China just K*LLED Claude AI.. - GoodAlexander & David Choi
GoodAlexander argues Chinese open-source models have broken the economics underpinning Claude and other closed-model leaders. After initially dismissing them as benchmark-gamed Opus distills, he tested models delivering Opus- or Sonnet-like performance on a roughly $10,000 GPU, even 3090s. Against protected Opus API usage at about $80 per terminal-hour versus roughly $2 for Qwen 3.6-class compute, the “40 to 80x delta” makes local inference and crypto-financed compute newly credible.
David Choi sees an enormous financing gap beneath the AI-capex boom: Nvidia has an estimated $500 billion of sales over the next five quarters, “99% of that is financed,” yet loans below roughly $200 million barely exist. He says 20-30% of those sales come from smaller emerging clients that cannot efficiently access securitized debt. USD.AI’s next-12-month target is $1 billion of GPU loans—enough, he hopes, to establish a global reference rate for the cost of growing AI.
The core USD.AI pitch is that GPUs should be financed like houses, cars, and aircraft, but their product cycles move too quickly for traditional securitization. Packaging 10,000 loans, tranching them, conducting a roadshow, and selling the instrument can take two years—long enough to move from Hopper through B200, Vera Rubin, and toward an unannounced Feynman generation. DeFi instead adds each originated loan immediately to the sUSDai NAV, creating what David calls “a debt perp.”
USD.AI’s proposed flywheel uses stablecoin economics to lower AI’s financing cost, not merely to put existing debt on-chain. If AI companies settle in USD.AI, the protocol can use the 3-4% earned on each minted dollar to subsidize borrowers while retaining origination fees and net interest margin. CHIP is framed as the protocol’s GP-like governance layer, directing fees, helping set rates across deal types, and potentially prioritizing redemptions—not as a fully disclosed utility package yet.
Compute scarcity is presented as both the reason Claude feels worse and the foundation of a new sovereignty contest. David says providers are rate-limiting users because demand is outgrowing installations; OpenAI’s aggressive compute investment was therefore directionally right. His “AI dollar” thesis is that countries will become net importers or exporters of generated tokens, and a nation without sufficient AI capacity will fall behind much as an oil-poor country did during the petrodollar era.
GoodAlexander’s crypto bull case spans three different clocks: Saylor-driven inflows in one year, capital flight in three, and crypto-native financial rails in five. STRC can bring retirees’ capital into Bitcoin, while failed attempts to suppress privacy coins demonstrate that privacy technology may survive a hostile policy regime. Longer term, taxation, money printing, CBDCs, and capital controls could make portable crypto assets attractive to wealthy people who cannot “get on a plane with a bag of gold.”
His equity view is almost the mirror image: stay bullish into the AI-IPO spectacle, then prepare for a historic reversal. He expects roughly $4 trillion of SpaceX, OpenAI, and Anthropic equity to come onto the market, with a $1.5-1.7 trillion SpaceX IPO as “the main show.” If banks support favorable AI comps and passive indexes rapidly absorb SpaceX, he sees that event as the “Solana $300 moment”—the beginning of a generational top rather than a permanent new plateau.
1. Compute—not model intelligence—is becoming AI’s binding constraint
David’s opening claim is that Claude gets “knee-capped every time you try to use it” because providers lack compute, not because model research has stalled. In the long run, he argues, the decisive question is simply: “How many Blackwells do you have access to?”
The speakers use a physical comparison: build 30 Las Vegas Spheres and, by their estimate, “that’s one data center.” They put one gigawatt at roughly $50 billion and describe planned sites reaching two or three gigawatts across dozens of global projects.
David says AI capex is now so large that, absent it, US GDP growth would be negative 3-4%. He further claims one year of AI infrastructure spending can exceed a decade of spending on oil, highways, and other conventional infrastructure.
USD.AI joins what he calls two megatrends: stablecoins, among the world’s fastest buyers of Treasury bills, and trillions of dollars of AI capex. The thesis is that the “fastest horse” in capital formation should finance the economy’s fastest-growing capital requirement.
2. Chinese open models have punctured the closed-model consensus
GoodAlexander says the prior consensus treated Chinese releases as stolen Opus distills that Anthropic would eventually shut down. He shared the skepticism—Chinese labs can “over-promote” releases and game benchmarks—but changed his mind after personally testing the latest models.
The break is hardware accessibility. Workloads that recently required elite closed APIs can now, he says, achieve Opus- or Sonnet-like results on a $10,000 graphics card, sometimes on 3090s: “Prior to two weeks ago, doing that on commodity consumer hardware was not really viable.”
His protected Opus setup costs roughly $80 per terminal-hour, or $160 while running two terminals, versus around $2 per hour for Qwen 3.6-class alternatives. That “40 to 80x delta” led him from “OpenAI has just won” to believing open source remains seriously underpriced.
In their discussion of newer models, GoodAlexander points to GLM 5.1 and Knowledge Atlas Technology (ticker 2513). David says GLM is, in his view, a better chat model than Opus; both criticize Claude/Opus for excessive verbosity and Anthropic’s overcorrection from sycophancy.
GoodAlexander connects the shift directly to crypto AI: closed models such as OpenAI and Anthropic are gated and unlikely to route meaningful compute spending through crypto, while open-source models make distributed, crypto-financed inference plausible. He expects explosive growth in Hermes Agent and other open-source harnesses.
3. GPU credit resembles housing finance before modern mortgages
GoodAlexander traces David’s edge to an earlier lending system for NFTs—the “worst credit in the world,” borrowers financing internet profile pictures. GPUs are warehouse-finance assets with an asset-backed-security element that he considers superior to private credit’s current treatment.
David’s mortgage analogy starts with liquidity. Banks can offer 90% house financing because thousands of otherwise illiquid mortgages are packaged into tradable securities; no equivalent secondary market exists for a single GPU, and “if you can’t get a loan for a single GPU, you can’t get a loan against a large cluster.”
Traditional issuance is too slow for the hardware cycle. By the time loans are accumulated, tranched, marketed, and sold over roughly two years, the collateral may have progressed from Hopper to B200, Vera Rubin, and toward Feynman—“the ones they haven’t even announced yet.”
Below roughly $200 million, David says institutional GPU financing effectively disappears. One emerging data-center operator received a five-page Guggenheim or Jefferies explanation for why its $200 million purchase could not be financed; private credit is retreating, while Oracle is portrayed as already “max levered out.”
4. Expensive equity fills the gap that asset-backed debt should occupy
Without dedicated GPU credit, borrowers resort to convertibles that can become equity. David’s example: early CoreWeave or Crusoe holders may suffer roughly 90% dilution because issuance becomes the only available way to fund equipment.
GoodAlexander calls this the “Bitcoin pizza dynamic.” If a company sells equity at a $1 billion valuation to buy $100 million of GPUs and is later acquired for $60 billion, it effectively surrendered about $6 billion of expected value for that hardware.
His Finance 101 conclusion: a founder bullish on the company should preserve equity and finance the asset directly. GPUs also carry large tax value because they are depreciable capex; David cites Meta’s roughly $20 billion of revenue alongside $20 billion of capex as an almost deliberate pairing.
David points to bonus depreciation and other tax incentives as an important part of the yield economics. The speakers’ broader argument is that cheaper GPU financing avoids dilution while preserving the depreciation benefits of the underlying buildout.
5. DeFi turns slow securitization into a continuously priced “debt perp”
David’s answer to “why crypto?” is speed of capital formation: DeFi can “spawn markets out of nothing.” A roughly $1.5 billion CoreWeave bond had traded only about $50 million since August, while USD.AI’s much smaller system allegedly produced around $15 billion of secondary volume in six to eight months.
The multiplier comes from composability. Once USD.AI and sUSDai exist, users can trade them, strip yield through Pendle, loop positions through Fluid, and create implied notional exposure far beyond the original loans.
Each new loan can enter the debt instrument or sUSDai’s NAV immediately rather than waiting for thousands of assets and a roadshow. David contrasts Hyperliquid’s futures perpetuals with USD.AI’s ambition: “We’re really like a debt perp,” continuously pricing and compounding newly originated credit.
The structure is intended to broaden access beyond white-collar bond desks. David’s claim is not merely that blockchains distribute old securities more widely, but that immediate pricing and settlement make a previously unavailable small-ticket GPU market possible.
