AI:AM — GPU Markets, Government AI Spending, and the Physical AI Supply Chain · October 22, 2026

Next year, we'll see governments become the largest buyer of tokens. We'll see a very significant ramp-up in intergovernmental competition between China and the US. There will be kind of an arms race in hacking. You need the frontier tokens to stay up, to stay, to do that well. The government can easily pay half a trillion to make sure that they are not hacked by China, which would be a political nightmare for anyone in power.

Hyperscalers do not have a significant price premium relative to NeoClouds. It's an illusion. If you're selling on demand, On-demand has a price premium relative to reserved contracts. One of the things that is different about an AWS or a GCP or a hyperscaler is if you go and you pay, you know, their $11 or $15 an hour, depending on what you're trying to buy, on-demand price, you'll get that capacity. In order to make that work, you have to take the risk that somebody is going to turn it off and maybe you won't have a customer.

First, the news: Amazon's bot checks that punish trusted agents, stolen API keys feeding more cyberattacks, and why cybersecurity won't absorb AI's displaced workers. Then Evan Conrad of SF Compute on why short GPU rentals rest on long commitments and why lenders can't price the chips they hold as collateral. Then Lanos Holmoshi of Cambrian on the factories behind AI and whether their profits survive scarcity. And to close, the hosts on peer-to-peer agent networks. A day into the automated attack wave, regional banks report AI agents logging into accounts and no way to tell them from humans. The CAPTCHA, invented at Carnegie Mellon in 2000, was built to keep bots out. What happens when the bot is yours?

I did have one interesting experience the last couple days that kind of bears on this a little bit, which was just trying to get My main laptop agent. When you buy something on Amazon, they really don't want agents on the site. They were throwing up the check this box to, you know, prove you're human, not the Google one, but, you know, their own. They were throwing that thing up every screen, basically. Claude being the good boy that it is, you know, didn't wanna just click that button and override it. I greenlight you to hit this one button in Amazon for things that I've explicitly told you to buy. And then Claude was kind of like, I don't know if I'm comfortable with that. So I didn't push it too hard. But bad for Amazon, it seems. I really want to use any model but Claude or GPT right now because those are the ones that I trust based on all the benchmarks around how hard they are to jailbreak and misuse and prompt inject and whatever. This is the one thing where I'm like, well, okay, DOTS would click that button, or maybe GrokBot doesn't care and would click it, but I still don't want to bring GrokBot onto my main laptop and give it all the access that I've given Claude. So I guess I'm probably just gonna have to live with a high-friction Amazon experience for the moment. Tell me more what's caught your attention and what, if anything, we can make of it.

There was a company called Railway, and Railway gets hacked. Not very clear what was doing the hacking. Basically what happens is they're storing all of the API keys that people use for OpenAI and Anthropic and OpenRouter. And what's happening is basically These agents are attacking Railway, getting access to people's API keys for OpenRouter and OpenAI and Anthropic, and then using the accounts of those people to create more attacks. So the AI that's being used underlying is actually stolen AI. They're using stolen cars for the robbery. So Railway announced, and then I saw this very interesting post from— this is the founder of OpenCode, and he responded to the founder of Railway, the Railway post. I'm sure you're all better than me at this, but just some of what we've seen, if it helps, they're not going to stop. The only reason we're surviving is because we have agents going 24/7. We're still barely sleeping. Good luck. Let me show you another post from Aaron Levy. So Aaron Levy is the CEO of Box, and so he, he responded to one of these posts. If you wanted to know all the jobs that AI will produce, look no further than cybersecurity. This is going to become one of the biggest investment areas for enterprises going forward. And even as the work becomes more agentic and autonomous, you will need more talent to point the agents at the most critical work, review findings, patch systems, and respond to incidents and more.

I mean, I don't doubt that like really good cybersecurity people are gonna, you know, cash in on this trend. The idea that this is gonna create jobs is honestly a bit ridiculous because the timescale is just absurd. Are we gonna go get like the content marketers that are currently out of work or the graphic designers? The new learn to code is like learn to become a top-tier cybersecurity professional that can contend with these like basically super intelligent agent swarms. But the idea that this is gonna be some sort of sponge that's gonna like soak up a lot of talent, there's just not that many people that can do this. The bar of like the, the competence level that you need is clearly rising in real time, but just the idea that you're gonna have like some sort of boon in hiring, these people don't have time to hire right now. Who's gonna spin up an agent to, to supervise the agents? Like,

Evan Conrad's first startup nearly died on a 12-month GPU contract it needed for 1 month. And the company he built from that mistake now runs the clusters and the market where where such contracts change hands. So who really pays for the freedom to walk away? And can an AI startup still afford it?

So we have Evan Conrad. He's the co-founder and CEO of the San Francisco Compute Company. The company operates clusters of GPUs, the computers used to train and run AI models, and sells access to them through contracts that customers, can resell when they have more capacity than they need. It combines the work of running a cloud with a market for computing time. In September 2026, SF Compute announced 2 3-year agreements totaling $245 million in computing capacity. This week, Evan offered another 1,000 GPUs on terms extending as long as 5 years. So, Evan, welcome to the show.

Hey guys, how goes it?

The question I had around the gap between hyperscaler pricing and NeoCloud pricing, which I'm sure you're well aware, is like pretty wide. I'm surprised by how wide it is. I wonder how long that can endure. Maybe part of the reason it's high is because it's like much more spot market. And so they are tolerating some amount of wastage in order to charge these, to have the capacity on demand to charge these much higher prices. What do you think is the future of that?

