Aaron Ginn on the Moral Idea Maze, China, NVIDIA and GPU Revolution

just read a biography. I think those tend to be much better at understanding the philosophy of a dictator or an authoritarian because you got to get into the psyche, right? Because it's not just like you can read abstractions and reviews here and judgments, but if you understand the psyche, you can easily more pick it up in terms of how it appears in other parts of the world. So China can only get access to certain types of Nvidia products. So if the government is already stepping in being like, These things are incredibly valuable and they create things that could be threats to national security. That's kind of next level in terms of innovation that we don't fully understand how it's going to flush out. Right. And these chips are so powerful and it's such a jump beyond CPUs that just nobody really paid attention to, that the applications I think are going to be much more creative. I mean, certainly that's where Nvidia is going with this, is they want GPUs used in healthcare and real estate and Finance and construction. They want to spread their wings and go way beyond AI.

Today, Aaron Gin joins me on Upstream. Aaron is an entrepreneur and technologist whose experience has bridged tech, business, and politics. He believes that tech has woken up to its ability to change society. In this conversation ahead, I talk to Aaron about intellectual diversity in tech, how the GPU revolution is a seismic change to infrastructure, and how it impacts foreign policy. Here's Aaron. Aaron, welcome to Upstream. Thanks for joining the podcast.

Yeah, thanks for having me, Erik.

Aaron, we're here to talk about GPUs. We're here to talk about China. Why don't we first give a little bit of an introduction to Your career spans both politics and tech and business. Why don't you give a little bit of background on when you look back at your career, what are the threads that you've kept pulling that have led you to start Hydra? Then let's get into Hydra.

Yeah, so I would say that I have much more of an eclectic background than the average tech entrepreneur. So my scope of my network expands Both San Francisco and Washington, DC. And I've always been interested in technology from the perspective of building companies is like public art to me. It's like an expression of an idea of what the future can look like. And obviously that intersects politics because politics fundamentally in Greek just means the way something should be or where you should live. And I think that for a while when I first was working in tech, There was a little bit of, which I know you know about, a little naivete about how politics actually intersects with technology as if it was completely unrelated and that the San Francisco people didn't need politicians, didn't need to be concerned about first principles or core ideas about governance. And I think post-Obama, obviously, they started changing as the intersection into actual how society works beyond just hey, we sent emails and you can Google stuff into systems that were actually alternatives to what the government already regulated, began to sort of break down those walls. And so I was already kind of really ahead of that because I saw sort of where this naturally was going as technology was expanding so rapidly that eventually it was going to start touching into more and more areas that are deemed as untouchables in terms of the political space. Right. And I was already crossing kind of both of those things just due to personal interests. And most notably, I was out front in technology working on the Mehrani campaign, but I've also been involved in a number of other governor races, Senate races, House races, other presidentials. And it's just kind of an unfortunate interest. I think maybe 5 years ago, 8 years ago, VCs would be like, no, you should be more interested in making money. Now I think that There's, as you know, a huge split within the VC community that there's those that believe that their venture is oriented towards societal change and those who don't. And so now, I get less of that because they're looking to actually fund companies that align to certain metaphysical directions that they view creates higher returns. And that actually the metaphysical concerns that go into the purpose of the business, what the founders believe, do, act, behave, form into whether or not it's going to reach that actual impact that the venture investor wants. And I think now actually there's been— when I first started working in tech 12 years ago, I would say it was much more libertarian, amoral, I think pretty naive about politics and how morality faces and ethics faces into technology. Now I actually think it's actually very high on the list, both in the sense of things I agree with and disagree with. That actually what you do and then what you build is much more related to the end outcome you're trying to achieve as a company, whether that's actually— whether or not something like someone actually wants to invest in or not invest in. So I decided to take the route of already kind of intersecting with that like 12 years ago and actually worked on the campaign, actually involved with community stuff, DC stuff, local stuff. Within California and Colorado and a bunch of other states. And on the technology side, I was working everything from social media to e-com to— I started other businesses that were not venture-backed. And the most recent company, Hydra, was focused on this area of distributed agnostic computing. And at first we were focusing on how this actually was going to be achieved from a virtualization perspective, from a cloud perspective. And then we drifted down deeper into the stack, finding other areas that needed to be innovated upon that were not virtualization, which is generally low margin, very high fixed costs to try to get the low cost within the virtualization model. We decided to go the direction of bare metal, particularly around GPUs, and that was obviously the machines are much more expensive. There's a scarcity, and people don't really need as much virtualization as they do within the CPU world. And there's a whole other world sort of growing within that space that needed the ability to be agnostic and also to be able to access different types of resources around the world. So that's what led us to building the software we did, which is kind of a combination of VMware plus Shopify. When I was helping run a fashion brand, I never thought I would actually be building alternative VMware. But VMware, from a provisioning management of metal perspective, and actually running everything from networking, security, user access, things like that, to actually manage an actual physical box in a location. And on the front-end side, we built a single turnkey solution to sell that GPU, whether it is on a private basis or on a public market basis, and be able to set pricing, scheduling, access, revoking access, payments, billing, invoicing, kind of put all that together to where it's, uh, one cohesive unit.

