GoodAlexander - The AI Trade Is Topping, Google’s Search Problem and the Long Coins Trade
- GoodAlexander says the AI trade is topping as inference growth slows and the ROI story weakens. Anthropic’s inference growth fell from 140% to 45% quarter on quarter; investors expect Fable 5.5 to reaccelerate Q4. He estimates adjusted inference gross margins at 50–60%, but a DeepSeek pro model matching Opus could push them to 10–15%: “That’s a disaster” at these valuations.
- Google’s search problem is losing monopoly pricing while AI investment consumes cash. Thread Guy cites a 13.5% selloff and search falling below 50% of revenue. GoodAlexander sees competing AI interfaces weakening monetization despite strong consumer spending; his challenge to YouTube’s growth story is pointed: “Why are you growing slower than fucking Cheesecake Factory?”
- CoreWeave’s 15% parent debt does not represent the financing cost of its entire GPU business. GoodAlexander bought debt because asset- and tenant-backed financing can clear around 6–9.5%, making the blended hurdle more manageable. The vulnerability is rental prices: if everyone expands into slowing inference demand, “we’re just gonna saturate the market with tokens.”
- Failed AI investment worries him more than a wave of AI layoffs. Accenture still has around 800,000 employees, and AI deployment itself requires hiring; his recession risk runs through disappointing returns, data-center retrenchment and weaker consumption. That caution extends to semiconductors: “The downside of tech up here is very, very high” if investors start valuing Nvidia as cyclical.
- Hyperliquid’s rerating rests on traditional finance moving on chain. TradeXYZ represents roughly 50% of its volume, varying daily, while crypto revenue growth itself remains unexceptional. The appeal is “early innings of blockchains becoming a TradFi settlement layer”; he prefers fee-generating HYPE and BNB to ETH and SOL, although HYPE’s rally has reduced its appeal.
- His long-coins thesis combines a liquidity rotation with fiscal and privacy demand. “Coins have bottomed out,” he says, while AI is losing its pull on capital. BTC benefits in his framing from roughly 6.5% annual US deficit spending and potential policy rescues; Monero benefits from triple-digit DeFi hack growth and easier access through Hyperliquid.
- Flat stablecoin supply limits his enthusiasm for ETH and SOL, while a blue wave threatens regulatory progress. Circle’s velocity explanation does not satisfy him: “We need to see the number go up.” Under a blue wave, he expects Elizabeth Warren to become banking chair and scrutinize Trump-family crypto ties—bad for integration, but potentially supportive of capital-flight demand for BTC and privacy coins.
1. Hyperliquid is being valued as settlement infrastructure
GoodAlexander reads Stan Druckenmiller’s TradeXYZ endorsement as support for a structural rerating: roughly 50% of Hyperliquid volume comes from TradeXYZ, varying daily, and traditional assets drive growth. Crypto revenue itself is not growing especially fast year on year. The appeal is being “in the early innings of blockchains becoming a TradFi settlement layer,” with Canton’s fixed-income story another expression.
The Robinhood comparison: overall revenue grows around 12%, while prediction markets and index zero-DTE options grow much faster. Customers favor “2% odds for 50x payouts.” But monthly users are decelerating, and the mobile brokerage revolution has happened; Hyperliquid offers investors a new infrastructure story.
2. Inference deceleration puts frontier valuations under pressure
The Anthropic S-1 mattered less to GoodAlexander than investor-day disclosures: inference growth fell from 140% to 45% quarter on quarter, with no Q4 guide. Opus 5.5 is “a tasteful and very good model”; investors expect Fable 5.5 to reaccelerate Q4. If inference decelerates, even the valuation’s retreat from $2 trillion to $1.5 trillion may be insufficient.
Vibe coding’s slowdown is an ROI problem in his telling: people thought they would make an app that made money, but didn’t. Traditional software engineers are now the people using AI tools. NPM download growth fell from triple to double digits. ServiceNow billings weakened as it dropped FDEs; Fortinet guided Q4 down and struggled to pass on memory costs, contradicting the cybersecurity-emergency pitch.