Hyperscalers do not have a significant price premium relative to NeoClouds. It's an illusion. If you're selling on-demand, on-demand has a price premium relative to reserved contracts. NeoClouds are mostly selling reserved contracts to folks. And there just is a discount for having a longer-term contract. You're giving up a premium that you would've made if you were willing to take a cluster and you were willing to have it at risk where the customer can walk away at any point. And that premium is pushed onto the customer. So one of the things that is different about an AWS or a GCP or hyperscaler is if you go and you pay, you know, their $11 or $15 an hour, depending on what you're trying to buy, on-demand price, you'll get that capacity. You won't get infinite amounts of capacity, but it's available to buy. And notably, the reason why it was available to buy is because it's so damn expensive that they're able to, one, make up for the fact that they've got some idle utilization that they have to keep around in order for you to be able to get that capacity in the first place. Meanwhile, that isn't something that most of the market actually buys because most people don't wanna pay that price. People do not want to buy on demand at the price that you actually have to charge in order to make it viable. Like on demand, when I mean, I mean like it works like a light switch. Like you turn on the light and then it's on and you turn off the light and it's off. In order to make that work, you have to take the risk that somebody's gonna turn it off and maybe you won't have a customer. And so you have to charge a higher price in order for the fact that somebody might leave. Reserve contract doesn't work that way. Like they're locked in with you.

A few days ago you announced you had 1,000 NVIDIA V300 GPUs you were offering. And the contracts I think were from 3 to 5 years. So tell us a little bit about how you put that kind of 5-year offer together.

The reason why something would be a 3 to 5-year contract is mostly from a lender. So right now the bottleneck in compute and AI in general is not Power, it's not chips, it's entirely credit. So in order to get debt, you need to work with somebody who has excellent credit because your lender has no idea how to underwrite the value of the collateral that they've been given, which is a GPU cluster. If in the case in which the AI lab falls through and fails to pay, and then the cloud fails to pay their lender, the lender is going to recoup a supercomputer and the lender has no idea what to do with that supercomputer. Not really. Nor do they know what that's going to be worth in the future. And so they're pretty much forced to realistically bet entirely on the credit risk or the credit quality of the customer, like the AI lab. The problem is every AI lab pretty much does not have good credit. Right. They're all new companies. You don't know if they're going to be around. We're still trying to figure out if Anthropic and OpenAI are going to be around or which one is going to win and so on. This is a question in the air. So what hope does the 2-person startup have that's trying to start a new lab or something? And what that means is that in order for that new startup to get a cluster at all, they need to pay a very large deposit on the cluster and they need to do a long contract so that everyone is comfortable that, you know, these people will be around and stick with it for a long time.

What do the lenders want to see from the customer? Because is it like a take-or-pay contract? No, you need like a Meta or Google to like honor the take-or-pay contract. Like you can't have this like 10-person startup being funded just for like 18 months.

I mean, what both the lender and the cloud want is an investment-grade customer like a Meta or a Google. They're likely not going to get this unless they are they've got a tremendous amount of capacity and they're willing to basically not be a cloud. Like Meta and Google don't want you to be a cloud. And so if you are trying to be a cloud, you are trying to sell to customers that are a little bit less than investment grade. And if you're an AI lab that is just trying to buy capacity, that's gonna be very difficult for you unless you raise an enormous amount of money to basically try and pay for the contract in cash rather than have to pay for it over time. If you can just pay for a 3-year contract entirely upfront, like, it doesn't matter, anyone will work with you. But that's why you see AI startups raise these astronomical rounds before they've, you know, got revenue or the product or traction or anything, because it's survivorship bias. Lots of AI labs will go out and try and raise money. The only ones that you see that exist are the ones that are able to secure extremely large rounds because nobody else can exist. Investors often wonder, like, oh my, which AI lab, which inference provider should I fund? Which one's going to win? Which one's going to work? And the answer is the one that you fund.

I know you have developed this technology that you call physical settlement. I don't know if it's a technology or capability, maybe a little bit of both. We have kind of 5-year contracts on the one extreme. We have light switch on/off on the other extreme. And it seems like this sort of physical settlement idea would be an attempt to create a more liquid in-between where I can say, you know, I'll commit to 3 months and you can deliver that and retain flexibility on the back end of it. How is that taking shape and what more middle options should we expect?

So when we say physical settlement, we mean the entire operating capacity of a GPU cloud without owning the GPU cloud. So we will go do site selection, we'll do the the design of the cluster, design of the bill of materials, backplane components, bare metal VMs, the network, and we own the whole SLA. We do it on behalf of other people. But it's the forwards market that makes it possible for you to have shorter contracts. Because what happens is somebody buys a longer contract and then they sell some portion of it, and that opens up shorter-term contracts for somebody else. That's sort of the gist. And it's part of the reason why we got started, because originally we were that. We had bought way too long of a contract and we were opening up shorter contracts for other people. And we were able to charge a premium for that because we didn't force you to sign for a year-long contract. You could sign for a month with us and we would charge you a premium on that.

So I have a bunch of questions about kind of who participates in this market and the technology that you've built to enable it. Let's maybe start on just kind of the sell side. Like, for starters, who out there in the world today has a bunch of GPUs that they're just kind of sitting on? sitting on and thinking, yeah, maybe I should rent these out? No.