6. USD.AI wants to discover—and eventually subsidize—the interest rate of AI
Chapter one is explicit: originate about $1 billion of loans over 12 months, becoming the reference rate for GPU credit. To David, that rate is “how fast you can grow AI”—the cost of capital and the amount people will spend for a return.
He expects ownership to spread down the stack. Rental providers are buying chips; token factories that amplify output per GPU want ownership; LLM companies already buy; applications will follow because indefinitely renting their largest cost is too expensive.
David puts compute at roughly 80% of OpenAI’s cost. He says $500 billion of Nvidia equipment becomes $600-700 billion after interest, making Wall Street effectively AI’s second-largest business behind Nvidia and ahead of OpenAI.
If customers settle in USD.AI, its 3-4% income from minted dollars can subsidize debt rates. Revenue still comes from origination fees and net interest margin. GoodAlexander frames CHIP as governing fee routing, rate decisions by deal size and off-take quality, and possible redemption priority; fuller utility details remain forthcoming.
7. Real customers are being pulled on-chain by financing, not ideology
David reports about 70 incoming customers, including large neoclouds, with demand increasing daily. The immediate constraint is enough protocol liquidity to serve them, hence the push to deepen sUSDe and USDe markets after CHIP’s launch.
Many customers have never used a wallet. The team introduces them to Coinbase Prime and, where European access failed, even sent free Ledgers; initial confusion gives way to amazement that they can borrow against GPUs, buy more GPUs, and repeat through a smart contract.
The founding insight came while David mined Bittensor. He says paying about $3 an hour for a rented chip amounted to more than $25,000 annually, while the GPU itself cost roughly $25,000: “Imagine if you bought a house for a million dollars, but your rent is 100k a month.”
The host disclosed that he previously held and traded CHIP, made money, and held none during the discussion; he remained cautious on crypto alts. David’s own emphasis was operational: “I’m just doing more loans,” with most fees expected from AI companies rather than crypto users.
8. AI capacity is becoming a sovereign resource and potential monetary anchor
David expects jurisdictions to become net importers or exporters of AI tokens, meaning generated inference rather than crypto tokens. A country’s token-production capacity will increasingly determine citizen productivity, economic output, and strategic autonomy.
His petrodollar analogy begins in the 1970s: oil powered cars, heating, factories, and national growth, so controlling its settlement conferred monetary and geopolitical leverage. AI now plays the analogous role because a citizen equipped with it may have “10x” the capacity of one without it.
David states the conclusion categorically: “If you don’t have AI, you’re not a country.” The extraordinary data-center buildout is therefore not just corporate capex but an arms race to keep the underlying resource continuously available at massive and smaller scales.
9. The crypto bull case rests on inflows, censorship resistance, and capital flight
GoodAlexander moved from mocking Michael Saylor’s promotion to embracing Taiki’s STRC thesis. He believes Saylor has “cooked this inflow machine”: an approximately 11%-yielding instrument aimed at retirees that turns older savers’ capital into mechanical Bitcoin purchases and can expand MicroStrategy’s mNAV flywheel.
He supports the demographic point with Asian retail behavior. Crypto Twitter is about 93% male, he says, while Japanese and Korean participation is nearer 50/50; older Korean women are meaningful speculators because years of weak local equities and popular forex trading produced a different investing culture.
Despite exchanges targeting Monero and pressure around privacy projects, GoodAlexander says Monero and Zcash have continued to work; Zcash’s presence on Robinhood reinforced his view that “you try to kill these privacy coins and you can’t.”
His horizon map is concise: one year of Saylor-supported inflows, three years of capital flight, and five years for the future to be built on crypto rails. Money printing, wealth taxes, CBDCs, and capital controls push wealth outward; unlike gold or oil, crypto can cross a border without a physical bag.
His long-run macro frame is darker: he believes US governance and property-rights enforcement have deteriorated, implying an eventual 11-13x earnings terminal multiple rather than roughly 29x. He says living in Puerto Rico and keeping wealth outside the system are ways to remain robust through that transition.
10. The AI-IPO supercycle may end by turning passive equity into high-FDV crypto
The host compared current markets to Solana’s “daily-runner” era: one speculative ball ricochets among Car, Hims, Texas Instruments, and other names, while the S&P’s relentless passive inflows distinguish it from crypto. GoodAlexander’s answer is that the frenzy can persist until its largest liquidity event arrives.
David estimates that SpaceX, potentially at $1.5-1.7 trillion, is “the main show,” with OpenAI and Anthropic following. He is more bullish on Sam Altman’s ability to engineer narrative and stock price than on Dario Amodei, and expects the sequence to be SpaceX first, OpenAI second, and Anthropic last.
David says banks and companies are incentivized to upgrade AI stocks and engineer impressive enterprise deals because weak comparables would threaten enormous IPO fees. He gives possible examples such as JPMorgan doing an enterprise deal with Cursor or Cognition selling a “kajillion” seats.
The danger arrives if SpaceX is rapidly added to the S&P 500 or Nasdaq, forcing retirement portfolios to buy it without company-level diligence. David calls that “imagine if EOS got added to the S&P 500”: a high-FDV crypto launch embedded in passive finance, followed potentially by a “generational top” and “the mother of all shorts.”
Full transcript
Alex, this is an incredible view, man. I’m getting view-mogged, dude. I have this new lighting setup, and I’m getting view-mogged right now. This is insane.
I feel like that would be your basement in this apartment. It’s just a different setting in the same location.
Yeah, there we go. It’s good to see both of you here. First and foremost, David, you were on what—maybe in October or November? You first came on, and I first met you. I thought what you were doing was really interesting. The whole USD.AI concept was pretty wild.
After that, a lot of smart people—Taiki, Alex Good, a lot of people—were telling me about it. Then you just had your TGE. You just launched CHIP, which is exciting because it’s the first hyped crypto TGE since, I don’t know, Aster that hasn’t gone down in a straight red line yet. It’s only been a day or two, but not going down in a straight red line is pretty exciting.
There were some pretty cool crypto names who were excited about it, GoodAlexander being one of them, so congratulations on the launch. Do you want to start with a high-level overview of what you’re working on and what you just launched? Most people have some context, and then we can get into the fun stuff.
For sure. For those who weren’t listening last time, we’re effectively trying to solve a capital issue that AI has. You see Claude getting kneecapped every time you try to use it because there isn’t enough compute. OpenAI seems to have caught up in terms of quality, and even though they have this Microsoft’s hosting in the back, they knew from the very beginning that, of course, you can have better and better models, but at the end of the day, it’s going to come down to access to how much compute you have.
You’ll never have enough compute. People might say it’s a power issue, or the quality of the LLM, but really it’s going to be how many Blackwells you have access to. That’s why you’re seeing this flip literally this week.
The problem is that we’re trying to figure out how to enable this abundance by making chips easily accessible—not necessarily just to OpenAI, but to anyone who wants to buy a chip. It’s kind of like a mortgage for GPUs. We solve this by using stablecoins to finance this wave of capex build-out.
A lot of people overlook the fact that T-bills are the world’s fastest buyers. Stablecoins are the fastest buyers of T-bills in the world. Right now, one of the biggest sources of spending in the U.S. economy is AI capex. If we weren’t spending this money on AI capex, our GDP growth would be negative 3% or negative 4%. It has that big an effect on the U.S. economy.
That should be financed with the fastest horse, which I believe is stablecoins. That’s why we built USD.AI. You have two megatrends: the capital formation of stablecoins, and the AI capex spend in the U.S., which is going to be trillions and trillions of dollars.
If you sum up the last 10 years of infrastructure spending—oil, highways, and everything else—one year of AI capex is more than all of that. The amount of money we’re spending is just infeasible. I used this metaphor last time on this call: if you were to visualize the Las Vegas Sphere, the most expensive stadium ever, 30 of those would be one data center. That’s just so much money being put into it.
Yeah, one gigawatt is like $50 billion, and then that build-out is two or three gigawatts. You have these being built around the world—50, hundreds of these. It’s like Project Hail Mary is literally being done. Have you seen the movie? But instead of trying to save humanity, you’re trying to have the best model. It’s an insane amount of spending.