Previously excited by inference growth, he became more cautious about OpenAI’s cybersecurity pitch. If powerful models force wholesale software rewrites, expenses rise without corresponding revenue. His Zoom example: customers still pay the same after a complete rebuild, so “Zoom’s margins are gonna go off a cliff.”
He rejects the claim that inference is given away free: AWS showed an 80% GAAP gross margin on inference, and after adjusting for converts, Thread Guy estimates margins might still be 50–60%. He says Amazon’s Anthropic shares are behind subsidized compute deals. But if a new pro model matches Opus at a much lower cost, margins might fall to 10–15%—“That’s a disaster” at these valuations.
3. Google’s search weakness exposes the gap between AI and earnings
Thread Guy acknowledges bad earnings but finds Google’s selloff jarring: 13.5% on a roughly $4 trillion company, with search below 50% of revenue for the first time. GoodAlexander contrasts single-digit Q4 AdWords growth with Cheesecake Factory’s 15% comp: strong consumer demand makes advertising’s deceleration harder to explain away.
His answer: AI search monetizes worse because a monopoly has become a four- or five-player interface market, including ChatGPT with around a billion users. That reduces pricing power. Google, he says, would rather discuss AI progress than search numbers while going free cash flow negative for the first time.
The market allows Elon to accelerate SpaceX spending because he has a track record of getting things done, relationships in Texas to build data centers, and has rented capacity to Anthropic. Physical AI supplies his distinctive pitch: a dolphin made 10 times smarter still “doesn’t have opposable thumbs.” GoodAlexander says physical AI training data is difficult or perhaps impossible to distill, describing individual data sales worth $5–10 million.
China is his counterexample to AI automatically helping earnings: a property wealth shock has crushed consumption despite strong models. Mooncake sales are down around 65% year on year; Hang Seng Tech trades around 10 times EBITDA versus 23 times for the Nasdaq. “You might have great AI models,” but if the consumer rolls over, “it’s all downside.”
4. Neocloud financing works only while rental economics hold
GoodAlexander says CoreWeave’s operators were Ethereum miners and had personally sold a couple billion dollars of stock over the last year before the company accelerated borrowing and investment. A roughly $100 billion, five-year backlog does not prevent near-term cash burn, even with three-year paybacks. Investors face opposing risks: acceleration brings competing chips and depreciation; deceleration undermines inference demand. Neither camp wanted the financing.
GoodAlexander bought the debt because the 15% headline ignores its structure: roughly 70% of a deal can be financed against GPUs and tenant commitments, such as Microsoft’s, around 9.5% or less, potentially 6% for a high-quality tenant. The last 20% uses parent debt yielding 15%. Vera Rubin looks capable of clearing the ROI hurdle, assuming market prices and rental rates hold.
His concern is that assumption: he says a GPU can generate $30 an hour running DeepSeek while rental costs are $10 an hour. If everyone chases that spread while inference decelerates, “we’re just gonna saturate the market with tokens.” Today’s attractive prices need not persist.
Oracle’s 14% rebound rewards investment cuts and a cap of $10 billion each on additional stock and debt. Management still claims two-year paybacks, free cash flow breakeven in 2028 and $30 billion of free cash flow in 2029. Credit-default-swap spreads are falling; CoreWeave equity still risks dilution if it must finance deals in a weak market.
5. The employment risk runs through disappointing investment returns
Thread Guy raises Microsoft’s 60,000 buyout offers after Block’s layoff announcement lifted its stock 15%. GoodAlexander pushes back: Accenture still has around 800,000 employees, and AI transformations require forward-deployed engineers. Commodity software workers are exposed, but “the idea that there’s mass unemployment is clearly false.”