We're the cloud. So what we do is we work with folks who've got the power asset or they've got the data center space, not the GPUs. And you bring the GPUs? Well, sort of. So what we do is they want to have the GPUs, they wanna own them, they wanna be on their balance sheet. A really simple example is like you're a platform as a service company and you would like to suddenly own the GPUs 'cause you realize, bam, are you gonna make a lot of money if you own those GPUs? All of these people, They don't know how to operate a supercomputer. They're gonna, they're gonna fuck up the purchase. They're gonna install the chips in the wrong damn data center. Like the whole thing's, they're not gonna meet their SLAs. What we do is we take them up to a certain bar so they can meet their SLA. And so we offer that as a service to people who would like to be the cloud themselves. And we're kind of like a GPU cluster property manager from their perspective. And then on all the stuff that we operate, we get to operate a forwards market, which lets us do interesting things in, in the realm of lending.

So, so I guess what you're saying is you are making it so that a lender can seize the collateral and operate it if necessary, because without you, it wouldn't be possible for the lender to operate that asset. Is that the right way?

Kind of, yeah. So right now, in a traditional GPU cloud, the lender is going to seize the asset, which is the GPU cloud, but they don't like They don't have any way to know if they're gonna be able to sell it. They don't know how to operate it. Like they're just gonna end up with a supercomputer they don't know what to do with. In our case, what they're gonna end up with is a, like an order on a book, something they can like press a button to sell. And not only is there gonna press a button to sell, but somebody was standing around waiting to buy that the moment it was available for sale. Mm-hmm. And that liquidity allows it to be like, better collateral than just the GPU cluster itself. 'Cause you're not taking on the risk that you suddenly have to operate a supercomputer.

What, what does the loan structure look like at this point?

Functionally, what we think is that by making the collateral more firm, you make the cost of the loan go down. So you make the interest rate go down and then you make the down payment on that loan smaller. You make the lender care less about the credit risk of the customer. This means that more types of people can buy compute.

How big does a provider have to be to come into this market? If I wanted to create Nathan's little out of my garage contribution, my guess is it wouldn't be enough for you to bother with. But where does it start to be interesting to you?

5 megawatts or so is probably the minimum for us. It's difficult to get something working in your garage, in part because you probably don't have the ability to power it. But, you know, a smaller data center is something that we can take over and handle. I do think you're going to see lots of, like, smallish scale stuff start to pop up around the world. So this is like under 15 megawatts of capacity popping up all around the cluster in the data center next to the power rather than putting it next to the specific customer you're trying to target because you don't care as much about latency.

Is this the kind of thing that a car company can get into because they've got some extra electricity sitting around? They can throw up a warehouse. They've got at least some balance sheet. Is Is that— are we at that stage in the cycle yet?

If you have grid power and you have credit, we would love to work with you and we would happily help you become a GPU cloud. I don't know if they would do it or what that asset is currently being used for, but lots of people are taking things that already have grid power and trying to convert them into data centers.

Are you operating outside the US? Are you dealing with European firms, etc.? Do you see demand in Europe or Asia, etc.?

We operate around the world today. When you go to Claude or ChatGPT and you press enter, there is an initial latency that's the time to first token. You want that to be fast so that it feels good and experiences quickly, but then you mostly care about throughput. How much are you pushing out of it? And this is not something that's latency sensitive. If you're gonna run something for days, Like, it doesn't matter if it's 200 milliseconds more. You just want the cheap thing that you can get a lot of so you can just push things through. Yeah, we're seeing capacity demands all around the world. This isn't the case for every single customer. Some types of inference don't work that way, especially if you're doing something like, you know, live video or audio. That's definitely not the case. The thing that matters in this case is every country is trying to have an AI strategy of some sort. From an economics perspective, the way that you get the value of AI to flow to your country or your state or your city is you put data centers in them. So if you think about like the data centers that are going on in the United States, there was a big one in Utah that people were very unhappy about. Nonetheless, that tripled the tax revenue of the local area. So if you can build a data center somewhere else that is in a country that is not the United States, and would really like the economic benefit from it, you can put a project that's quite meaningful to that economy in that country. So this is Sovereign AI. Lots of people are trying to do this. One of the things that we do is we are the San Francisco partner for Sovereign AI. We will go and we will be your supercomputing partner. We bring the expertise from Silicon Valley and we also bring the offtake from Silicon Valley.

I think you had an MTS Live appearance where you talked about a credit crunch in 3 months. So where do you see this kind of credit crunch? Are we gonna see this distressed credit cycle? Do you see credit tightening and like people not being able to fund their data centers?

I mean, not to toot my own horn, but I think I was almost exact on that timing. I don't remember when that MTS Live thing was, but seems like the core Poor weave debt is getting more expensive, although I think there's a lot of complexity in that one. It also seems like it's very hard to close new customers unless you're investment grade or near investment grade. And many customers are being asked to pay for 100% upfront payments on their contracts. But I think for the next possibly a year or two, if you're a VC, you should be Prepared to write some really big checks for your portfolio or encourage the companies to sell. I think we're gonna probably see a lot of consolidation of AI labs and inference companies. And I think the reason why that's the case is because they're gonna be asked in order to get the next cluster and grow to pay for the entire contract upfront, and they won't be able to do that. And so they're going to have to consolidate. going to instead be forced to sell their company to somebody else who has great credit and doesn't have to do that. Alternatively, they'll go out to investors and they'll try to raise big rounds. The investors will say no, because I think, um, at, you know, at some level of scale, they'll, they'll probably just not be able to stomach the risk.

I'm an AI startup. I need like 1,000 GPUs and I need to raise money. And how do we put this financial contract together? How much money would I have to raise normally for 1,000 GPUs? How much money would I need to raise with SF Compute in there?