I, I want us to go through the, the idea maze on, on a couple fronts. So one, I, I want to ask about Hydra idea maze and how you sort of thought about that as like, you know, where the highest leverage is and how you should spend your time. But first, actually, I'm curious on the, the moral idea maze that you mentioned where you sort of transitioned, you know, from sort of the libertarian views and aesthetic to do something a bit more right. Why don't you talk about that journey? Because it's a journey that a lot of people make.

There was something that happened in the '90s that sort of thought that like, if we just teach these kids to make money, everything will just work out, right? And that was like the training in high school, training in college was like secularism is this sort of détente and an armistice or something, that we just are amoralists and everything will be great, uh, which obviously was a red herring because it's not true that the idea of secularism is based on morality. It's based on things about, like, within ethics. That is not the lack of it. It actually is an argument from ethics and morals that you can be secular, right? So it's not an absence, it's actually the inclusion of statements about what life is, like, you know, objective statements about people and how you should treat people that creates the ability to be secular. And that obviously failed. The whole experiment failed. The great expression of that is wokeism, that something filled the void. And instead of framing it from, here are the presuppositions that require secular transactions that creates the market mechanism, we just sort of removed the upper story and just only took the lower story. And then that created a sweeping in of another higher story, right, that actually is eroding the lower story, to use a sort of Thomas Sowell framework. Uh, and, and I think that that's one of the transitions that, like, I saw whenever, uh, so the book that, like, changed my perspective on politics was when I was 17, uh, and I was being a Christian when I was 17, uh, was The Terrible Truth About Liberals, written by a libertarian. And it actually was not a very libertarian book, but what it actually argued was how people talk. It was about being a polemic. So how do you actually present arguments, how people talk about politics? And that was actually really eye-opening to me about what actually is being discussed, what's not being discussed. And as you know, as a Girardian expert, that's also a very good investing philosophy. Fundamentally, what we're doing is we're building a company, an enterprise, to create social change, right? And that social change is both in the function of things that I would agree with, right? Which is like providing value, either reducing costs or innovating. But also people take that money and do something completely nonproductive with it, right? And do social change that way, right? But fundamentally, you're creating these vehicles of value creation and value transaction. And I think that in, say, Web 2.0 maybe, or Web 3.0, I don't know. There was kind of a moralist perspective about technology, and that then filled itself with, I think, very moral-filled— and I use that not from a subjective but an objective meaning that they believe themselves moral, obviously disagree with a lot of things they're doing. And that's what created that because it filled it with this sort of, I would call it, an atheistic puritanical rage. and created a conflict point with the rest of society and the globe. But if we would've just created an upper story that had the values that continue to support free markets and human flourishing, I don't think that would've ever happened. I don't really believe that angle of entrepreneurship or businesses. I don't really think anyone— it's valid anymore. But that was like, if you were an '80s baby or '90s baby, that was what you were taught in high school and in college. Now it's not. But, but that— I think that that was the biases that went into, um, the technology community that, uh, that, you know, has now had a massive split in it between, you know, sort of 2 diametrically opposed worldviews.

When you think about the worldview that is not wokeness, do you think for you, is that more traditional religion, or is that Is it, you know, looking towards the past, or is it something else? Because I think that's a broad question, is like, do we go back to what, what we, we knew but is perhaps discredited by some, or do we go towards something new? But if so, you know, what, what is that thing?

Yeah, I, I mean, I'm, I'm a big believer in C.S. Lewis's chronological snobbery. So, you know, the Ten Commandments are pretty dope, uh, you know, and And reinventing the wheel is not necessarily something that I'm interested in, but I do think that the application of those values obviously have some sort of subjective relevance to it, right? Some judgment that's applied to it on how we sort of go forward next, right? But I firmly believe in innovation, and I firmly believe in progress in the human experiment. And sort of how faith fits within that. You mean more specifically morality. I do believe that what was presented in a lot of the founding fathers stuff, which was radically innovative at the time, is still the relevant code that we should follow today because it has worked, obviously. We are the dominant force in the world. We still are. And we have competition, but we still are number one. And been the only country in human existence, in all of human history, to create a multiethnic experiment, multi-regional, multiethnic experiment that has been pretty successful. There are lots of other countries that have tried this that absolutely failed. And as in modern experiments, historically, it just was never really true in terms of human civilization to have something like America. And so I believe that those work and we should just continue to reinforce those. there may be a refreshing about how we land the plane. And I think that's where you see these people, both good and bad. I'm not endorsing any of these people, but Jordan Peterson, Rogan, Tate, different voices on Twitter that are very popular, like Walsh and Jeremy and kind of like the newer age versions of the Ann Cultures of the world. And I think that's how we sort of continue to continue the foundational principles, the first principles, But then apply the fresh element. There are new things I think that we should actually incorporate. I think a lot of the elements around Girard, as a big fan of Girard, should be taught in school to teach people that victimization and the cancer that is victimization is only cured with scapegoat theory and mimetic theory, showing that actually this is a product of control and power. It is not actually true. And scapegoating and understanding how that's the natural desire of all of humanity to do that, I think, is a cure to a lot of those pathologies. Not just saying you're not a victim, it's actually understanding the metaphysical psychological framework that goes into those things that makes people into where they feel like they're helpless. And therefore, it's like we're in this upside-down world where the victims are the abusers. And almost like real victims are forgotten. And I think that that's where you have to show that the human desire and human heart is drawn to these things, and that's what we have to correct. And there are recent thinkers that are really good at that, obviously Girard, that can showcase how there's a different path. And the path of that is residing in forgiveness and grace and not in a power struggle right on identity category.