Bank of America illustrates the scale problem: he cites $300 million of Mythos savings, rising to $700 million next year, against a $2 billion money-laundering charge and 6,000 compliance hires. His bigger employment worry is failed AI CapEx, data-center retrenchment and negative GDP, turning Cheesecake Factory’s positive 15% comp into negative 15%.
For SOXL and semiconductors, he flags a January Huawei–DeepSeek training-chip launch. Vera Rubin upgrades look exciting assuming inference demand holds. But Korean equities trade at six times earnings; repricing expensive Nvidia as cyclical could be painful. Asked whether buying semiconductors here offers good risk/reward: “No, I think it’s terrible.”
6. Long coins means selecting fees, fiscal hedges and privacy
His preference is fee-generating HYPE and BNB over ETH and SOL, although HYPE is less compelling after its run; BNB still looks reasonable. “You generally just want to own the cheapest fee-generating assets you can on-chain.”
Monero’s specific demand driver is noncyclical, triple-digit year-on-year DeFi hack growth. The Wagyu bridge lets Hyperliquid users reach spot XMR without KYC, opening incremental demand. GoodAlexander keeps the operating uncertainty explicit: “We’ll see if it stays open.”
BTC benefits in his framing from roughly 6.5% annual US deficit spending and possible fiscal and monetary action if growth weakens. AI was sucking liquidity out of crypto; now, he says, “coins have bottomed out.” If crypto hacks decelerate while TradFi hacks accelerate, crypto’s forced security adaptation could become an advantage.
His counterweight is the stablecoin supply chart: Tether and Circle aren’t getting incremental volumes. Circle’s velocity explanation does not satisfy him: “We need to see the number go up.” He says materially increasing stablecoin supply would make him more excited about ETH and SOL.
7. Political backlash could split crypto’s winners
The next two years could entrench crypto in TradFi, making it harder to unplug. Under a blue wave, however, he says Elizabeth Warren would become banking chair and the focus would turn to World Liberty Financial and Trump-family ties. He worries about integration while expecting capital-flight demand for Bitcoin, Zcash and Monero.
His final political call draws on what he describes as Palmer Luckey’s genetically engineered war dogs: religious voters will turn against AI over “playing God.” With the pope already against it in his telling, he thinks that backlash probably puts a damper on the political right in the US and globally.
Full transcript
Yo, yo, yo. What’s up, man?
Doing well. How are you?
Dude, I’m fucking incredible. It’s good to see you back, man. It’s been a while.
Been a while.
You’ve got the background, the full mog. I love it.
Oh, man. I’m thinking of turning it off because people might roll up on me. Yeah.
I love it. Dude, it’s good to have you back. There’s a lot going on. I wasn’t planning to talk about cryptocurrency at all, but the Hyperliquid news—I figure we might as well just start there. You made a tweet about it. They got Big Stan, I guess, the number one KOL. I’m not sure how much bigger you could get than Big Stan coming out and making his first tweet ever, shilling TradeXYZ and Hyperliquid. What do you think?
I mean, his daughter works at Palantir, right?
Really?
Yeah. He’s hard-shilled Palantir before, and the last time he did that, he was like, “My daughter works there,” and “It’s gonna go up.” That obviously worked, right? So, if you follow his track record of overly promotional statements, it’s worked out well for investors.
I think the Hyperliquid take is that, if you look at Robinhood, for example, monthly average users are decelerating. Hyperliquid has turned into this thing where, if you look at the volumes, it’s about 50% TradeXYZ at this point. It varies by day, but the actual TradFi metrics are driving growth for Hyperliquid.
The story at Robinhood is not very good because you’ve got legacy equities decelerating. Actually, all the growth at Robinhood is prediction markets and zero-DTEs—
Which is insane, by the way. How much volume—
Exactly.
How much money they’re making on prediction markets. I was shocked.