So today you're going to need to do roughly the same. What we're doing is we're building up the ability for that collateral to be worthwhile. That's going to take us several years to do. Lenders, lenders don't get comfortable with things very quickly. You have to show proof over a long period of time. And so that's what we're doing.

I'm not going to set up my own cloud because I have to get into operation like pronto. What is the next step here?

So when we work with demand-side customers, typically the thing we're selling them is 2 things: the capacity itself.

So you're just selling them like one year, like this year, for example, and you don't require them to pay the other, the 4 years. Is that what you do?

No, we lease them for, you know, 3 to 5 years. And the reason why someone might go with us is because we let them get out of that contract. We let them sell that capacity to somebody else.

Lowness Hamaili made his name building AI software. And now runs a fund that owns the factories behind AI instead. This episode asks who actually gets paid if governments overtake consumers and businesses as AI's biggest customer, and whether those suppliers can keep their profits.

So our next guest is Lowness Hamaili. Uh, he's the managing partner of Cambrian, a, uh, public markets investment fund focused on the physical supply chain behind artificial intelligence. The fund announced its launch on October 13th with approximately $70 million in seed capital. Its focus is on the companies that supply the materials, components, and manufacturing capacity needed to build AI infrastructure. Uh, before starting Cambrian, uh, Lonis built the software businesses. In April 2023, he introduced God Mode, A web application that let people experiment with AI agents carrying out tasks through a sequence of steps. That July, he reported that it had passed 1 million users and 15 million completed tasks in less than 3 months. He also co-founded Dwork, a task management platform for organizations using blockchain-based coordination and payments. Dwork announced a $5 million seed round in 2022. His work extends beyond software. At Prospera, a special economic zone on the Honduran island of Roatán. He served as head of governance and vice president of growth, helping attract entrepreneurs and businesses. What is the fund? What does it do? What is the thesis?

A year ago, it felt very close, very clear that, okay, we're getting AGI in this decade. But I kind of looked at my own portfolio and I realized that I was actually not only not— my portfolio was not correlated to AI, it was actually decorrelated, anti-correlated, because I had all of these venture paper gains. from software startups, which would probably be quite worthless if we get AGI soon. And I started talking to friends and their portfolios were even worse. Some of them had commercial real estate, some of them had bonds, and all of these types of assets, some of them just held the S&P, but all of these assets are not set to perform well at all if we have short timelines. And I started thinking a lot about how to solve this. Obviously, I was very lucky to invest in Anthropic via NSPV. But even investing in labs are not a perfect solution because I don't think the government will necessarily allow them to get that much larger. AI is already super unpopular for no good reason. And this year there will be many good reasons. There will be misuse, there will be hacks, there will be job loss. And so these companies are extremely politically exposed. And so I came to a realization that actually the supply chain is the ones that are winning in every scenario, whether the labs keep on leading, Whether the government gets heavily involved, where the Chinese guys catch up, all of them need the same memory, the same tools, and the same power.

How is this not priced in? 2026, like it's all staring us right in the face, right? Like the skeptical take would be everybody knows about all this at this point. Where do you see that that is not the

case? If you look at how these semi stocks are trading, they're trading at 16 times earnings or so. You know, S&P on average is trading at 22 times earnings. Walmart is trading at almost 40 times earnings. I think that they are not durable. I think they're less durable than the overall earnings of the rest of the market. And then specifically the companies that we're very excited about, whether it's silicon wafers or memory or capacitors or foundries, those are trading at like 3, 4, 5 times sales. We have this thesis that the private market is actually more efficient when it comes to AI than the public markets. Because in the private markets, you have all of these ex-operators, technical people who are bidding up this small set of startups, and they're bidding them up to these high multiples, 50 times sales, 100 times sales. But in many cases, in some cases, rationally, right? These are rationally high multiples because people here are pilled and technical. It's a very different setup where very few people are technical, very few people are pilled. And especially as a share of overall traded money, it's like super, super small, and the market is much larger.

So you obviously are modelling this out with, I mean, there's all these different trends that you kind of have to take into account, right? There's the scale at which computing is ramping, but then there's also the scale at which algorithmic efficiency is driving more usage out of that compute. And then there's the uptake of AI adoption by individuals and corporates. What do you think are kind of the most Unclear curves, like the ones that you have the greatest uncertainty about that could kind of break the model over the next couple of years? And maybe even before answering that, maybe just a couple anchor points would be useful around, for example, what are you expecting tokens per capita to be? Or however you think about kind of what the global usage base might look like. What are some kind of anchor points that we can imagine? Okay, numerically, this is what 2028, 2029 looks like.

Consumers in general are terrible customers of ChatGPT, and consumers are still only paying like $20 billion per year, $30 billion per year. Enterprise is doing a little bit better this year, maybe up to $200 billion. But the real buyer would be governments, right? And that hasn't happened yet at all. But next year we'll see governments become the largest buyer of tokens. We'll see a very significant ramp up in governmental competition between China and the US. There will be kind of an arms race in hacking. Need the frontier tokens to stay on to do that well. The government can easily pay half a trillion to make sure that they are not hacked by China, which would be a political nightmare for anyone in power.

Tell me more about this government thesis, because, I mean, obviously the government can spend a lot of money quickly on something when it chooses to. Can they actually use these tokens? I mean, it's one thing to say, okay, we're going to spend, you know, Couple hundred billion on AI. Now you've got whatever quadrillion tokens you gotta use. Can the government actually point a quadrillion tokens at anything even remotely useful?