Hey, we'll continue our interview in a moment after a word from our sponsors. Looking to give your startup a competitive edge? Meet Mercury. Mercury is banking for ambitious companies, a modern partner that can provide you with the tools and resources you need to turn your startup into the best version of itself. Say goodbye to the friction that comes with traditional banking. Mercury moves at the speed of startups. From creating an account to wiring money, a few clicks is all it takes. Mercury isn't just a place to hold and send money. It's software built to help you scale with safety and stability when you need to, whether you're a team of 2 or a team of 1,000. And Mercury goes beyond banking to remove the roadblocks to success, providing you with the connections, network, and guidance necessary to make your ambitions real. As a startup and Mercury customer myself, I've loved how easy it has been to scale my company's finances with Mercury. Visit mercury.com to join more than 100,000 startups on Mercury, the powerful and intuitive way for ambitious companies to bank. That's mercury.com. Mercury is a financial technology company, not a bank. Banking services provided by Choice Financial Group and Evolve Bank and Trust, members FDIC.

Compliance doesn't have to be complicated. In fact, with Vanta, it can be super simple. Trusted by over 5,000 fast-growing companies like Chili Piper, Patch, Gusto, and Juniper, Vanta automates the pricey, time-consuming process of prepping, For SOC 2, ISO 27001, HIPAA, and more. With Vanta, you can save up to 400 hours and 85% of costs. Vanta scales with your business, helping you successfully enter new markets, land bigger deals, and earn customer loyalty. And bonus, Upstream listeners get $1,000 off Vanta. Just go to vanta.com/upstream. That's vanta.com/upstream.

Totally. And as we see this sort of conflict between 2 worldviews, do you expect that the, the business of this increasing polarization will thus mean there's a left-wing X and a right-wing X? And by X, I mean the I don't mean Twitter, I mean literally everything. How do you think this will happen?

How it will flush out?

Yeah. Will we have a right-wing parallel economy for everything?

No, I don't think so because I don't think the other side works. I don't think it actually produces prosperity and healthiness in a civilization. I think it produces destruction. And my family fled communism and real communism, right? Not the type that works at Starbucks and wears a hat, right? Like real communism. And so my family has seen what it's done. Murder, stealing, theft, right? The abuse of authorities, no human rights. And I don't believe that is a sustainable solution to anything. So I always want to bet on the good values and the principles because that's the only way I'm going to win. And I just want companies to behave too. And I think that that's where there is some legitimacy to, I think, the alternative economy when it comes to, I think, maybe non-moat-oriented companies like brands and things like that. I think that's totally fine, like Bulletproof Coffee. I think that's great. But there are things that are in the alternative economy that are just things that you should have, like gun companies. Gun companies are alternative economy, but it's like, Are there left-wing companies I don't know about? Right. So I think that the areas that I would agree with it would be endorsing brands, endorsing companies, because that's your right to choose as a consumer. Right. But I think on the technology side, there's less application around that because those tend to be more parabolic outcomes. And that's where that doesn't necessarily apply. I just want those companies to be more You know, behave better, right?

Yeah, it'll be interesting to see how this all plays out now that there's more intellectual diversity within the tech ecosystem. And you were really early to figuring out that that will be important, that it'll be important to have a view sort of on what the company is trying to do, and that politics will play a role in these companies. You can't just ignore it. You know, you may not be interested in politics, but politics is interested in you. As the, as the saying goes.

Yeah. I mean, they tax you. So they fundamentally are, right? Yeah.

They tax you. They regulate you. They determine what, you know, what you should be saying, et cetera. Let's segue into GPUs. Why don't we talk about NVIDIA versus the public cloud? Why don't you sort of give a bit of an overview there?