It’s the most we’ve ever spent since the build-out of China. It’s hard to see because people keep talking about data centers, but I don’t think people realize the hyperobject of this.
Thirty Sphere build-outs for one data center is nuts. Alex, by the way, why are you mogging me right now? You looked fresh last time. What’s going on here?
People kept commenting on my background. They were like, “You’ve got to leave your room,” and I’m like, “I don’t want to leave my room because you don’t want to see what’s behind me in real life, because you’ll feel bad about your location.”
Now that we’re in the mogging era, I thought it was appropriate to reveal the state.
It’s got to be the double jaw. I haven’t pinpointed it yet, but I’m talking to Razmig Bacarian [?], and we’ll figure out what it is.
Give me a primer before we get into some of the details. You’re definitely one of my high-signal proxies for AI stuff. If you say something about AI, I generally pay attention to what you’re talking about—maybe except for Baba Long—but in general, I do.
You were, and have been, pretty excited about USD.AI, CHIP, and the whole ecosystem. Explain why you’re here and why you’re excited about it.
David and I have lived next door to each other—we’ve been neighbors since 2021.
I’ve been super into this ever since David originally had a financing platform, a blockchain-based financing platform for NFTs. He built this really elaborate credit mechanism for the worst credit in the world, which is basically guys borrowing other people’s profile pictures on the internet. It’s the exact dude you don’t want to lend to, so he built a really good system for that.
When NVIDIA started talking about this whole sovereign AI thing, what you might realize at a very high level is that the holders of Tether—the UAE, Malaysia, and these international jurisdictions that are mostly the stablecoin holders—are also extreme investors in data centers. They’re extreme investors in NVIDIA graphics chips specifically.
This is the perfect crypto financing use case because there’s weird credit, warehousing issues, and a completely new category that you can invent. For example, Canton Network is a very hot coin.
Yeah.
That’s the old economy. That’s trading Coca-Cola fixed income on a blockchain. There are a lot of people who trade Coca-Cola fixed income without a blockchain, and you’re like, “Well, anon, check out this cool new private network.” They’re like, “How much can I farm?”
With USD.AI, it’s more like, “Nobody really knows how to finance graphics cards right now.” Before David showed up on the scene, a lot of this was done through private credit and unsecured lending.
What got me excited was that these are warehouses. Previously, when I worked at Palantir, I used to work in trade financing. I studied Standard Chartered, receivables, credit, and trade financing, which is one of the biggest forms of financing in the world.
GPUs are firmly warehouse-space credit products, so I thought, “This checks all the boxes.” It’s a new category, it has an asset-backed security element that’s superior to the market’s current treatment of the asset through private credit, and the existing holders align with the Tether holder base.
I counseled David back when we were in a sauna. He said, “I’m thinking of pivoting into AI,” and I said, “You should absolutely do that.”
Obviously, I was younger and happier, and things were very optimistic. But as we’ve evolved in the market, shit’s real. Everything is no longer a meme. All these things are actually happening, and that’s how I got into CHIP and USD.AI.
Right now, it’s exciting because even a couple of weeks ago, the consensus in the market would have been that Opus, Mythos [?], and OpenAI were the god-tier models. Random people buying GPUs, and a lot of the things people in crypto were saying, were misguided.
You’d ask, “Do you really need to buy your own GPU if OpenAI is the best thing in the world and you can just access it through an API?” Probably not. But in the last 2 weeks, there’s been this nonstop barrage.
What caused the shift?
For a while, people were saying that China was just distilling Opus. The narrative was that these Chinese models were just Opus distills, that they were stealing our shit, and therefore they weren’t worth paying attention to. What was going to happen was that Mythos [?] was going to be a thing, and they were going to shut down the distills.
So Anthropic put out these releases saying, “With no more distills, we’re shutting this down.” And then there are all these releases coming out that are better than Opus in a lot of different ways—in terms of efficiency, in terms of utilization. The fact that you can run one of these models with Opus-like or Sonnet-like performance on a $10,000 graphics chip is a complete game-changer. There are guys doing this on 3090s now, right?
Damn.
A nicer card than that.
Yeah, and so it’s like—the local models. These are real. I didn’t even believe it because, with Chinese stuff, they do have a tendency to over-promote these releases and game benchmarks. But I tested this stuff out because I didn’t believe it, and these models are amazing.
I actually think the market still largely hasn’t realized what this means because people are still very much in this mindset that we literally aren’t going to be able to afford Opus. My Opus bill, I was saying, is $80 per hour per terminal if you’re actually using Claude with proper zero-data-retention protection—ZDR—so your IP doesn’t get transferred to Anthropic and you have strong protections. If you’re using that via an API without the Claude Max plan, or whatever, and you’re just paying per token, it’s $80 per hour per terminal. A lot of guys run two terminals, so that’s $160 an hour.
Jeez.
Right? That’s extremely expensive. To put that in perspective, this machine—the new models on Qwen 3.6—is $2 an hour. You’re looking at something like a 40–80x delta, and a lot of these models are state-of-the-art, SOTA.
I’m like, “Oh, shit.” And no, I haven’t really used GPT-5.5 a lot yet.
Yeah.
But I think that’s what’s new in the market. I think everyone—SoftBank and a lot of the Anthropic secondary market—all of this stuff has already had this crazy run. I just don’t think the open-source stuff is really priced in or appreciated by a lot of people in the market.
It’s really good for crypto AI because we aren’t structurally capable of serving closed-source models. Especially with the Mythos launch, where they’re gating it—imagine trying to buy Mythos on Venice AI, for example. It just doesn’t make sense, right? You have bio-authenticated, closed-source models. Crypto AI would really be in a bad spot if that were the future.
So, yeah, I think this is great news. I’m super excited for the first time in a while because I actually thought OpenAI had just won. And, yeah, that would suck.
That would suck.
David, go ahead.
Yeah, I just want to add that it’s kind of why I thought having both of us on would be nice. We both come from Puerto Rico and seem to share something just like Jeff from Hyperliquid. It’s a very miserable place, but I love it at the same time, to some degree. It’s terrible, but it’s lovable.
This is kind of like the new wave that we want to push. It’s like the exciting times of early DeFi. I think there’s such a rut in the ecosystem right now. People are surprised that our project did decently well in the market, and I’m like, it’s because there’s kind of been a cesspool of horrible projects, obviously, and the post-memecoin era, with all this venture-backed vaporware whose goal is just to extract money from public launches and hedge it on perps.
But there is still a future. I think a lot of people have lost that glimmer of hope. You’ve seen so many OGs leave the space. Crypto AI is definitely where we’re seeing interesting use cases. You see them with Venice, people are starting to see them with us, and soon with Post Fiat. There are actual users coming about and organizing around this new wave of hope.
What can we do to show the world that the things we’re doing actually have an impact—not necessarily just in the sector we’re in, but out toward the frontier, toward what we actually see in AI?
How big is the market to finance GPUs?
I’ll give you an example. NVIDIA has $500 billion of sales in the next 5 quarters. Ninety-nine percent of that is financed.
Next 5 quarters. Okay.
Yeah, and about 20–30% of that is from emerging clients—that is, smaller participants. They aren’t the hyperscalers like Amazon. We were small at one point. One of their loans was around $50 million.
It’s basically all these participants. You really can’t finance anything under $200 million. We were chatting with this pretty emerging data center, and they had a 5-page pitch from Guggenheim and Jefferies explaining why they couldn’t get financing for a $200 million GPU purchase. They had a call with us.
And why can’t they get financing for it?
This is best explained through a metaphor. When you get a loan for your house, the bank you get the loan from isn’t the one holding the loan. When you get a mortgage, it’s crazy, right? You put 10% down and get a 90% loan on your house because they batch it up, package it with 10,000 other loans, and make it tradeable.
It’s much like wrapped stETH with ETH staking. You’re staking your ETH and it’s stuck, but if you have a derivative, then—
Let’s stick with the mortgage market, David. Maybe that’s not the one you want to go with.
Yeah, but I mean, The Big Short—that’s all interesting combinations. The main idea is that you can make something really liquid for an illiquid asset. Houses are not liquid, right? How many house sales are there versus the value of the houses? It’s very, very low.