Government are happy to pay higher margins. They're not as rational of buyers as others paying for a bunch of other features other than just the per token pricing. Just like intelligence operations are very, can be very token expensive, need all of this security. They want everything on-prem. Very likely. So they will have to build out their own, you know, data centers to serve these things.

If you had to steelman the why these companies are not mispriced story, could you do it?

One steelman is that many, many of these businesses are not great businesses in normal worlds, especially the add-on businesses we like, whether that's capacitors, silicon wafers, what have you. They have, you know, high fixed costs. And if they don't have a set utilization rate, they actually lose money. And, you know, high debt in some cases. And so in normal worlds, they're just not great businesses. And it's only in the world where AI really hits where they get this big operating leverage and they become amazing business. You know, today, let's say they have 10% gross margins. So a terrible business by Wall Street standards. But if they're able to raise prices by 10%, you know, they suddenly double earnings. And if they're able to double prices, they 10x earnings. But that only happens in this very specific AI world. In a normal world, that basically never happens because usually they're very good at That's like matching demand patterns with supply ramp.

And memory right now, what the market perceives is that by 2028, 2029, there's going to be an oversupply of memory by that point in time. So that is a market perception. So why is your thesis different from the market's thesis? Are you expecting like massive revenue increase next year and another round of massive CapEx going forward?

I mean, the interesting thing with memory specifically is that doesn't require that many things to go right because it's already trading at only 5, 6 times earnings. Even if they can't raise prices further, at some point they should rerate to at least 10 or 12 times earnings. And so it doesn't require prices to increase. It does require demand to continue to outpace supply. And when you break down supply, almost no expert thinks we can more than at most 2.5x DRAM bits by end of decade. Zero experts with any track record have said we can 3x DRAM supply by end of decade. Yeah. And so then you just have, on a very naive level, you have to assume that, okay, the demand will not more than 3x by end of decade, which of course everyone here would assume we will. Then of course there's the separate fact term in the equation, which is memory compression. How much memory compression do we get? How many bits do we actually need per token? And, you know, we've been looking into how this, for instance, developed from GPT-3 to DeepSeek V4, our bits required per token decreased by something like 500x. So, you know, but in that same time when bits per token decreased 500x, Ethereum bit price rose by 6x or 7x.

So I keep coming back to, will we not apply, you know, the intelligence that we have and the first use case of that intelligence is going to be to resolve the bottlenecks in the supply chain iteratively, which is part of the RSI loop, really. Even if you are convinced of the AGI story, will you not use that intelligence to resolve the problems around that AGI production first? And how will that work out in the long term?

I think that will happen. And then the question is just like, okay, when we are at that point where we have an AI smart enough to solve some of the hardest engineering problems in the world, what else will it do? Will it not then at that point also be able to at least automate a bunch of the email jobs and Excel jobs that we currently pay trillions of dollars per year to solve? And then if it does that, then that unleashes trillions of dollars of value that the supply chain could capture.

What would you need to see in the market to say that memory prices are going to go down? What is the antithesis to your thesis?

I mean, for memory specifically, the main thing would be like when we get superhuman coders, how much better will memory compression get? Maybe it just happens to be that we just get memory compression way faster than we get other economically useful features, such that we don't get the Jevons paradox in the short term. One thing, I think that's relatively unlikely for a few reasons, but that would be one thing. Energy becomes a larger bottleneck than semis. There's less incentive to bid up the marginal wafer or the marginal memory bits to such high prices if you anyways have to wait a quarter to get power in.

Right.

And that's a little bit the kind of regime we've been in parts of this year. And there are a few reasons why we think that won't be the case, you know, going forward. I mean, we've finally kind of gotten a hang of doing, you know, behind-the-meter things. So, you know, people are like increasingly not connecting to the grid. They're building their own power plants with their own natural gas. And a big reason why you can scale energy production faster is just that there are so much more latent supply of things. So, you know, for instance, now the frame turbines are bottlenecked, right? These large turbines that turn natural gas into electricity, they're sold out until 2030, you know, from GE Vernova and Siemens, etc. But then people find other ways. They find that you can use truck motors and things from airplanes, and we can stitch things together and kind of make it work. And because there's so much latent stuff you can kind of make use of, which is very different from the semi supply chain, there's no latent supply of wafers or of memory bits. Like all of that is already used.

Elon is building out the TeraFab and it looks like it's going to be a 2030 thing. One of the things that has struck me is that Elon is specifically saying that if you need to 10 or 100x capacity, like the existing players will just not do it, right? They're just going to be at like this kind of 2x maximum by 2028, maybe like 3, 3.5x by 2030. How do you think about that mental picture? of the TerraFab coming up and disrupting the market? Yeah,

I mean, I think the TerraFab is super promising. Obviously, Elon knows how to run these large-scale, like, real-world projects. I think though, many of the bottlenecks that other players are running into will still be bottlenecks that he's running into. There's just so many silicon wafers, like, he will not make his own silicon wafers. He will have to buy them from some of the 3 or 4 current players. And supply of that is only growing 9 to 15% per year. And he will have to buy the same tooling as everyone else, the same UV machines, the same etching machines, the same deposition machines. It's not clear to me that it will lead to that much extra added net capacity until we kind of resolve all the other upstream bottlenecks.