Obviously, this is one of the most interesting times relates to like infrastructure. And, you know, there's like a kind of a colloquial joke that like investors like you are buying Nvidia chips in terms of the companies that they're investing in. And I think that that's somewhat true and somewhat not, because it's true that this is the biggest infrastructure change we've seen in 15 years, right? And it's actually touching at something that will affect the entire internet, from devices to laptops to companies where they host. Because I'm a big believer that GPUs is basically what we always waited with quantum, and GPUs is happening, which is just Simply put, like parallelizing computing, right? And that this wave is going to continue. And it's just like that the brand around ChatGPT is what kind of sparked it, right? To say, oh wait, there's another chip out there that we can buy, right? Because NVIDIA has been around for a while, right? There's a chip we can buy that we can use to do different things with, right? So the public cloud has built an entire business model on low rent, low cost CPUs, right? And they've invented their own chips, their own CPU chips. They've designed a whole networking system around non-parallelized computing, and they have a whole cost energy model, cooling model that's only designed for CPUs, right? GPUs are like, it's like if they have a CPU, like I was just using car analogy, like a CPU is like a Skoda, right? Let's use a Volkswagen Group example, like a Skoda, right? That'd be like the nicest CPU you can get, right? The average GPU that you'll just buy from Nvidia is minimally like an A3 Audi, right? So the difference is we're talking about for like a mega data center in Virginia that is only optimized for that thing, right? And now jumping into supporting Audis is like a huge difference. Now you're talking about like DGXs, right? You're talking about like Lamborghinis and Porsches, right? That you can't put that in a Virginia mega data center. Like the energy needed to run an H100 versus like average CPU is like magnitudes difference, right? sometimes 10 to 30 times greater, right? Then you have the cooling problems, right? And then you have the fact of you're dealing with a box that's like half a million dollars versus a box that's like $2,000, which would be like a really nice CPU, right? Public cloud is not ready for this, right? And then you then add the other layer to it of the physical area, public cloud's not ready. And then you get this other area of the networking side, which is the real value behind a lot of GPUs versus how you actually interconnect them all. How does public cloud sell that? How do they sell that? They don't sell racks. And then now add not just racks, but also the interconnections between those. And that the reason why they get the cost so low within S3 or whatever Amazon service you're using, whatever buckets you're using, is because they control the networking absolutely. And so now you got this other factor coming in now with GPUs coming in, Where it's like, what gives the advantage a lot to GPUs is you give over networking, right? Where the companies have access to it and going to do whatever they want with it within the rack. And so I think those are things that are breaking apart between public cloud, where public cloud will be a thing, just like IBM's a thing, right? But the next wave of cloud and virtualization and infrastructure is basically going to be, I believe, highly modularized, highly regionalized GPU-focused data centers. Like, is it data centers are focused only on cooling, data centers attached to green energy like nuclear power plants, geothermal, like data centers that have like their data centers in the Nordic countries that the towns will pay the heat from the data center into the town, right? So that is where GPU hosting is going and public cloud is not ready for that. Like they're still focused on their big mega data centers because Yeah. When you're dealing with boxes that are tens of thousands of dollars, right? And you're just going to run them for 3 to 5 years on a low energy model, right? And how do you switch that over to where you're running a super hot, where you only can have— and some data centers in Phoenix, right? They're not optimized for GPUs. They're running only 2 H100 servers in one rack, right? So previously that would be maybe 8 or 12 CPUs, right? because it's so hot and takes so much energy, they can only put 2 in there, right? So how do you deal with, as a mega center, how do you deal with that, right? So that's why I think it's going to break itself apart, where GPUs and GPU-hosting is going to go against the virtualization direction, against the public cloud direction. And also there's another area that is kind of like the people in the business know about this, but not many people outside of the— say if you're buying GPUs, you don't know about this. Nvidia doesn't really like the public cloud companies. The only reason why they do partnership with Microsoft is because Microsoft has agreed to not build the chip. But Amazon and Google have all agreed to build the inference chip. So Nvidia is like, I don't want to sell to this, right? And they also understand what happened to all the other chip makers for the CPU revolution. What happened to them? Nosedive in stock, right? Once an infrastructure company came out, they controlled access to the customer. They just squeezed and squeezed and squeezed them, right? So Nvidia knows that, right? And so they are not— they're using— since they're the king of the hill in the tech world right now, they can do whatever they want. Because 80% of all GPUs sold are Nvidia, right? No one cares what anything else is made. Literally nobody cares, right? So the only people buying the other GPUs are people who are cost conscious and don't want to spend the money. Everybody wants Nvidia because it still is the best. Regardless of what AMD says, there is nothing that can compete with an A100 and H100. We'll see about L40Ss and lower grade stuff, but it is still the Lamborghini on the market. And they're just also really fast innovating. Like when AMD came out with a cheaper inference chip and NVIDIA is on fast-on's tail, basically because they're so good at making high-performance chips for training, they basically just add more storage to their GPU and now it can be competitive on the inference side. So NVIDIA wants to create footprint. really want density in a couple of places. So they're trying to spread out as far as possible, get as many data centers to have HGXs, have DGXs, have as many workstations as possible around the world, right? That's their focus, which is a very smart play. They are not trying to just— they could theoretically just solve all the shortages on one platform, right? And just give it over to them, right? But they know that they want to control the customer, Right. So this is why they're coming out with these SDKs. This is why they're trying to get these relationships with these customers, because they saw what happened to the previous chipmakers and they do not want to lose access to the customer. So I think a lot of the shortages that you see on big platforms will continue. And rather what you're going to see is these data centers are going to start popping up, which is why people use our software. Data centers start popping up, being able to sell H100s at below market rates because they have a direct relationship with Nvidia and no one knows they're around and they have, let's say, 32 H100s available and it's just locked behind some access ports because then they don't have an API, they don't have any way to securely log in. And I think you're going to see more and more of that where these individual companies start selling their own GPUs. Then that's also where the public cloud is not really these things are so expensive and you have obviously companies, probably some that you invest in, that are buying $5 to $10 million systems, right? So that's very different than buying your own bare metal CPU, maybe a couple hundred thousand dollars, right? And so they want to resell that, right? And you can go talk to these companies. They're like, yeah, my friend's using it today, right? Because he doesn't have access to H100s. That's also like a new thing too, right? And it's almost like back to the 2000, 2001 era, right? And that's also where public cloud is not capable of really competing is that a lot of these companies are just going to resell to the market. And I don't believe we're overbuilt. I've read that a couple of times because I think that's a very narrow view about— plus the economics are wrong. Most of these public cloud companies are making 15 to 20% margin on GPU sales. So that's not correct. But yeah, I see this kind of decentralization appearing with GPUs because they are so expensive. Virtualization is not how you best use a GPU. Public cloud is not optimized based on data center design to support GPUs. The newer data centers are the ones that are more capable of supporting GPUs. Those are typically smaller, more agile. And Nvidia does not want to sell to public cloud. They want to sell to all these other ones dispersed all around the world so they get their footprint in to That's well said.