Yeah, it’s truly astounding. A lot of those markets are—you know, there are all kinds of studies on it—low-percentage YOLOs. That’s the popular market. People love buying 2% odds for 50x payouts. People also like short-term price action prediction markets.
Yeah.
If you’re honest about the actual growth driver of Robinhood, it’s zero-DTEs on index options, which are growing almost triple digits. Overall revenue is growing about 12%, but prediction markets and these degenerate bets are growing massively, at triple digits, and index options are growing really fast. Crypto is not growing fast.
It’s similar with Hyperliquid. It’s not actually growing super fast year over year. Same thing with Pump.fun. If you look at annual year-over-year revenue, it’s not particularly compelling. October is fine; year over year, it’s not that good.
The reason the market is rerating it is that there’s a question around the sustainability of crypto revenues. In TradFi, people are saying, “Robinhood isn’t sustainable because it’s not providing core infrastructure. The mobile revolution has already happened.” With Hyperliquid, people are saying, “Okay, we’re in the early innings of blockchains becoming a TradFi settlement layer.” That’s why Canton is pumping this morning too, right? People are thinking, “Maybe we’re early in putting the fixed-income market on-chain.”
I think it’s one of the few non-cyclical stories in the market. AI was supposed to be this non-cyclical growth story, but we’re seeing signs of that decelerating. Everything moving onto a blockchain is a story people want to chase, and I think Stan Druckenmiller putting it out there is going to help with the story.
That’s a pretty good take. I agree with you on Robinhood. I was shocked when I saw how much they did. What was it, like $300-something million in options volume to about $175 million in prediction markets? I was like, “Fuck, really?” It feels like that number is tight. But I think it’s a good take.
But anyway, I’ve already talked about Hyperliquid a lot. We don’t have to do a lot of crypto stuff. What I really want to talk to you about is everything happening in stocks, AI, and Google. The big news this week was Anthropic’s S-1, and I know you’ve been talking about the prospect of Fable 5.5 and how important that’s going to be, whether it’s amazing or not. How do you feel, a couple of days after digesting Anthropic’s S-1, about the implications that’s going to have when the IPO hits the market?
Yeah. Anthropic’s S-1 didn’t contain a lot of super-material information for investors, in my opinion, because the thing people were specifically concerned about was a deceleration in inference.
Yeah.
Right? That came at the investor day. Basically, there was no Q4 guidance.
No.
Inference was growing 140% quarter over quarter, and it dropped to 45% quarter over quarter. The consensus right now, if you poll investors, is that Opus 5.5 is quite good. That’s objectively true. Opus 5.5 is a tasteful and very good model. The story is that Fable 5.5 is going to drop and cause Q4 to reaccelerate, because right now, if you look at the overall state of the inference market, Claude Code and vibe coding are starting to compare very negatively quarter over quarter. npm, where you download code packages, has basically gone from triple digits to double digits.
Why do you think?
I think people aren’t getting—I think vibe coding was hot, and now it’s not hot. Basically, people didn’t get the ROI. They thought they were going to make an app that made money, and they just didn’t. Now the people using AI tools are traditional software engineers.
I think the other thing hitting sentiment—and we can talk more about it—is that a lot of these AI-driven narratives from publicly traded companies aren’t demonstrably paying off.
Yeah, yeah.
Google just imploded. Its advertising business is decelerating. Fortinet said AI was going to cause an acceleration in cybersecurity. It didn’t. They just guided down Q4.
ServiceNow just dropped a ton of FDEs—people working full-time deploying AI at customer sites. Billing declined, right? So you have three different segments: advertising showing a deceleration; enterprise SaaS, with ServiceNow, showing a deceleration; and cybersecurity, which was supposed to be this massive acceleration, showing a deceleration. All three of them.