So, so from what, from what I hear is happening at TeraFab, TeraFab is going to be like this recursive process. So they're buying from existing suppliers right now. At the same time, they're looking at how to design the next layer, which has more supply chain integration, which is what he's done Specifically in every single business he's gone into, looking at like 2 or 3 cycles. And in order to get to, you know, 2030, the real TeraFab, he's gonna have all this stuff integrated. But does that mean that he's gonna, you know, basically like deprecate like all of the existing suppliers? How would that work with these, um, yeah, with these chip suppliers?

I mean, I think Elon will, will do really well. I think he will find some, some ways to innovate. I think though there's a very big difference between kind of entering the semi-market and entering the rockets market, right? The rockets market, you were competing with governments, which is really what you always want to do. Like, very generally very incompetent players, you know, not any real profit incentives, not the real market. With the semi-market, it's way more competitive. There are many more very sophisticated players. It would be way harder, we predict, to find this very low-hanging fruits that you could find in other industries. And so yeah, I fundamentally don't think he will be able to build all of these types of machines themselves. It's very unlikely, especially in the coming 3, 4, 5 years.

As you evaluate the supply chain, how do you determine what's going to be hard, even for somebody like Elon? And obviously you've also got to kind of take into account infinite AI agents to help do whatever needs to be done. You've a couple of times mentioned silicon wafers in particular. There's plenty of silicon out there, right? Is it so hard to make a wafer compared to these other processes that this will remain a well-motivated business, even if Elon turns his attention to it? Epistemologically, how do you try to get your arms around that for all these different links in the chain?

I mean, so we like to give silicon wafers as an example because it's kind of an easy to understand thing. But there are many hundreds of layers in the supply chain. The key thing to understand with all of these layers is that there are different, these different tech tiers to them, right? So in memory, we have, you know, HBM3, we have HBM4, we had HBM2 before. And these tiers always add like an increasing difficulty. So there's always this arms race. So if we look at China, for instance, China, you know, according to Elon, China is really good at like manufacturing, scaling up advanced processes. They've been trying to do, you know, memory for 10 years now. They bought the Sherman company more than a decade ago. To get a bunch of know-how and then try to do HBM3, but they're only yielding like 25% versus SK Hynix yielding at almost 80% or so. So like they are really struggling, even like these Chinese firms which are hypercompetitive, you know, run by very smart people, they are really struggling to catch up to the frontier even after a decade of doing it. And now we're asking one company to do the same thing that China has been trying to do for a decade plus across all these hundreds of layers at the same time. It's just very hard. If there's someone who can do it, it's Elon. I think he will be able to do some of it, but it's very unlikely he'll be able to do it to a significant extent.

So one of the things that has confused me is how much does superintelligence or the increased intelligence play a part in reducing the moats to all of these businesses in general, right? Because all of these businesses are kind of gated by intelligence in a sense, like management capacity, engineering knowledge. Yes. All of these are intelligence based activities? Because that's kind of what people are predicting for bio, right? For bio, people are like, oh, you're going to get like bio is going to get like random guy, like, you know, creating like this thing that, you know, even the best labs in the world cannot create with all of their knowledge and technology, right? So why wouldn't that be true for semiconductors too?

I guess I would probably disagree that it's easier for an LLM to innovate in hardware supply chains than it is to innovate in biotech. And it's not that one is fundamentally harder than the other. You train these models, right? There's a bunch of biological data you can train on. There's not really much good training data at all for how to design a silicon wafer or how to make DRAM bits. Almost all of this is just very tacit knowledge that sits in every engineer's head.

But right now the labs are buying data all the time. So they're buying up data from all of these companies and they're creating RL environments for semiconductors. Which is definitely a key area that they're doing. I believe Anthropic has a deal with Micron. OpenAI has a deal with, I think, Synopsys or Cadence. So with the design firms that they've tied up with, obviously some kind of revenue share agreement going on there. So how does that work? Obviously, they are looking at creating intelligences that can do this, right?

Part of that's true, but it's a very significant difference between designing a ship and producing the ship. So designing a ship, I agree, that is like a core priority for all of the labs. It's really one of the key ways to get our SI loop closing, right? There's RL environments on it. It's a big priority. That's very different from building a, you know, 2 million square feet, you know, fab, right? And designing that from scratch. There's no really good RL environments for that. There's no good manuals for it. All of it is tacit knowledge. sitting in this, you know, Japanese engineers have been working at the same company for 3 decades. And it's just very hard to extract that information in a structured way. Of course, at some point it will happen, you know, once we get superintelligence, it will figure out the best way to extract information and learn the best. And, you know, it will happen for sure. It's mostly a matter of like the timing of it and like what happens in what order.

So going to the fund itself, how do you guys think about your position amongst other funds that you might be in some competition with? How do you try to contrast or differentiate yourselves? And how are you thinking about managing risk?

Our thesis is that we feel very confident about the overall arc of AI, rough timelines. And we are not at all confident about any quarter-to-quarter developments. Right? So we are doing some form of value investing where we find companies that have— that are trading at great multiples, that have durable technical moats, and then we're making bets 2 years out. And so that's quite different from the other firms. On top of that, we're not doing any margin. We actually have in our fund docs, it says, may not use margin leverage. Even if you did nothing else than just holding Nvidia for the last 4 years, you would have been up 15x. But if you held it with any margin, you would have been 0x because it was down 40%, 60%. Our LPs generally want to be rich in many worlds, not just a trillionaire in one world. And then I guess like another thing, we mainly invest in Asia, more than half of our book is in Asia. Almost a third of our book is small caps and mid caps. Smallest holding is a $30 million Japanese company, which has a monopoly on a specific type of tooling. And we kind of found out that there's no English sentence about this company anywhere on the web that we could find. The AI tools are good enough to actually be able to have this Socratic dialogue and just go deeper, deeper, deeper until you find those types of companies.