It's a great overview. Why don't you share kind of like, what is the biggest question you have in terms of what you think is still unsolved or not figured out in terms of, let's say we're having this conversation a year from now or 6 months from now in terms of how you think this is all going to play out?

I think the surge continues because no one really knows how to optimize on these chips. And I can tell you, this may give a lot of investors heartburn, But there are many times that companies come to me and be like, hey, I want an A100 80 gig. That's by far the most popular chip right now, by the way. H100 is not the most popular. A100 is the most popular. And they're like, yeah, I want an A100 80 gig, and I got to do all this training and inferencing on it. And I look at it, I'm like, well, I mean, you could do this with an A40, right? You don't really need to run. It's like, no, but I read from a friend that did this and this on it, right? And so that's going to continue for a while. And they don't know how to optimize on these chips, right? So lower-grade chips, you can do a lot of work on, like a lot of great work on RTXs. You can do a lot of great stuff on it. And you just got to know how to use a storage server combined with the card, right? And we're seeing more of that, especially on the refurb market. That's like a whole area of great value for individual companies. I'm actually not encouraging them to buy new. I'm encouraging them to buy with the refurb.

Yeah.

Because it's just like 1/3 off and it lasts still for another 3 years and it's great value. And it's like, you know, versus like an A100 80 gig HGX right now would be like $180K. You can get like a 6-month used one for like $125K, right? So, and that will run for another 3 years. Like, you know, Nvidia can clean and pull it and give you another like warranty on it. So, so yeah, I think that like how all of the supply works together in this orchestration. And I think that the negative view on GPU, which is unknown, is whether or not there'd be another use case. Will it expand outside of AI? And these chips are so powerful and it's such a jump beyond CPUs that just nobody really paid attention to because it's in gaming and stuff. That the applications, I think, are going to be much more creative as the refurb market and other chips come online and pricing continues to pull down. I mean, GPUs are always going to be more expensive than CPUs, but as it gets pushed down, I do believe you're going to see more people just adopt it and they'll see how powerful it is and will start using it for more different reasons. I mean, certainly that's where Nvidia's going with this is they want GPUs used in healthcare and real estate. Finance and construction. They want to spread their wings and go way beyond AI, which is smart. That's also why Sequoia Postman Viral was not viewing this properly. There's a ton of companies buying GPUs that are not in that business and they're in other businesses that require virtualization. One of our companies that uses our software, they use it to make weapons. So that wasn't factored in that what's left in the being overbuilt, right? There are different things being done with it, and I do think that will continue to grow. But that's the biggest open question of when a company becomes really hot that use GPUs to do something other than AI, then another wave will happen. And I don't know what that wave is going to be. And so it's highly brand and press dependent. But somebody is, somebody's going to describe their infrastructure and then the people associate their infrastructure with product success. And you also know how I know that happened is that's what Netflix did, right? So the famous story around Netflix and Amazon is that the reason why AWS can be successful is they basically supply Netflix with all their infrastructure, right? And so that's something that's going to happen again that's not AI, but uses GPUs. And that's what we don't know. And the resale market is going to be— the resale market around refurb will be, I think, change a lot of people's perspective about GPUs. Because if you're looking at something that's half or a third of the price and it's still a great box for another 3 years or even longer, if you just use it for bursting into training or something like that, then it's a very different value proposition than when Tesla bought, I don't know how many, about 4,000 H100s, right? It's very different than that sort of thought. I think investors only see that. They don't see how the rest of this market is developing as the price comes down. As you know, it's classic innovation, right? They're like the first car to ever receive— I'm a big Audi fan, which is why I use the Volkswagen Group example. The first car to receive those cool headlights, right? Like Audi's known for its headlights, was the R8. That was the first car to receive really cool headlights, and they built a whole brand around that. And then now it's in the A3, right? And it took a decade. So we're going to see that innovation spread. And only focusing on what training and inferencing is doing, I think is very narrowly focused on what actually is happening within NVIDIA and actually how— you have to give them props, like how they know they're the king on the hill and they're trying to enforce a strategy to make them to where they're a permanent fixture within big tech. Big tech as in like, unicorn tech going forward and is not going to go the way of Intel or IBM or something like that.

That's a great overview. I want to segue a bit to China, and I want you to talk about how this all plays out or how you see us interfacing.