Somewhat notably, on the Fortinet side, you have this question about memory, right? SK Hynix came out with very, very strong earnings, and they’re saying, “Look, we have a ton of pricing power. We’re getting extensions for these multiyear pricing contracts.” Then you listen to Fortinet, and they’re saying, “Yeah, we’re having a hard time passing memory costs on to executives.” You’re like, “Wait a second. I thought cybersecurity was an emergency. Why are you having a hard time nickel-and-diming executives over memory costs?” The stock is down more than 20% today.
I think the market is seeing a consistent story, whether it’s code downloads, interest in inference, or OpenRouter traffic: “I’m putting everyone into AI, and I’m not necessarily getting a return on investment.” That calls a lot of the CapEx into question. I think the current story in the AI market is that the ROI isn’t there.
Now OpenAI has already pivoted its marketing to investors in its fundraising round toward cybersecurity. In the last couple of weeks, the companies have basically come to the realization that maybe inference is decelerating. We know that if inference decelerates, $1.5 trillion is too high for Anthropic’s valuation. It was at $2 trillion, and now we’re talking about $1.5 trillion. Then we’re like, “Okay, maybe $1.5 trillion is too high.”
There’s been a dual shift in the narrative. As we move out of inference, the first thing they tried was biotech. They were like, “We’re going to cure cancer.” At Oracle’s investor day, they spent about 10 minutes talking about how people at a hospital think AI is going to cure cancer.
The other narrative is cybersecurity. OpenAI is running around saying that—Sam Altman is tweeting things like, “I’m so scared of the new model’s capability,” because it’s literally hacking things at the hardware level. So you’re going from, “We’re going to sell you a lot of inference,” to, “Actually, AI is going to hack everything so badly that you’ll need to rewrite all of your software from scratch, and the only way you can do that is with a frontier model.” That’s going to be their demand story.
That’s so bearish, because think about what happened with Fortinet. They told everyone that was true. It just wasn’t true.
Management didn’t pay up for it, and now the stock is down a lot.
Mm.
If it is true, and everyone needs to fully rewrite everything that’s ever been written, that means everyone’s expenses are going to rip and their revenue isn’t going to accelerate, right? Nobody’s going to notice that all the software has been rewritten.
Yeah.
If you have to rewrite Zoom from the ground up, you and I are still going to pay the same—
Doesn’t give a fuck, yeah.
Then Zoom’s margins are going to go off a cliff, right? That’s where my risk-reward analysis shifted. I was getting pretty excited about this inference acceleration back when inference was comping at 50% quarter on quarter, maybe even 100% quarter on quarter. That’s what people saw in Q2. They were like, “Oh, inference margins are dope.”
There’s this FUD, this Ed Zitron thing where they’re like, “Oh, it’s actually all convertibles.” When you do the math, AWS has basically printed an 80% GAAP gross margin on inference. Then you dig into it and realize Amazon has this massive bag of Anthropic shares that’s causing it to do subsidized compute deals—
Yeah.
—with Anthropic. But even with that, even adjusted for all the converts, it’s still maybe a 50% to 60% gross margin on inference. What Zitron said—that they’re literally giving it away for free—isn’t true.
No.
It’s 50% to 60%. The problem is, if you look at the cost curve of the tokens and the capability of something like DeepSeek, it’s so cheap relative to Opus. If a new Pro model launches at a similar capability level to the current Opus, then you might have a situation where margins go to 10% to 15% gross margins. That’s a disaster. You can’t have a business trading at this valuation.
The overall AI trade is showing fatigue. The reason is just that everyone’s at peak investment. But I’ll stop ranting there and let you ask follow-ups.
Can I follow up with Google? I think you understand the search advertising business really well. That’s the story right now: it’s down 13.5%. On a $4 trillion stock, that’s down about $400 billion. Unfathomable amounts of market cap erased. I know they had bad earnings, but it still felt like there was some whiplash here. I think the metric was that search is under 50% of their revenue for the first time ever, and it’s just down only. What’s your take on what exactly is happening with Google?