That exists for airplanes, houses, and cars. You can get a loan against a Honda Civic, right? But you can’t get a loan for a single GPU. And that’s a problem. If you can’t get a loan for a single GPU, you can’t get a loan against a large cluster.
Why is this? It’s because you don’t have the thing I just described: a mortgage-backed security, something that batches up all the loans and makes them tradeable. Why do you have this for airplanes, cars, and houses, but not for the most important commodity in the world today—the basis of national sovereignty, AI?
Why don’t you have this? It takes about 2 years to issue that instrument. You have to get 10,000 loans, cut them up, tranche them, go on a roadshow, and then trade and sell them. I explained the whole thing. That just takes too long.
If it takes 2 years, you’re past the Hoppers, the Blackwells, and the B200s. You’re past the Rubins, the Vera Rubins. You’re now at Feynman, which they haven’t even announced yet, because of how fast technology can improve.
Because you don’t have the secondary debt market, which is the basis of how you get that easy, simple mortgage, this is like U.S. housing in the 1910s. You didn’t have mortgages. You did a 5-year loan, and the banker could call that loan at any time. That’s how it worked. You put 50% down, as opposed to 5%.
This, too, is like the Bitcoin pizza dynamic, right? Imagine if you’re Cursor and you’re raising a billion-dollar valuation to buy some GPUs. You buy $100 million of GPUs, sell it at a billion-dollar value, and get acquired for $60 billion. You just paid $6 billion of expected value for $100 million of GPUs.
If you’re actually bullish on your company and selling at a high valuation—right now the venture market is very hot, but the reality is there’s a reason it’s hot, and all of these companies are going up a lot. A lot of these Korean stocks are up massively. These equities are crazy. People are saying SK Hynix, Samsung, and some of these other companies are going to earn their entire market cap in the next 3–5 years.
The earnings of a lot of these companies, unlike in the 1999 tech bubble, are quite high. So, do you really want to sell equity? At least you want to have the option to finance the GPUs, right? You don’t necessarily want to sell equity or other types of credit when you have this alternative financing mechanism.
It’s unfortunate because, for the most part, people actually do have to sell equity to do these deals. It’s a weird category because a lot of the companies that would be going into the traditional debt market have underwriters saying, “Who are you? What is Cursor?”
Yeah.
You know? I think it’s kind of a cost-of-capital thing, right? Very basic Finance 101.
You're bullish on your own company. You don't want to issue unsecured credit. You know that you have to make an asset purchase. Why wouldn't you do asset-backed financing like every other industry? This is common.
I used to work in the e-commerce space, and a lot of really basic things, like lawn mowing or garage work, where you know there's going to be a return—all of that is asset financed, not only because it's efficient, but because of taxes. This is something that people, I think, also don't understand: these are fully depreciable assets with huge tax implications.
Because of the new tax acts, David, I don't know if you want to talk about this, but it's a material part of the yield component of this credit instrument—some of the recent tax bill changes.
Yeah, that's kind of why you see Meta have almost a perfect pairing of, say, $20 billion of revenue and $20 billion of capex. It's kind of on purpose, by the way, because they can offset all of that.
I said it completely.
Yeah, because of bonus depreciation. These are capex assets. We're really trying to encourage the development of stuff, and I think us both being in PR kind of already enforces that: you can definitely change human behavior with that kind of incentive structure.
But for those that don't understand, when you buy—I mean, a G-Wagon also qualifies, by the way, because it's a weight class where it qualifies as a credit—
But then only—
Yeah.
Can you—sorry, finish, finish.
Yeah, I would just say, too, that we're pushing whatever we can to encourage the development of AI capex because it is such a massive black hole of money. There's so much money being put into it, and we should definitely put ourselves in the center of this.
Can you explain, David, why crypto is necessary to solve this? And why is crypto uniquely positioned here?
Crypto is unique because DeFi, specifically, is the sector that I tackle. It is the fastest horse. It forms capital in a way that is hard to recreate even in TradFi. You can just spawn markets out of nothing, which is difficult to do—you need to have a lot of infrastructure and a lot of participants.
I'll give you a good example: a CoreWeave bond that traded $50 million since August. It's a $1.5 billion tranche. It's a tradable debt instrument. We ourselves were much smaller, but in the last 6 to 8 months we traded about $15 billion in secondary-market volume.
Because the moment you create USDai or sUSDai, you can put it on Pendle to strip out the yield, and then you can trade in the implied notional value. You're doing looping on Fluid or these other platforms, and you end up creating this massive volume.
What the tech really develops is, one, perpetuality. Of course, you have HYPE, which is the futures perps. We're really like a debt perp. We can instantly price and compound things, which is why whenever there's a new loan, it gets added the moment it's originated.
You don't have to wait for 10,000 loans to be batched up. The moment a loan is created, it's immediately priced on NAV and given to our debt instrument, or to sUSDai. So it's just a really, really good financial-tech product, which is the way finance should be run, because it makes things extremely fluid and tradeable.
You don't need to have just white-collar guys trading it; anybody can do it.
Is TradFi struggling with GPU funding, or is that one of the hardest? It is. TradFi is struggling with it, especially on the lower end, with the smaller contracts, like you said.
It doesn't even exist on the lower end. It doesn't exist. You can't get financing. If it's under $200 million, you might get it if you work with a private credit fund.
By the way, every private credit fund is in retreat. Oracle is fully tapped out. They're max-levered out. They've maxed out their credit card and their term loans, because you have revenue and you can borrow 6× against it.
So you're literally seeing the very edges of pretty much every financial vehicle to raise money to build out AI capex reaching its limits. It's like you're playing hot potato, where you're trying to make sure your hand doesn't get burned, but people are playing solo, right? They're literally just holding the hot potato—
So, do you just get stuck with suboptimal financing because you're the only one that will underwrite these GPUs?
Oftentimes you have to do a convertible, meaning that you get a loan, but then they can turn it into equity and you get diluted. A good example is if you were a seed investor in CoreWeave or Crusoe, you were diluted almost 90%—10 times more equity had to be issued because that's the only way to finance it, because you don't have any other option.
So having non-dilutive financing is pretty interesting, but generally the solution doesn't even exist. It's why we exist as a company. It doesn't even exist as an offering in traditional ecosystems.
People have described us as a monopoly because nobody else even comes close to offering this stuff, which is smaller than Apple. I'm not going to say who, but it was definitely not—
I'll ask you in a minute, David, but I'll go back really quick: what is the win condition for the company at large, and what is the purpose of the chip? How do you 10× or 100×—not price, but overall size—from here? How does it get massive? This is to you, David.
Well, there are 2 angles for us. There's chapter 1, which is the next 12 months, where we do $1 billion of loans. That's our ultimate goal. We want to do enough loans that people start referring to us as the reference rate for what the interest rate should be.
That's what we're trying to do: if you do enough loans globally, you discover this interest rate. When I say interest rate, I mean the cost of capital—how fast you can grow AI. That is what you're trying to discover.
It's really hard to know what the interest rate of artificial intelligence is. It is literally the cost at which things can grow: how much money you're willing to spend for a return. Once you discover the interest rate—which is us doing $1 billion of loans—how do you then weaponize this to penetrate into the rest of AI?
I'll say this, too. We're starting to see that every AI company we talk to will eventually own GPUs. We started talking with these rental companies a year ago, and the GPU rental platforms are starting to want to buy chips.
They want to rent them out to token factories, these guys who can really amplify, like the Base10s of the world. They amplify how many tokens you can generate per GPU. Now they want to buy chips, and LLMs are already buying chips. The apps are going to buy chips because it's really expensive to rent them.
What we're trying to do next, after we discover this interest rate and offer financing for smaller participants, like applications or LLMs, is figure out how to subsidize it—how to lower that interest rate more than the market, like a central bank in a sense. Lower rates, things grow, and people spend more money.
The number-one expense in all of AI is compute. Just to clarify that: 80% of OpenAI's cost is compute. Salaries are very high, but compute is way more expensive than the salary part.
This is expensive because Nvidia chips are expensive, but also because all of it is financed. The $500 billion of sales, like I said—it's actually $600 to $700 billion because of interest rates, because it's expensive.
Much like a house, you're paying $3 million for a $1 million house across 30 years. In this case, it's just really high interest rates, and everything is debt financed.