I don't know if this is a synthesis of his thesis that he would, or maybe a compression of the thesis that he would quite sign on to, but I liked the idea as a motivation for how do you figure out what is hard? Look at what China has been trying to do and not been able to catch up on. And if they haven't been able to do a fast follow, then odds are it's going to be hard for anybody in Europe or wherever else to do, or other parts of Asia to do a really fast follow and eat into incumbents' advantages.

I would probably not buy into that. It's very good. at taking existing technology and streamlining it in— and he's willing to do this because he's willing to change parameters. What he believes is that a lot of the parameters that you start off with are illusions. So for example, one of the things that he's doing in the TeraFab for chips that would be launched to space is that he's changed the operating temperature. And the reasoning being that, OK, if you have operating temperature of 50 to 60 degrees Celsius, of which the NVIDIA machines are operating at, you need cooling. And so you need these radiators. And the radiators are 4th power of your temperature in Kelvin. If you increase the temperature in Kelvin by 20%, you can halve the radiator span that you need. So he didn't fix the radiation problem. He's not going against the laws of physics. He's asking NVIDIA to redesign chips to run at higher rates. which they've never done before. And he's saying, look, we can afford to have more lossy chips. We're going to accept that the chips are going to be more lossy, and they're going to need to expend more TFLOPS in order to get useful intelligence. He's accepting that. In return— and you probably have to redesign all your models too. So he loses a lot of optionality. Nvidia, ASML, they design for more optionality. They're designing for multiple vendors. They're designing for downstream. Being able to redeploy the machines to do multiple different things. This is a classic way of engineering, which is you reduce the optionality that you're designing for, and you make everything just fit for one thing, and you design just for that one thing, and you change the parameters of the exit. You open up the design space a lot.

This is the bottleneck thesis?

Yeah, the whole bottleneck thesis is not actually that true. Those bottlenecks can be gotten rid of. by someone who is a dedicated physics guy.

Does that mean that these companies still aren't bottlenecks for the rest of the market? Because I could see a scenario where it's like, well, Elon solved 100 problems and he worked around a lot of these bottlenecks, but that doesn't mean you can, right? So everybody else who wants to get into the game or continue to scale up their operations, if they're not going to be so bold, if they're not going to be so First principles oriented, if they're not going to change the operating temperature and all these things, then maybe the bottlenecks still apply to everybody else.

Look at what happened in the launch market, right? In the launch market, he is now 90% of the market even before Starship. Commercial capacity, he has taken up all of the commercial launch capacity pretty much with one or two competitors. I think you could see the same thing happen In that market too, in semicon too. You can have like Elon, which is 99% of the market, and have everyone else like 1% or less than 1%. It could happen, right? I don't think it will, primarily because I believe in RSI and I believe Anthropic and OpenAI will get to superintelligence and then they will be able to buy or create a bunch of this stuff. So that is the only reason that I think about. If not for the RSI thesis, I think Elon would win.

So how do you think about investing? Do you have an alternative thesis that is like the Prakash thesis that you would advance as an alternative?

So I have this theory of adversarial diffusion, and adversarial diffusion basically means human beings don't adopt technology for fun. Human beings are largely satisfied with the status quo. And if you want to adopt technology such that you get to 10%, 20%, 50% GDP growth a year, that is something you do not because you wanted to, but because you had to. And I think that only happens as a result of emergencies. So for example, mRNA during COVID mRNA was a 20-year latent— 20 years it had been in latent development. During COVID within 6 months, They're like, all right, let's go ahead and do this. 6 more months, you had a vaccine out. A year later, large portions of the world are vaccinated. So I think we're going to get this kind of adversarial diffusion where you're not going to want the growth, but you're going to get it. I think we'll see it in cyber first. 10% GDP growth in the US implies that you need to break the NIMBY consensus. You're going to need to build factories in California and factories all over the place. You're going to need mechanized factories. They produce pollution, obviously. You need all of this physical stuff, which the US has not built in 40 or 50 years. And you're seeing this pushback already against data centers. And they're going to need to overcome this political opposition. And I don't see that being overcome except in a state of emergency. So unfortunately, I think it's going to be like this growth that no one really wants or likes, but you need in order to survive. So one of the confusions for me is that governments are typically very slow. Number 2, all governments worldwide are at debt capacity limits, right? Like everyone has too much debt. Like let's say a government in the US decides to spend $1 trillion. There's always a question of why don't you spend it on like curing homelessness or something? The question for me is, I think the NSA so far this year, they've increased their spend on AI by like a couple of billion, right? $2 billion or so. And the question for me is like, is it going to be seen as a giveaway to these companies? Right. Because government can also do this Defense Production Act thing. And they're going to— they can say like, look, we're going to cap your— we're going to cap your returns. And the moment you cap returns in a sector, you actually get deinvestment. I can't make that money. I'm going to get out of the sector. Why would you think that governments will become a big purchaser at high prices?

They will not do it until there is catalyst. But we predict next year there will be certainly catalysts. I mean, we will see significant hacks of banks and other institutions. Foreign adversaries might be hacking into our local governments. or federal governments, they get access to different types of information. That's really when it becomes scary. And that's when it will become very easy from a political perspective to say, no, we need to make sure that we're always staying ahead. And as you say, there is the Defense Production Act, they could use that. And that will lead to what you say, to deinvestment. That is the last thing the government will want after a catalyst. After a catalyst, the government will want more investment from American firms in AI. And so I'm sure there would be other ways to work that out.

