is because it is it is a niche issue. It is something that maybe a couple hundred thousand people really care about and and the data center it affects even though it does bring jobs it does bring the these types of opportunities to uh cities that previously never had a position to support a data center which is great it is really kind of a narrow set of people that are really benefiting uh and if you go look at uh you can go read the bills that are in consideration across America I it the America is not acting like it's very serious about building day centers the power profile files, the approvals that are happening, like it's it's just not happening versus the other countries, they they're generally more serious about it. Um, and many of these announcements that you as as Jensen disclosed, they they're they're fundraising announcements. They're actually not actual projects yet.

So, so they're trying to let's say they have an equity already lined up, but but the equity generally is like 10 to 15% of these projects. So, you got to go raise the other portion of it. and and the offtake side outside of the hyperscalers there's not a lot of confidence the credit side has outside of the the those hyperscalers. So so the so the debt side of the world actually has been quite conservative on the these these and the private credit. So that's why I never really thought like you know the quotequote bubble talk and bursting it it like we're nowhere really there if you go talk to lenders like we're nowhere even close. uh the type of capital that's being deployed has actually been very conservative has been aligned with contracts has been associated with high paying credit you know top tier customers we're not seeing you know the Jordy credit union in the middle of uh you know SoCal doing black you know B300 loans and that happens then I'll be like yeah you know

working on the like with the so so so [clears throat] that's that that's where if you look at the legal side if you look at the power profiles If you look at what's happening and the political nature of it, I I don't think many of these projects actually scale to what they need to. But I do agree with Gavin that like we actually do need to get serious about it because it's not like the world would be worse off with more power, more comput capacity. If everything all of all those things get cheaper, our entire life gets better.

And just as it's cheaper today to get a banana than it was 100 years ago, and that means there's more bananas. That means people consume more bananas. So, so the logic will still apply.

Let's give it up for bananas.

We love banana.

I'll give you a layup for that.

I love banana.

Uh, great to see you, Aaron.

Software is eating software is eating bananas. Thank you so much for coming on the show. We We always We shouldn't have booked 15 minutes. We should have booked two hours cuz it's always a fun chat. Uh, thank you so much for taking the time. We'll talk to you soon. Have a great rest of your day. Merry Christmas.

Merry Christmas.

Merry Christmas to graphite.dev. Code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. What's under the tree this Christmas? It's graphite.dev, baby. Put it in your It's a great stocking stuffer. Um, up next we have Matt Callish, the co-founder of DraftKings. He's a board member at