So if you can lower this expense, which is very simple because you're just paying Wall Street—Wall Street's getting the second-largest business in AI, call it: Nvidia, Wall Street, and, I'm sorry, Nvidia, Wall Street, then OpenAI. That's $20 billion of revenue versus $500 billion, and then you have $150 billion right in between.
How do you lower this second-largest expense in the AI sector? We think it's by settling in a stablecoin. I described a debt instrument as USDai. We're giving out loans against chips.
But what if people started paying in a stablecoin that I created? This is what USDai is—not the staked version. If you can start paying in my currency, I can subsidize the interest rate, because I earn 4% or 3% on every dollar that's minted.
Much like Circle and Tether, they just pocket that, right? They just keep that money.
I'm saying, why don't you have the entire AI sector settle in a stablecoin by subsidizing their number one cost that they can control, which is debt?
How do you make money?
Make money? I mean, we still earn an origination fee on the debt. We also make a net interest margin. But eventually, in the longer run, we're not giving 100% subsidies. It's much like the growth pains of any stablecoin beginning. You're always cutting deals, but eventually you just earn the yield.
But if you can subsidize the interest rate of AI, which again is how fast AI can grow—the cost of capital—I think this is a second line item in AI, but it's also the second-largest business.
And what is the purpose of the chip token? You talked about it real quick.
That is kind of like the GP position, in a sense. This is the best metaphor I can give. We haven't really detailed it; we will over the next few weeks. But it controls the fees that are related to the protocol and how they're routed.
It also determines what the interest rate should be, which I think is very powerful. You want to know what the rates are for small deals, large deals, better off-takes, and worse off-takes. There will be some prioritization structure as well around redemptions. That's usually where the best token utilities are, when it relates to prioritization.
Transaction fees—there are a number of use cases that it governs, of course, but it also has decision-making power in how the fees are routed. But that's up to the users, not necessarily just us.
Cool.
On the crypto topic, Alex, I feel like, broadly, you are one of the more optimistic, forward-looking, crypto-specific people on my timeline recently. I don't know if that's surprising.
Yeah, actually, I think Taiki got me into the thesis early on. I kept seeing these Michael Saylor yoga-instructor videos, and I'm like, this is just late-stage Madoff. It's so bad. I think I pinged Taiki and said, “This is just late-stage Madoff. It's so incredibly bad.”
He's like, “No, you don't understand.”
And I'm like, “What do you mean?”
He's like, “STRC can't go bankrupt because it's a really smart convertible preferred stack, which is based on the mNAV.”
And I'm like, “Okay.” So I looked into it, and Saylor has cooked up this inflow machine with STRC.
I know everyone is wrecked. Everyone has given up hope. But Saylor can't give up hope. Maybe it's because he owns too much Bitcoin, or he's an older man without a wife. Who knows what the reason exactly is? But he has developed a very, very effective yield-inflow mechanism.
There's a funny Economist chart showing that the Silent Generation is substantially wealthier than millennials. It's pretty bleak. Millennials and Gen Z are pretty poor relative to older people. If you really want to get inflows into an asset class, you need retirees to bid.
To be honest, not to derail on CHIP, Upbit volumes are very high. This is something I've kept track of with XRP. A lot of the buyers are older. Older Korean women specifically.
Women?
Yeah, it's a big retail market: the hajumma.
It's very different from U.S. retail. Japan and Korea both have a more gender-neutral composition. If you look at crypto Twitter demographics, it's 93% male. If you look at Japan and Korea in terms of the retail composition, it's closer to 50/50.
On a capital basis, on a trading basis, you might be seeing an even higher female skew because of cultural dynamics. Japan, for the longest time, and Korea have had very poorly performing stock markets over very long periods. That led to forex trading becoming popular.
Both of them have famously bad work-life balances, and it's led to women adopting financial speculation in a way that hasn't been true in the United States.
These demographic points are important because, with Michael Saylor, the question was always: who's the incremental buyer? All the memecoin guys are cooked. None of the crypto AI protocols, at least as of Bittensor, have excited me. I was like, “Okay, the AI buyers are not here.”
The Koreans are bidding AI stocks in Korea. So the question was always: who's going to buy this thing?
The STRC catalyst was big enough that you're like, “All right, this is getting multiple billions of dollars of inflows.” If you were to put a billion dollars into the current tape, it's going to move the market. Furthermore, if you've seen MicroStrategy's mNAV expand, that creates a flywheel.
This is just mechanical. A lot of the recent bounce has been ETH/BTC near the lows. This is a mechanical BTC-driven pump.
The other element is that there was a head-fake conceptually, where everyone bet on the institutions. They bet on Trump, they bet on the banks.
Yeah.
And it was just a rug, right? World Liberty Financial kind of was a rug.
The good news is that all of the reasons why crypto existed prior to that are still there. Historically, if you study the crypto asset class and ask, “What is the driver of the price of Bitcoin or crypto market cap over time?” it's mostly people moving money out of the system.
It's Chinese whales mining Bitcoin. Over the last year, you've seen basically every exchange come after Monero. The Europeans came after Monero. Spagni's done prison time, and Zcash has worked. Monero has worked despite everyone coming after them.
Zcash is on Robinhood today.
Exactly. You try to kill these privacy coins and you can't, which is exactly what you want to see for the forward outlook of the asset class. If you try to kill these things and then fail, you have a picture that this is going to last and be able to work in a regime that is not friendly to capital.
What does that mean? It means the Democrats winning, it means CBDCs, capital controls, wealth taxation, and unfriendly policies toward capitalists.
From here, what is going to cause us to go incrementally higher? Right now, everyone is convinced the S&P is a moon machine and that you don't need crypto because you have equities. In the near term, that's great because it means Saylor is going to keep issuing STRC, selling 11% yields to 75-year-olds, and getting billions of dollars to max-bid BTC.
In the near term, we're going to be fine because Saylor is literally an ATM that props up the market. In the long term, I also think it's going to be fine because we have the privacy technology, and we have the outlook of the G10 turning into China.
You're going to have extreme fiscal pressure because the fiscal situation isn't getting better anywhere in the world, and Iran is making it substantially worse. Across Europe and Japan, with energy subsidies, the global fiscal situation was quite bad before, and it's gone from bad to much worse.
The way that's going to be paid for is ultimately through taxation and money printing. Those are things that drive capital out of systems.
I'm like, “Okay, where's the money going to go?” It's going to go into gold. It's been going into oil. But ultimately, you can't easily leave a country with gold or oil. You can't get on a plane with a bag of gold. We saw this in Dubai.
Saw this in Dubai, yes.
And there's going to be a bid, in my opinion, from the ultra-rich to move money out of the system, and that bid is going to go into crypto assets.
That's sort of my thesis. It's not necessarily optimism about CHIP. I think the stuff like CHIP is gravy on top of it. We are in an internet economy. There are open-source technologies and transactions that can be built on top of blockchains, and it doesn't make sense to build a new credit market without a blockchain.
Why would you construct a new credit system in this legacy technology stack when you know that you probably need to move to the UAE in 3 years or move to Latin America? We know that the future is going to be built on crypto rails, which is maybe a 5-year story.
In a 1-year story, we have Saylor. In a 3-year story, we have capital flight. So on a 1-, 3-, and 5-year time frame, I feel very, very good. Right now, it's quite contrarian, and you should be there pounding the table when other people are depressed, because that's when you make money.
How do you not drive yourself crazy with this bold thesis sort of cemented in, where everything is pretty fucked? They're going to hyper-print, it's late-stage capitalism, but crypto—and risk assets in the interim—are going to pump really, really hard. How do you think through that? How do you not drive yourself nuts thinking through that?
I mean, I think a lot of the really exciting areas to speculate in such an environment are—you know, I think the answer is you keep yourself busy.
Like things like Avis rental, right? This recent 850% increase—like a lot of the retail feeding frenzies that you're seeing—are exactly what you would expect in a late-stage capitalist economy, right? It's kind of like all financial meaning has been reduced to meme stocks and meme-like asset price movements.