Yeah. If you look at the consumer economy, Norwegian Cruise Line says there are 20% more people on cruise ships year on year because people are cruising in the middle of a war. Cheesecake Factory put up 15%. People are buying deep-fried Buffalo Wild Wings at Cheesecake Factory: 15% comps. Chili’s is at 10% comps. The idea that Google AdWords is decelerating to a single-digit comp in Q4 while all these consumer stocks are ripping—the idea that consumer spending is slowing—GDP just came out, and consumer spending was very strong. The idea of advertising decelerating in that market is strange. And by the way, Netflix is down 50% year on year.
Yeah.
The story was supposed to be that people are only watching short-form video. YouTube was the “elephant in the room,” and Netflix is down 50% year on year because YouTube is taking share. If YouTube’s taking share, why are you growing slower than fucking Cheesecake Factory? What is this AI story? If AI is so good, why is your Q4 decelerating?
The answer is that monetization on artificial intelligence-based search is lower than traditional search. The reason for that is obvious, in my opinion. Google used to be a monopoly with basically complete market share in search. Now you have 15 different agent platforms, whether it’s Muse, Grok, Google’s own agentic search platform, or ChatGPT. ChatGPT is a billion-user product with search enabled in it, and everyone’s doing it. You’re going from a one-player market to a four- or five-player market in terms of the user interface for search, and that’s causing a decelerating ability to charge, which is very bearish.
Well, Google, if you listen to their earnings call, they don’t even want to talk about search.
They’re only talking about Argon and RSI and their new TPUs. This is every earnings call, whether it’s ServiceNow, Fortinet, or Google. All these guys that are down double digits are always saying, “We only want to talk about AI and AI progress, and we don’t want to talk about our actual numbers, which all look like shit.”
The story is, “We’re investing so much, and maybe you’ll see it show up—not next quarter, definitely not next quarter, because we just guided down—but maybe at some point, if we hit RSI, this is going to be awesome.” I just think the market has already shown us with Oracle and CoreWeave. The market already forced Oracle to cut its investment.
Oracle just had an investor day, and the market pumped the stock 14%. The reason it pumped so much was because they promised to stop investing and diluting so much. Before, the story with CoreWeave and Oracle was: “Look, if you buy a house, what’s the rental yield? Do you get paid back in 10 years, 15 years? With these data centers and these chips, do you get paid back in two to three years? I’ll do that trade all day. I’m getting paid back in two to three years. I’ll just do unlimited amounts of that.”
The market was like, “I don’t know about that. 2029 is a long time from now. Your debt is due next year.” Oracle finally slammed on the brakes. They said, “You know what? The most we’re issuing is another $10 billion of stock and another $10 billion of debt.” Google is going free cash flow negative for the first time in its history.
That’s crazy.
Google is still keeping its foot on the accelerator. SpaceX has its foot jammed on the accelerator. They’re adding so much capacity to the market. The interesting thing about the market right now is that it’s allowing Elon to do this completely insane CapEx.
Why do you think?
He’s got a good track record of actually getting it done. He has the relationships in Texas to get permits and build the data centers. He was early to build data centers and rent them out to Anthropic. The fact that Elon is renting this stuff out to everyone is sort of like—Sam Altman and Dario might have beaten him on the model side, but at the end of the day, they’re paying rent to him.
Got it. Yeah, yeah, yeah. Yeah, yeah, yeah.
He’s sort of the Chad in the situation. If you look at his Q3, his Q3 CapEx at SpaceX is higher than his entire calendar-year CapEx. He’s gone from a completely cash-flow-reasonable situation at SpaceX to completely blowing it out on CapEx. And he’s talking about acquiring Tesla.
He’s gone from a disciplined investment process to a completely crazy investment cycle. That’s actually what Oracle did last year, and they got crushed. CoreWeave now, when they’re on the road, are saying, “By the way, if we turned off all of our capital expenditure, we’d become profitable.” That’s the story now, which is a very late-cycle thing, and Elon is going full blast.