Outside of like, when are we eventually going to have some kinetic interaction with them? Yeah, it's like kind of destiny at this point. One is that this is also the first time, not factored into that, The other reason why I referenced that Sequoia analysis is because it went viral. I think also that's something not factored into that analysis is that GPUs are considered probably one of the first times like a national security concern, right? Where they're banning the sale of certain GPUs to countries, right? So think about that. This is like fighter jet level controls that they're applying to things, right? So China can only get access to certain types of Nvidia products and they are hoarding still. They're finding ways, black market, refurb to buy and hoard because everything you have access to sucks. So they're trying to get ahead of this and of all these embargoes and blockades. But that's also something we have to think about with how if there is— and I don't think, for all of my belief that politics is like the show Veep, there are things that happen incentive-wise that are aligned with what people think the future is going to be. So if the government is already stepping in being like, These things are incredibly valuable and they create things that could be threats to national security. That's kind of next level in terms of innovation that we don't fully understand what's going to flush out. Right. And my guess is that somebody in DC was talking to some senators about— probably came from the Senate. I don't fully know how a lot of these EOS flushed out, but this sounds like something that would happen in the Senate. About how GPUs can be used to make an AI model that attacks the stock market continually, like basically a more advanced DDoS. And so we have to prevent the access to that, which is actually pretty probable, right? I mean, we've heard about Chamath is kind of a little bit of doomsday on GPUs in terms of being used as hostile things against us, right? And I don't want someone to do that, but I can definitely see someone doing that, right? I don't know how successful it would be, but if you make an LLM to tell you how to make cookies, I guess you can make an LLM to how to break into the NSA, right? I mean, it kind of makes sense. I'm sure it's being tried right now. But the conflict point between us and China continues. The asymmetric stuff, which I consider embargoes asymmetric, continues to ratchet up. It continues to increase. And that will also continue to impact how GPU sales go and the scarcity that it creates. And also just like general competition too. They kind of defaulted Nvidia into regulatory capture as in a permanent fixture of what is valuable to the West. And it also goes into their whole chip strategy, right? The one reason why they're shifting production to H100 is they want to reduce exposure to other chipmakers that possibly have exposure to the CCP, right? And we're already seeing it in our own platform. people requesting non-CCP exposed parts, right? And that's coming a lot from CIBRs, which are basically like government innovation projects. And so we're seeing that now happen too, where previously everything's a shortage, doesn't matter where it comes from. Now it's like, oh wait, we need to think about this because the real value and what we're creating with the GPU could actually be spyware, right? Basically the Huawei down, like there's a Huawei example there. Right. And I personally haven't experienced that. I haven't had someone buy a GPU on our platform and be like, oh my God, there was a CCP agent on it. Right. And also our software cleans and pulls and it completely erases all that stuff. But you don't know in terms of physical hardware if there's something attached to it. Right. And I think that becomes increasing concern as well. But China, certainly, if you look at the scale of capabilities on the asymmetric side and digital side, I mean, I would put them either at our level or better than us. So this continued issue around GPUs and who they go to and want is going to, one, hopefully accelerate our ability to defend ourselves and create offensive asymmetric, basically messing with them as they mess with us. I have a couple of friends that are Navy SEALs. Navy SEALs just don't swim, right? just jump out of helicopters. Right. They also do espionage. And for those that are deployed in China and associated countries, they know and they will tell you how much China is assaulting our country. Right. All asymmetric, something you don't notice. Right. You don't see a plane or anything like that. But I mean, it's very hostile to the point where Russia definitely Does stuff to us, but they they sort of do it as in like the way I would describe it is what they told me would be like like your uncle that's kind of like a bit kind of messed up, like can me look kind of a drunk and they kind of like wow like like you know kind of like hits you hits you in the shoulder right? But he doesn't really like want to like like push you off your throne right? Versus like the the efforts that China does is like massive social destruction, like that's what they're seeking. To our country. Right. And they don't really think, I think, beyond that. Their ability to look beyond that is, I think, minimal. But the efforts in terms of conflicts between the two, you could see by what people are doing towards these adversaries, tells you what their motivation is. Right. And certainly, Russia's has definitely escalated in terms of post-Ukraine war, in terms of that proxy war they were involved in. But the China thing continues to increase regardless of any other proxy war they were actually in with them. And there are these rules of the road that Russia still follows, right? Even though in fact we actually haven't followed them in return, by the way, but they still follow these lanes because we have a long history with them. But with China, there's no lanes. It's just like, we're just going to win, right? And they don't care, right? And consequences, they just don't want to be kinetic, right? And GPUs kind of fits into that orientation where we are issuing an asymmetric conflict with them around the computing power. And now they're going to try to go find it elsewhere and try to scoop these up because there's nothing that they can come even close to making to what Nvidia has made. Nothing. Look at their efforts to build like a 5th generation fighter in a commercial airplane. There is no way they're going to be able to recreate what Nvidia has. And so in that sense, what the White House did with these EOs is saying it's like Boeing, right? Nvidia is Boeing, right? Or something like Nvidia. Sorry, like Boeing, right? And Lockheed and Raytheon. Those aren't companies, right? I mean, they are, but they're not, right? They're state assets, right? So I can see Nvidia picking a side. I mean, it's going to be our side because this is where they make all their money is in the US anyways. And saying this is the country we're going to go to, right? And I've seen the CEO, he's gone to Hong Kong, he's gone to Singapore and did stuff. He knows who's in charge, right? The West is in charge of the GPU market and there's not enough money that the CCP could bribe them with that would make it to where it'd be even compelling And obviously Taiwan sort of situation too, where we protect Taiwan. So that's the game theory around where the GPU market is going, which is why the $20 billion missing hole thing I think is a bit of a red herring because it's not seeing actually how all these things are intersecting. This literally is not just like a chip made in Silicon Valley anymore. This is literally viewed as a weapon and it must be controlled. And that's where the whole market and how it flushes into startups and AI and investing. And we have these H100 companies just all selling just H100s. It makes it for a very fascinating next 2 or 3 years as the computing market entirely shifts over. I believe it will into using these new times of parallelizing computing because You know, the governments of the world have said like, this is the priority.

So that's well put. What other foreign policy concerns or predictions or questions come to mind for you as it relates to GPUs and how things might play out?