I think SpaceX, OpenAI, and Anthropic—you've got like $4 trillion of equity coming onto the market. And so it's going to be the biggest period of speculative IPO excess in financial history by an order of magnitude. None of this is in the S&P 500 right now, right? So if you think about it, the S&P and the passive-investing markets need to absorb $4 trillion of bids, right? And you're like, okay, I don't know if that's even possible.
The bleak thing is that you can predict there are going to be a lot of people pushing assets really hard. There are going to be a lot of fundamental investors who get short-squeezed on things like Avis. And you're going to have a lot of dislocations where people are screaming, “This shouldn't happen,” and all kinds of GameStop-like events.
By the time it ends—when it does end—it's going to be the mother of all shorts, right? It's going to be just like 2021. We know that equities have turned into crypto, and as crypto people, we're super jaded. We see these AI conferences and kind of know, “Yeah, this is 2021. I know what happens next, and trust me, it's not great.”
Most people don't know that, right? So I think what's exciting to someone like me is, okay, you're going to monetize a lot of the insanity until the SpaceX IPO and the OpenAI and Anthropic IPOs. But after these events, you're probably looking at a generational top in equities, where you're going to make so much money shorting this thing into the [__] ground, right?
Ultimately, my simple frame is that governance in the United States has declined substantially. I think the government is not trustworthy and doesn't enforce property rights effectively. Even things like the supply-chain risk classification toward Anthropic shows a government that is capricious and random. In the long run, we know what that model looks like in China and other countries that are not property-rights-friendly.
It's an 11-, 12-, or 13-times earnings terminal multiple, not a 29-times multiple. The long-term playbook for me is to assume that. That's why we live in Puerto Rico, right? We don't believe that the system in the long run is sustainable. And the reason I'm in crypto is because I don't want to have money in the system after it blows up.
Yeah, yeah.
And so that's how I think about it. I'm just building a life that's robust to the system going through a major transition to something that isn't that awesome.
Let me ask you one more on this, and then I'm going to go to David. You're kind of rolling. How does this play out in the interim? One thing that's been fascinating to me is that I've never really paid much attention to geopolitics, and I've spent the last 2 months trying to figure out how to monitor the situation and these headlines.
You see fake news everywhere. You see a market whipsawing on Trump tweets, most of which are fake. And you see a market that's basically completely shrugged off a war and just a generational bid on the S&P. What happened today was nuts. The bid is so insane, it makes you start believing in a higher power almost.
At the same time, you see this—I almost compare it to the daily-runner era of Solana—where there's a hot ball of speculative money that whipsaws from Car to Hims to Texas Instruments to thing to thing to thing. And now we all know how this ended on Solana.
But there's a difference where our majors are BTC and ETH versus the S&P, which has, like you said, this passive, forever, nonstop, aggressive flow. They're printing and they're buying, and they're printing. It is nonstop up-only. So how does this manifest in positioning and price action over the short term here? How long can a hot ball of money whipsaw around?
Well, I think the very near-term catalyst—I think SpaceX is going to suck a lot of liquidity out of the market in the U.S. My crypto proxy is that SpaceX is the TRUMP coin.
It is. It's like, yeah.
It's obviously the TRUMP coin.
The face, right?
It's like the xAI thing—manufacturing the financials to be ridiculous. The reason Elon acquired cursor is either to manufacture $10 billion of revenue or capacity utilization for SpaceX, or to juice it and allow him to point to an enterprise billing curve similar to Palantir that he can hype up.
He's going to announce all these things. He's trying to front-run Sam Altman and say that they have a $600 billion AI research lab. So when you're buying Starlink, you're buying the future in-space data centers. You never want to fade Elon.
I actually think Trump is probably amateur hour in terms of Elon. But to be fair, this is a $1.5 trillion to $1.7 trillion IPO. In my opinion, this is the main show, and everything leading up to it is basically a dance before the main DJ comes on the set. We're still on the pre-set. The main Tiësto is coming, and that will be the main set.
My drugs haven't even hit fully yet.
After that, it's over. I think he's going first, and you're going to have this series. Then you have 2 more catalysts: the OpenAI IPO and the Anthropic IPO. They're going to be systematically pumped.
I'm more bullish on Sam Altman's ability to engineer narrative and manufacture stock price than I am on Dario. I think Dario is kind of a weird guy who crashes out and sends weird emails. The Anthropic thing has had a lot of self-owns.
So basically, Elon's going to go out first. Sam is going to go out second. And then Anthropic will be the last one, like, “Oh, that's it.” And I think after that—the answer to your question is that after the SpaceX IPO, you start getting very bearish on equities, right?
That's sort of similar to Solana. Basically, the SpaceX IPO is going to be the Solana $300 moment, right? That's where we are right now. We're in this max-bid environment where anything's possible. Everyone is incentivized—every investment bank is incentivized—to come out and upgrade every single AI stock.
Insane.
Because they're going to get so many [__] fees when SpaceX IPOs and OpenAI IPOs. They don't want any of these comps to look bad, right? So JPMorgan will do some insane enterprise deal with Cursor, or Cognition will sell a kajillion seats to financially engineer these IPOs. Everyone's in on it, and everyone is incentivized to make these numbers look good.
Everyone's in on it.
And when it ends, nobody has any reason to play anymore. So that's my mental model: it's bullish until you get $4 trillion of liquidity dumped.
The wake-up—the realization that people are going to have—is that Elon is basically working with the S&P and the Nasdaq to fast-track inclusion of SpaceX into these passive vehicles. This is the really big moment where we finally see what we see in crypto.
We know what a high-FDV scam is in crypto, right? But the S&P has historically been immune to the high-FDV scam because there's a committee, and they're like, “I can't list this. It's too spicy.”
But in this case, it's such a big deal and there's so much money on the line that they're going to jam SpaceX into the S&P way faster than they should. And basically, the entire retirement—
—is [__] SpaceX.
Oh, yeah. And then everyone will max-bid SpaceX. They'll own a ton of it, and they won't even know it. They won't have done any of the due diligence.
It's like an EOS ICO. Imagine if EOS got added to the S&P 500. That's the crypto equivalent. After that, everyone's going to be like, “So it's going to be like us in crypto, but with equities, right?”
After TRUMP coin went down from around $20 to around $7, they're going to be like, “Oh, man. This is all a scam. I can't believe Goldman Sachs did this to me. Why did they put a buy—” It's like, “Well, I mean, you're a passive investor.”
You trusted the market cap of the stock, and that's what you did. So, yeah, I think that's my thought process: we're in bull mode until SpaceX. SpaceX is the Trump ICO, and then after that, it's the beginning of the downfall. That's sort of my equity view.
For crypto, I'm like, it's obviously kind of the opposite because after this kicks off, there are going to be a lot of people trying to get the hell out of Dodge. There are going to be a lot of people moving to Latin America. There are going to be a lot of people in Argentina and Uruguay, and you're going to start seeing very rich people, very soon, starting to look for exits.
Jesus, man. How could Goldman Sachs do this to me? Yeah, everyone's going to hold fucking SpaceX in the S&P?
Yes.
Damn. I was telling the chat before you came on, by the way, that you came on Twitter 24, and I was reading them my one-shotted tweets that I had 30 minutes after you got off. I was debating whether I should tell my waiter about Truth Terminal and fucking—everyone was fucking clowning me, and now they're like, "Yeah, yeah, yeah."
David, can you explain to me why my Claude feels like the models are getting worse, by the way? Do you have a take on this? You mentioned it briefly at the beginning. Why is my Claude so horrific now? What's happening to these models? It's seemingly getting worse.
I mean, they even say it to you: they're rate-limiting you. They rate-limit you, and then they add in all these excuses for why it's getting worse. It's kind of like a Big Mac over the years. They're slowly adding in all these excuses when they're just trying to survive the fact that there's too much demand and not enough compute.
It's entirely—if you know, there's a constrained resource, and this is access to how many tokens they can generate and how much compute they have access to. They saw the users growing like this, but their compute is only growing like this. You see this with the pricing of GPUs as well. Installations are getting slower. There was the Helium thing with Lizard to Hermes [?]. You have a bunch of supply shocks.
What happens is that installations are stale, they're slow, and demand is higher. So the equilibrium is fucked. You just don't have enough compute. We're in a compute-constrained environment. The scarcity is real. If we had infinite compute, everyone would have a very good model, but they're growing too fast.