I thought it was interesting what Elon said. It may be an interesting story, and this is something that’s happening in Chinese AI as well. Elon said, “Look, if you made a dolphin 10 times smarter, it still doesn’t have opposable thumbs. So it isn’t able to manipulate its environment.”
Yeah.
“Even if it’s able to recite Shakespeare, it doesn’t really matter. It doesn’t have thumbs.” His story is physical AI. He’s like, “We’ve got Teslas.”
Ah. Yeah, yeah.
In China, it’s the same story, actually. The Chinese government is incentivizing physical AI, and there are some technical nuances with physical AI that make it more or less impossible to distill physical AI training data.
So in the private markets, there’s this big bubble in basically maids with cameras strapped on them—
This is nuts, by the way. I’m seeing this everywhere.
Yes. The story is: Anthropic got distilled, DeepSeek can just steal Claude or whatever, but you literally can’t steal this robot training data. So you’re seeing this data explosion where people are making $5 million or $10 million just selling a piece of data to a lab.
It’s insane.
That’s where the market is. They’re rewarding physical AI; they’re crushing software AI. The consumer is extremely strong. It’s almost like the guys who just weren’t monitoring the situation and didn’t know about AI are doing great. They’re on cruise lines. They’re in a hotel. They’re at the Cheesecake Factory. They’re like, “What’s AI? Why are you so concerned, Anon? I’m just at Cheesecake Factory. It’s fine. Why are you buying gold at Costco? That shit’s going down, right?”
I think that’s where we are in the markets. I have a view, obviously, that if you look at the U.S. consumer, it’s a total aberration. If you look at the Chinese consumer, they’re in a depression. Louis Vuitton is down 45% to 50%. The Chinese real estate bubble basically crashed, and you had this huge wealth effect loss in China where consumers just aren’t spending. Mooncake sales were down around 65%. Mooncakes are like the equivalent of when shitty corporations send you caramel popcorn.
Yeah, that’s like their—
Mooncakes are kind of like that for Chinese people. They have like 2,000 calories. You should never eat a mooncake. But they’re down like 60% to 70% year on year, which just shows you how bad it is in China. You have the Hang Seng Tech Index trading at around 10 times EBITDA, and the Nasdaq trades at around 23 times.
Whoa.
In China, you can see what the market looks like without a good consumer, and it is no bueno. They have really good AI, too. I think that’s the interesting thing about China: they have such a powerful AI sector, and it just doesn’t matter. It doesn’t actually drive consumption or productivity. I think that’s the bear case for the market. You might have great AI models, and AI might work really well, but it’s totally unclear if that’s good for earnings. If the consumer rolls over, it’s all downside.
Mooncakes. That was crazy. Okay, can you tell me this? I feel like I understand, but I don’t know the intricate details of the neocloud business that well. I was following CoreWeave, and that was going crazy. Everybody was panicking, and CoreWeave and Nebius got destroyed. Then I’m following the Elon situation. Can you high-level explain the panic and what the risks are? I know you posted an article about CoreWeave, which is a really good one. How is this going to play out, and what’s the historical precedent for this in the early 2000s?
I mean, I think maybe as a crypto guy, you might understand this better. The CoreWeave guys were Ethereum miners.
It’s crazy.
And they took off a couple billion personally in the last year selling the stock. Going into it, you have to keep that in mind. With companies like Cerebras or CoreWeave, you have these getting-older guys selling a couple billion, and you’re like, “If you’re so excited about this, why are you selling billions of dollars of stock?” So going into it, there’s just a high bar for this to be a thing.
Yeah.
Right?
Yeah.
Of course, what happened was that DeepSeek launched and put numbers around it. Essentially, even if you’re running DeepSeek, you can make $30 an hour renting out a GPU.
Okay.
And the rental rate is $10 an hour. By the way, old GPUs, like H100s, are still renting out at high rates.
Crazy.