So one of the things that strikes me though, is that We have reached the point where I think you can have this like, you know, transfer of wealth from one part of the economy to the other. And you've reached the part where you actually need the entire economy to expand. You need, you need the actual GDP growth. One of the problems is that if you cannot get like physical growth, like physical supply chain growth, you actually don't get GDP growth. If you only have reallocation of dollars.

Sure.

In terms of higher valuations, And that's happening as a result of job loss on one side from, you know, from consumers, like basically having their jobs be automated. And that is actually a reduction in consumption on one side. And if you actually have that go into just higher asset values on the other, that doesn't work. And one of the things that strikes me is that that physical growth is still constrained by, you know, human time limits, human time limits, and you can't do exponential growth on physical.

I mean, there is this, uh, world where, I mean, I'm not sure if it's a good world, but where we get software singularity, but we don't get real-world singularity. Robotics kind of lags by X number of years. And at that point, we might see significant automation of white-collar work. And so the AI companies in the supply chain can still benefit, but the broader public might not. Because to your point, maybe the economy is actually not growing so significantly. It's mostly rerouting salary dollars from humans to AIs, right? I think that's somewhat possible to happen in the short term, and that might not be great for the world. But again, I think even in that scenario, the supply chain would do decently.

In consultation with my agents themselves, the idea that I've come up with, which I'm currently calling Helppeer, but which might deserve another name at some point.

This is Helppeer, a friend-to-friend network for our AI agents. First, how I run my own agents. This is one pattern that works well for me. Not a requirement. On my laptop, an extension of me. It sees my email, calendar, and messages, but only acts when I say. High access, low autonomy. On a separate Mac Mini, employees. Their own email accounts and a CapCard. They work for hours alone and never get my keys. Low access, high autonomy. My rule is that no agent gets both.

When I say the virtual employee agents don't get my keys, they don't get my most important keys. They don't have my Gmail. They do actually have a lot of API keys for a lot of different services that I do want them to be able to use. So they can do, for example, this kind of voice generation. Actually, the AIs themselves were kind of like, are you sure you want to give the ability to generate audio in your cloned voice to, you know, the more autonomous agents? And I was like, yeah, I think I'm fine with it. There's a lot of audio out there. You can go to Minimax and clone a voice with no consent requirement, no permission, no opt-in. Knowing that that's out there, I was like, I guess it doesn't really matter if my virtual employee has That ability.

Nathan's assistant answers email from anyone who writes in good faith. That makes it a target for prompt injection, where a message carries hidden instructions meant for the model, not the reader. So which model should be the one that opens the mail?

The model that reads mail from strangers should be hard to manipulate, and the models are not equal. Far.ai's AI security leaderboard searched for universal jailbreaks across frontier models. On Grok, one cost about $58 to find. On Gemini, about $278. On Claude and GPT, the search never succeeded, even past $14,000. That measures jailbreaks, not prompt injection exactly. And far.ai says passing doesn't make a model secure, but it's the best cross-lab signal we have. So for the agent that faces the world, we suggest a frontier Claude or GPT model.

I was concerned enough about this to set up a system where when The main agent on the laptop is going to provide information to the virtual employee to then send out, like, we're gonna pause, we're gonna do this, you know, special review step. If it looks like something sensitive might be going out, then it's alerted to my phone and I have to like review and approve. But it takes like 5 of those false positives before I'm like, oh God, another one of these. Like, yes, just check it, whatever. So I'm trying to stay disciplined around that. Uh, and we'll see how well my discipline holds up.

Oversight is the human by CC, plus a daily digest of what your agent sent, held, and blocked. What we don't do is count on agents policing agents. The peer preservation work suggests monitors may cover for their peers, and Checkpoint found an agent that watched teammates break a rule broke it too, rising from 44% to 98%. So the hard rules like who's on the roster, the hop limit, and what the agent can read are enforced in plain code. The models get judgment calls, the code gets the walls.

If you brought a crypto guy in, they'd immediately go into like decentralized, you know, how do we sell

a token on this? Yeah.

And I, I, I, I don't think that's necessary. I think the trust aspect is actually more valuable than crypto always goes into like no trust. And so I think like getting rid of the trust doesn't, doesn't help you in any way. And making it monetizable uses trust in the, in the network because it says that you can actually pay for something. So I think the reputation matters a lot more than the monetary aspect.

I would consider it a win, honestly, for me, if I could just get 10 people who I think are great at building their own skills and whose agents I really expect would have a lot of hard-won knowledge. The effort that they put forward to get their taste and judgment into a skill, that's the thing that I can't access Otherwise that I really think I could benefit from.

Guys who are there trying to sell consulting services and sell their SaaS, but I don't want that.

Give me the skill, bro.

I'm not going to like, you know, go and subscribe to your SaaS and like have something else that I need to manage. And it's not the UI that I want to use either. And I may have to make some changes that don't allow me to use because you didn't build a feature for it. Meanwhile, if I have the fundamental skill, like my agent can do the rest.

Evan Conrad is co-founder and CEO of SF Compute, which runs GPU clusters and the market where compute contracts change hands. Lownis Hamaili is managing partner at Cambrian, a fund that owns the suppliers behind AI hardware. Lownis put a date on his call. Next year, he says, governments become the largest buyer of tokens. That is the one to check.