Technically, as much hate as OpenAI got, investing in compute was the right move. They knew very well that it was going to be the constraint in the future.
It feels like there's an OpenAI vibe shift happening, aggressively.
Yeah. Even for a lot of Chinese models, there are a lot of GPUs being built in Asia, too, so they're getting a little better. But we are in a compute-constrained environment.
Speaking of sovereignty, Alex is saying that eventually people will just be net importers and net exporters of tokens—tokens meaning obviously related to AI and not crypto.
Yeah.
They're literally developing infrastructure to import or export how many tokens they can generate for each jurisdiction. It's become a sovereignty thing.
I guess, more toward what the purpose of AI is in the future: obviously, we have the petrodollar. This was extremely evident in what I think is a manufactured crisis that we made in the Strait of Hormuz. We just take out our military and see what happens. Of course, they'll go to the U.S.
But just to reinforce it, the petro is still the world's top concern, whether it's [peripheral Asia?], because they don't have access to oil. In the future, having proper AI for your citizens is going to determine the efficiency and national well-being of your nation. If you don't have AI versus a country that does have AI, you're going to fall so far behind. You're like a backwater country if you don't have it.
Can you explain the petrodollar, but the token-dollar thesis?
Yeah, I guess it's more the AI dollar, the compute dollar.
Yeah.
This is what happened in the 1970s. Obviously, we moved off the gold standard. They wanted to ease credit. I'll try to explain this as high-level as possible, because once you ease credit, you can start enabling ownership of different things, whether it's houses, credit cards, and other things, because you can have some elastic control over how credit expands in your country and your currency.
But the most important thing that makes a reserve dollar globally is whether you control the most important resource that determines your ability to grow as a nation. That was oil. This is why it was kind of a technological feature of a nation in the '70s: you had to control oil.
Yeah.
Just compare a citizen in China who didn't have oil versus a guy who has a car, heating, factories, and everything like that. Oil was literally the determinant of what made a nation survive at that current moment in time, and what made a citizen fall behind for the next 20 or 30 years. It was a commodity that was effectively the mortality of a nation.
The same thing is true today. If you have a citizen in a different country that doesn't have AI versus you, who do have AI, and all things work, your capacity is 10x. It's the same thing as somebody with oil in the '70s.
The entire purpose of the petrodollar is that you had to control oil in order to have sovereignty. If you didn't have oil, you were not a country. If you don't have AI, you're not a country. You are significantly behind. Your citizens are fucked. If you don't have AI, the outputs, GDP growth, and everything else—the well-being of the nation—is dependent on compute, how many tokens you're able to generate per unit of compute, and the infrastructure you built to enable that.
If you don't have enough, you just fall behind. Your sovereignty, as a result, is subsidizing the AI wave. Every dollar that we're spending in the U.S.—yes, maybe it's a little petro, but at the same time, it's also making sure that we win this AI war, this AI arms race in a sense. That's why we're spending so much money building out these massive data centers.
It really is kind of like Project Chimera. It's an arms race. It's the petrodollar all over again. It's a national resource, and we have to make sure that it's continuously being built, both at a massive scale and at a smaller scale.
It was a good explanation. Thank you.
I guess, as a follow-up, one of the last things I'll ask you, Alex—I'll let you guys go soon and go back to David afterward—you posted an article today about how these Chinese labs are white-pilled. Can you give us a high-level view of your take and your thesis? I was going to read it on stream, but instead of making you guys wait, I'll just bring you on.
Yeah. Essentially, before the last 2 or 3 weeks, using these models on your local machine or on a cheap GPU—and when I say cheap, I don't mean an H200; I mean a $10,000 GPU—wouldn't generate something acceptable for what people are using most of these models for, which is AI coding.
Specifically, everyone is using Claude Code and everyone is using Codex, and Elon has acquired Cursor for $60 billion. So we know the product-market fit in the market. The reason why the software index is getting smoked is because everyone is writing code with AI.
Prior to 2 weeks ago, doing that on commodity consumer hardware was not really viable.
The big shift, of course, is that this has become viable because of these efficiently distilled—or whether they're distilled or not—versions of Opus. These Chinese models are a new thing.
It's relevant to the crypto space because if there were ever going to be widely distributed AI spend through crypto rails, it would almost certainly be open-source models. The likelihood of OpenAI or Anthropic agreeing, given some of their comments around national security, to deliver spend through their platforms just doesn't make sense.
A lot of the Tether arguments, or even these stablecoin arguments, are like, "Oh, we'll have these bio-authenticated, KYC'd users." It's like, "Okay, you're going to have zero fee capture with that."
The chunky spend—the stuff that's leading big financing deals with USD.AI—if there were a world where we only had Anthropic and OpenAI, and they were the only buyers of compute, they would be doing debt-financing deals with Goldman Sachs, and you would never hear of chips. You would never be talking about Bittensor or any sort of distributed compute protocol whatsoever.
And so, basically, the market is totally off-sides because no one thought this was possible. The result is going to be explosive growth in Hermes Agent and all these open-source harnesses. I'm so used to using Codex that I was testing this on Codex, which is—
Long Alibaba right now?
Yeah. Yeah, I mean, I think I'm also long all of the Chinese names. Baba is the worst-performing. Alibaba came out with Qwen, and it's just—the stock is a piece of shit. So I was like, okay, this—
The worst stock ever?
It's generally part of—I mean, if you look at Qwen in the Chinese internet index, it's down-only tech, because of a lot of governance issues with the CCP. It's all the same shit as usual.
Yeah.
But I think the thing is, if a company is doing models like this, and they show that level of expertise, the likelihood that they have something closed-source that they can deal to enterprise clients is really high. You're already seeing random devs popping up on Twitter being like, “Oh, I quantized Qwen, and now I can run it on an even smaller thing at 100 tokens per second.” The likelihood that Alibaba hasn't figured that out internally, if they haven't developed a model, is very low.
Suddenly, people used to buy Meta because they had Llama, right? So I'm like, okay, this thing is the most consequential AI release since Llama. Alibaba stock is hated. Everyone is balls-long OpenAI and closed-source tech because that was the current meta. I was long SoftBank—I was shilling it to everybody who would listen. Now I've had the catalyst. We've had the big blow-off top, and I actually think the next big narrative is going to be—obviously, GLM 5.1. I'm not sure if you're familiar with it.
Okay.
Knowledge Atlas technology company, ticker 2513. A lot of these are Hong Kong stocks, which are weird. But—
You know, it's just an amazing model. I think it's a better chat model than Opus. With our own cryptocurrency, we have a community of people interacting with these models, so we run our own benchmarks with real user data. It's like, “Okay, which of these models are good, and which ones are generating—”
What you might have noticed on Opus is that it's just so fucking verbose. It has a tendency to say—like, it'll send you four pages. I'm like, “Bro, just please tell me. Just tell me the answer.”
It's like one of those women that sends you a 20-page document, and you're like—
“What is this?”
You know, it's funny—the Claude thing worked backwards. At first it was very much, “Wow, it's so personalized, and it's very nuanced, and it's hitting all these details.” But then the personality hits this pendulum too far out, and all of a sudden it's a very fine line. Then all of a sudden it's like, “What the fuck am I supposed to do with this?” Every answer is, “What is this?”
They also hired a woman from OpenAI who, originally, people got too upset about what you call sycophancy, which is them kind of glazing you.
Yes.
OpenAI overcorrected, and they made a model that's like, “Well, you might not be thinking of these seven nitpicky points.” And you're like, “Dude, okay, I'm just talking to you. I'm trying to get answers. Don't give me these fucking nuances.” They hired that woman at Anthropic.
Okay, this is my biggest pushback with Claude, by the way. I don't know how important the instructions are. I should maybe update mine, but it used to be way too glazy, right? Claude is not so much, but it's very frustrating. Now every time I ask it anything, it's like, “Sure, but what you're obviously not considering—” and I'm like, bro.
I asked it about USDai, and it gives me an answer. I'm like, “What do you think of this question?” It gives me an answer, and it's like, “But what you're obviously not considering is that it's a crypto—” I'm like, bro, just answer. Is it a good question or is it not? I don't need your life story here.
It actually hates crypto.
I'm like, what? Just answer. Is it a good question or is it not? I don't need all your life story here.