You know, sometimes what happens, like, with the history of China, there's a great book about China called Mao. I think it's just called Mao. And it's just like a 900-page biography about Mao. So I've read it. I think I read it twice actually, and I actually read a biography about Lenin twice too. Those 2 people will tell you a lot about non-freedom, non-free market thinking. Just read a biography. I think those tend to be much better at understanding the philosophy of a dictator or an authoritarian because you got to get into the psyche, right? Because it's not just like you can read abstractions and reviews here and judgments, but if you understand the psyche, you can easily more pick it up. in terms of how it appears in other parts of the world. And the history of China is that they're not very good at generating out of whole cloth new things. They're very good at copying and iterating. And there's a great story. One of our buddies that worked at a German manufacturing company and he lived in Colorado, And he told me a story about the Shanghai Hypertrain to the airport. And he goes, yeah, we went there and we demoed the train, our train set. Here's our train, cost whatever billion dollars, right? And China didn't pick it. And then the CEOs went to the launch of the train and they're like, oh, this looks really familiar, right? This looks really familiar, right? Like the design, there's small cosmetic things that are different, But the whole train design just looked like what they made. And I think, is that innovation? No, of course, you're just copying something. And you see these derivatives of things. The great is that their new commercial aircraft that looks exactly like an Airbus 220. And it's a different sizing, but it looks very similar. And the engines, by far, which is the hardest part of a commercial aircraft, is still produced by the West. So it's not getting off the ground unless we decide to continue to allow them my parts, right? So they're very good at that. And there's other examples of this, like how they got their 5th gen fighters, how they got their 4th gen fighters, how they got the submarines, how they got their first nukes. These are all examples of game theory. One of the greatest examples of this is— I love the story that Stalin did not like Mao. Which is kind of amazing to think about, right? Because they're both insanely brutal dictators. But Stalin thought Mao was crazy, which is saying something, right? And so during the early post-World War II and the Korea conflict and stuff like that, Russia viewed it like it had to defend China because it didn't want the West to be on its border. And so Mao would make up these kind of arguments about Knowing that, saying like, oh, you don't want Russians to come into China and to defend China for you, right? Because he knew that Stalin didn't want China to become like a Western ally. He didn't want Korea to become a Western ally. So literally Mao would pitch him things like, oh, just send over your weapons. You can station them here, send over your scientists, right? So how he got subs and nukes was essentially convincing Stalin that you just don't want to send Russian troops to defend Chinese territory. Right. And in fact, how he got the nuclear weapons, and this is in the Mao book, was that Mao sent essentially alerting, angry messages that he knew Eisenhower was watching and made all these threats. Right. And then Eisenhower then says to defend Taiwan that the nuclear option is on the table. Then Mao then goes back to Stalin, says, you need to give us nukes because if they nuke us, you have to come in and defend Chinese territory. Right. And Stalin agreed. And that's how they got nukes. Right. Then they stole the technology from Russia. So that's how they got submarines too. Right. So that's the pattern of the CCP. Right. And that's why this whole reopening thing with— I generally agreed with Deng Xiaoping and Nixon and all that stuff. I generally agreed with all those things. But we sort of got too far into it, really, in the '90s and 2000s, where we were super naive about IP protection and what the CCP has a history of doing. And now we're much smarter, thanks to really only one person, which is President Trump. He changed both parties' view of China. Absolutely did. That was one of his greatest victories as his administration. I understand a lot of people may be watching this don't like him, but that's objectively true. And both parties would say that.

Right.

So that has changed now, right? Where we now view them from a much more hostile perspective. Can you imagine, Erik, like 10 years ago I was talking about, let's say, amazingly successful Chinese social media company that we're talking about banning? That would have never happened like 10 years ago. Now it's like the hot take. Yeah, we should ban TikTok, right? Senators and states and everyone, it's like the hot thing to do is just to ban, right? So much so that on All In, the resident Democrat like Chamath, who I guess Jason too, has said that, yeah, it's going to happen. He's like, they're just going to ban TikTok. It is the conclusion. It's not going to be that big of a deal as people think. That's saying something coming from as loyal as a Democrat is. And obviously David is full on board. Right. So I think that we're entering this new phase of regionalization and almost like I don't know if the right word is deglobalization, but maybe more specified globalization where we're withdrawing back into our own territory and our own borders and are working with people that we trust, which will mean we have to be okay with this as people who use risk-based capital, like longer-term inflation, because we are onshoring things back. And in the cheapest place in the world for America to produce something, It's not China, it's Mexico. And so that whole re-enshoring thing, I think it's going to be China-plus strategies. But like GPUs, I can see the same thing happening. I can see manufacturing moving for those chips to the West. I can see more and more high chip production, as you saw with the— I forgot the name of Biden's bill, but the Biden bill to re-ensure chipmaking. Now that's going to increase price of all these things, right? But I think the West has decided, including me in Europe, that that's worth it. And that's a whole new wave of innovation. I know Andreessen has a fund devoted to this. I'm sure some other people are as well. But that's a whole new wave of companies that need to be made and supply chains that need to be rebuilt. And maybe that's the new alternative second economy, right? Is this just like re-onshored version of what previously was offshored? But it is happening. And you can talk to people who make not software things, but real things, like physical things. And every one of those companies will tell you that that's what they're doing, that the lockdowns triggered it, and that the whole break in supply chain triggered that to where they were interested now in other supply chains. They developed those new supply chain relationships, Now they see that, like, oh wait, this other country that we relied on basically stabbed us in the back, right? Kind of screwed over our company for a year or more. So that's a risk now I have to factor into my lean supply chain. And they're just going to go to places where they know it's not going to happen. So they're not going to go there. The pricing, they're going to factor in that risk calculation. And it could be like Malaysia, the Philippines, other places that are more allied to us. But yeah, I see Mexico as this new rebirth of re-enshoring and other low-cost manufacturing regions in America, including Colorado and Texas and the Plains states as well, to really bring this back locally. And I hope that will include GPU production. And Nvidia does US-based only chips and stuff like that.