Yeah, I mean, that's the—have you—you're a psychologist, you know, based on our earlier conversation. Have you seen Boris Cherny?
Yeah, I have, actually. What are you saying exactly?
He's the guy. If you look at him and understand that the system prompts are written by this man, you should understand exactly what's happening. At an organizational level, they're getting a bunch of dudes who are making this product. It's weird. It's SBF-backed Anthropic because they're a bunch of fucking weirdos who got along with him and his effective altruist friends.
SBF is victory-lapping that all the time, but it's also like, dude, it's because they're weird people. These guys are weird. They're in polycules. They're just a group of—they're odd.
They're in the Bahamas, ripping the freak-offs and stuff.
Yeah, I mean, it's sort of like, if those are the people generating your chat models, it's going to get weird. Actually, I think Opus 4.7, as a base model, what you'll see about their corrections when they're putting out releases—I think they leaked Claude Code.
You look at Claude Code and the stuff that got leaked, and it's like there are 20 script files about generating different animal shapes for different users. You're just like, what is this tool? What were you doing when you made this, and why?
You can tell it's all vibe-coded, too. They keep shipping all these products that are clearly vibe-coded, and they clearly think they're sick, but then there are all these guys who post viral threads saying, “Yeah, the UX doesn't work in 15 different places.” I think what's actually happening is that they're a world-class model company trying to be a consumer company, and it's turning out really strange.
Strange, right? Strange is a very good word for it. I think we over-indexed on the Sam hate, really. He's coming out of this looking kind of based. Elon can't build a model that works. Claude's gotten fucking weird. The only thing was that OpenAI was too fucking nice—GPT was too fucking nice. And all of a sudden—
Well, everyone crashed out because he worked with Trump. I think there's a nuance that people maybe miss: you kind of don't want an effective altruist designing your work product. You kind of want the guy who says yes to the military building your coding tool, because he'll actually do what you ask it to do.
If Hegseth is like, “Blow up that building,” Anthropic will be like, “I don't know. Are you sure? That building is very ancient and has a lot—”
This was always the thing with the war, by the way. If you looked at it from the Anthropic angle, the Department of War thing made sense. All right, fine, Dario says, “I don't want my AI models to be used in automated killing weapons.” But then at a certain point it's like, all right, are we going to call Dario to get approval?
If you're the guy in the jet over Iran, are you calling Dario? It's a very fine-line, weird spot on this war stuff.
Yeah, I think Sam Altman is a very opportunistic guy. I don't frankly trust either Sam Altman or Dario. I think that's another really good thing for crypto AI: you shouldn't trust the CCP, either.
Your three options with the closed-source versions are: you can trust Sam Altman, who's Sam Altman. People who knew or are close to him have New York Times articles saying he's a pathological liar. He's a killer. Then you've got Amodei, who's an SBF-type character. And then you've got fucking Xi Jinping.
I don't want any of my data going to all three of these guys, because I just don't trust them at all.
Give me Eric Voorhees.
Well, the crypto advantage is that Eric Voorhees can serve Chinese models without you sending data to the CCP. That's the whole point. It's not that I trust Eric Voorhees. It's that people like Eric Voorhees can make crypto protocols where you can actually say, “Okay, did my data go to the MiniMax server that has an agreement with the Chinese military?” No, it didn't.
It’s like maybe I don’t love the Eric Voorhees. I tried looking into Venice, and all their terms of service and how they’re doing onion routing—it’s not very transparent to me. I don’t have a strong view about that, but on the surface, crypto is likely going to be the rails for Chinese open-source models.
Yeah, that was honestly a sick discussion. I don’t know if you froze, but I was going to ask David a question anyway. David, as a sign-off, now that you guys are live, the token is live, and you have a bunch of contracts, what is your number-one focus going forward? What are you most focused on?
I’m just doing more loans. That’s pretty much it. We’re very much a product-driven company, which might sound odd. We really are trying to build something that TradFi hasn’t been able to solve. I think that really comes down to how you deliver a service to the market that’s unique—something that traditional finance or other offerings haven’t been able to capture.
It’s kind of like Hyperliquid, right? Hyperliquid really is an instrument or product that’s hard to replicate on traditional rails. That’s why it becomes a very unique offering. People come from outside the space and specifically use that product.
For us, we have about 70 customers coming in right now—large neoclouds—and that number is increasing every day. We’re just making sure that we can actually service them by having the capital. That’s why we launched the token, but we also want to start building out more liquidity for sUSDe and USDe. That’s also why we’re launching a salon event.
Sweet.
Once you start serving these customers, then, to Alex’s point, people will just use crypto protocols out of need. Much like our customers, they’ve never used wallets before. Our 50 to 70 customers literally don’t even know what a wallet is.
We introduced them to Coinbase Prime, but it doesn’t work in Europe, so for one of them we had to send free Ledgers. They were like, “What the fuck is it?” It’s his fault. Then once they do it, they’re addicted. They’re like, “You can literally just borrow against your GPUs through a smart contract, get money, buy more GPUs, and get more money.”
It just blows your mind that you can do this—just like leveraged looping in DeFi, but using GPUs as your lever loop, because it actually generates money. I don’t think people realize how much money GPUs make.
This is actually how we found this sector. I was mining Bittensor with GPUs, and I realized I was paying $3 an hour for this chip. You do the basic math: there are 700 hours a year or something somewhere around there, and you’re spending $3 an hour. It ends up being more than $25,000 a year, but the chip itself costs $25,000.
I thought, “I’m paying this much a year to mine this, and this makes no fucking sense. I should just buy the GPU.” So I bought the GPU. I actually just bought the GPU. You’re paying 100% of the value of the chip. Imagine that for real estate: if you bought a house for $1 million, but your rent was $100,000 a month, you should just buy the house.
Yeah, but you can’t unless you have debt. It’s too expensive.
Got it. That’s kind of our aha moment. We thought, “Yeah, we’ve got to just buy the chip, and then get debt against it to lever it up and actually scale it.” We’re just getting more AI participants to borrow against us. Most of our fees come from AI companies; they don’t come from crypto customers.
It’s also why Alex and I are on the same pod: to illustrate that there is a use case out there. There aren’t a lot of good use cases, but Venice is one of them. Obviously, what Alex and I are building is part of a new wave of going past the jaded world of meme coins and unproductive things and actually starting to build real stuff.
I do think it comes from where the growth sectors are, which is probably not RWAs or whatever they call them. You’re just taking dinosaur assets and bringing them on-chain. That’s fucking sick. You should really be pushing this research at the frontier, which really is just AI today. But yeah.
Sick. Well, look, thank you so much for coming on. Alex was awesome. You were incredible. Congratulations on the launch and on a successful launch, at least up to this point. I’m looking forward to what you guys do.
It’s about time we have some fucking talent, some sauce, and some excitement in the crypto hierarchy, if you will. Maybe until part 3. Mr. David, it’s an absolute pleasure, man. Thanks for coming on. Thanks for your time.
Thanks for having me on. Good stuff.
All right, brother. Have a good one. Peace.
I thought a lot of that was Jenny. Did you? That was kind of Jenny, bro. Alex is a lunatic, and the USDai thing is really cool. Alex is Jenny, bro. He just has—I mean, it’s hard to refute. Honestly, tomorrow we should sit down and do some math and really figure out what we’re going to do into this SpaceX IPO. I don’t feel like we’re taking it seriously enough. Genius is generational. I don’t feel like we’re giving it enough credit. I’m not going to lie.
I like David a lot. Smart guy. Really smart guy. And Good Alex is one of the best. If you want, we could do a solo with GoodAlex in the future. By the way, I don’t hold any CHIP. I did hold it, traded it, made some money, and got out. I’m not the most confident in crypto alts, but I do like it.
If you want, I will do a solo with Alex, and we could talk more in depth about Luton Rob and a lot of this stuff, because I think he has some really good takes—good takes on privacy. Alex by himself is too intense. Maybe we’ll throw D-Nap in the corner or something. He is very intense.
Do you understand why I got one shot now? I would imagine it’s now making sense to you why that happened. Okay? I got fucking—I lost it. I lost my mind because he was talking about how AI is going to ruin—we’re going to—yeah. You get it.