That's a helpful overview. What we were talking about Trump— I'm curious if you could briefly compare Biden and Trump in terms of what are the biggest overlap and biggest differences substantively policy-wise. In many ways, it seems that Biden is sort of a continuation of Trump and even harder or harsher or sort of more Trump than Trump on maybe some of the foreign policy stuff or China stuff in particular. But, but in other ways, it's a bit different. Why don't you unpack that?

Well, one, I mean, I guess We don't really know who's making decisions. I'm certainly not Biden. We've all seen how he walks. That person is not making decisions. He's essentially as that one guy, that comedian says, he's a Roomba. So he's just cleaning the grass as he walks outside. So somebody is doing this, but also the neocons have to go somewhere. They're certainly not Republicans now, so they're going back to the Democrat Party. And I think that's where a lot of this hawkishness still is continuing. But also it's like an easy scapegoat. Going to my Straussian theory, it's easy. It's so far away, they look different, they sound different. Easy scapegoat theory.

Right.

And I think that that is often the desire when a country is internally stressed. And we are. We have high inflation, high interest rates, unsure whenever we're going to go to a recession. Everyone's been thinking recession for the last 2 years. Hasn't happened, which is kind of interesting and which makes me think that maybe Powell was Smarter than everybody thought. I mean, we'll see again next six months, but it's also kind of impressive, right? That like maybe he does soft landings. If he does, there will be like a literally whatever he did will be named the Powell method, right? If he doesn't, then you know he goes the way of the 2008-2009 crisis, right? I think that there's they there's things that are consensus that both parties will accept because it's sort of viewed as a non you can't win elections over. So that's generally what I view when policies become bipartisan. No one is going to vote against you or for you for that issue. And so they give ground on that because they just don't really care and they can find some easy path to find some compromise. So China is not like— there are certain hawks that will vote for China and those generally would be Republicans, but that's not going to win an election. Right. And so especially for If we're looking at this next cycle, charisma matters the most, right? And midterms doesn't matter the most. It doesn't matter because people generally vote for the top ticket and then it flows down. So as next upcoming election, I think it will be a lot more about basically what do we want to be as a country? And so Biden, if he runs, which I'm skeptical that he does, I just don't know how you do this now, right? It's to the point of We keep referencing All In, but it's the number one podcast for our sector, our industry. But even if you're losing the support of Jason and Chamath, that's saying something, right? That they look at him and it doesn't breed confidence, right? And so I think that they shift to have somebody else come in that's more younger, more vibrant, that resonates with some of their identity categories they used to get a lot of people out to vote. But the overlap of the parties, I think, is generally focused on things that you generally wouldn't win an election on and doesn't hurt the other person. They generally create lines on things that they think they win over and create wedge issues. And China, thanks to the Trump administration, just became the thing that everybody just kind of agreed to. And there's differences here and there, but remember, I'd freak out about the Trump tariffs and all this crap. We just sort of accept it now. Right. And that's because I think, one, post-lockdowns, that changed also people's view of China because just how dishonest they were and they were not being forthright with what was going on. The second is it's not the cheapest place to make anything anymore. There's a ton of other countries that's cheaper, that are much more friendly. Even communist ones like Vietnam are significantly more associated with these More favorable patterns to our country. So much so, I think Vietnam will actually become a democratic country in a couple decades. So yeah, I think that's actually created this fissure where there's a consensus built because there will be other issues that they can pick as wedge issues, and China is just a super easy scapegoat. And they're actually like, no. A scapegoat generally doesn't justify its behavior, but will, uh, in this case, like they know they actually are our greatest adversary. So yeah, thanks, Erik, for having me. Uh, and I appreciate it. Uh, it was a good conversation. And yeah, if you want to follow me, follow me on Twitter, uh, AGINT, or check out my company Hydrohost if you are interested in good GPUs at wholesale prices. So thanks, Erik.

Upstream with Erik Torenberg. Is a show from Turpentine, the podcast network behind Moment of Zen and Cognitive Revolution. If you like the episode, please leave a review in the Apple Store. Turpentine is a network of podcasts, newsletters, and more covering tech, business, and culture, all from the perspective of industry insiders and experts. We're the network behind the show you're listening to right now. At Turpentine, we're building the first media outlet for tech people by tech people. We have a slate of hit shows across a range of topics and industries, from AI with Cognitive Revolution to Econ 102 with Noah Smith. Our other shows drive the conversation in tech with the most interesting thinkers, founders, and investors, like Moment of Zen and my show Upstream. We're looking for industry-leading hosts and shows along with sponsors. If you think that might be you or your company, email me at erik@turpentine.co. That's erik@turpentine.co.