If you’re in CoreWeave management, the bear case for your business is depreciation: you’re caught holding the bag, owning and renting out these GPUs. But if we look at our historical GPUs, I’m not seeing any signs of it.
So what CoreWeave did when they saw this was say, “We’re going big. We’re going to take out a ton of equity, we’re going to take out a ton of debt, and we’re going to radically expand our offerings.” If you look at the chart, they got a backlog of about $100 billion over 5 years. They grew their backlog like crazy and started investing like crazy. Their earnings per share went off a cliff because with these businesses, even if they have, let’s say, a 3-year payback—
Uh-huh.
—your EPS goes off a cliff in the short term, you’re burning tons of cash, and you need to be able to finance it via debt and equity sales.
Investors were like, “Look, we’re seeing all this. In the bear case, OpenAI is talking about their new chip. CXMT is talking about discontinuous memory development. We didn’t think it was possible for them to do HBM3 memory without advanced lithography. Now they are. Everyone is launching a new chip at the same time. Google is launching TPUs. The idea that historical chip appreciation is going to last into the future, after everybody on Earth is going into chip development, isn’t believable to me.”
Yeah.
That’s sort of what investors were thinking. They were like, “Dude, I’m not going to do this. You’re issuing all this debt and equity. It’s too dangerous. The volatility of outcomes here is too high. Z.ai, the open-source lab company in China, is down 80% off its unlock.”
Yeah.
From its high, yeah.
Wow.
MiniMax, too. MiniMax is a great video model, and its stock is getting crushed. So investors are kind of like, “Yeah, I don’t know about this.” Oracle’s already—
I think investors, even on the sort of accelerationist path, have two paths. If you’re super accelerationist, kind of like these RSI guys, chip depreciation is going to rip in an accelerationist world where everyone makes their own chips. If you’re a decelerationist, then inference is going to go off a cliff, this is a fad, and then you’re going to—so you have these two camps. Neither of them wanted CoreWeave equity or Oracle equity or debt.
I see. I see.
Then you had this ramp in financing with the optics of manager share sales, and then you had a panic.
It’s just all—
Basically, what happened is they clicked on the accelerator right after they personally sold a lot of stock into a market that was skeptical of—
Yeah.
—the ROI, and people are just like, “I don’t want this. I don’t want this debt.”
The other thing, I think, is the reason I got long the debt was that the meme was that CoreWeave is financing at 15%, so their ROI needs to be higher than 15% or it’s going to zero.
Mm-hmm.
That’s not true. The way these debt deals work is they have asset-backed loans. It’s called DDLT 5.5, and they basically have chips backing the debt, and they also have the tenant backing the debt. Investors will go into a CoreWeave deal, and let’s say 70% of it is financed—
Oh.
—with this Microsoft tenant agreement to pay, plus the underlying assets, which have been appreciating, and that clears at around 9.5%, or even less. For a high-quality tenant, it might be 6%. So the blended cost of CoreWeave debt for these deals: they only need to finance the last 20% with parent company debt, and that’s the crap that’s yielding 15%.
Ah.
The question is, can they hit that ROI hurdle? With Vera Rubin, which is Nvidia’s new line, it looks very much like they’re going to be able to hit that ROI hurdle, assuming market prices stay the same. Assuming the future spot rate for rental stays where it is, they’re going to be printing money even at current interest rates.
The problem with this in the bear case, and the reason I started getting freaked out, is I’m like, “Wait a second. If everybody in the world is making $38 an hour selling DeepSeek into a decelerating inference market, I don’t think that $38 is going to hold. Everyone—we’re just going to saturate the market with tokens.”
Just destroy it.
And then all of a sudden, if the break-evens go negative, it just gets really complex, right? I think the market freaked out because of equity and debt issuance that the market didn’t have appetite for—
Yeah.
—combined with the optics of the insider sales.
Got it.
And then the market has transitioned into Oracle actually calming people down. They said, “By the way, our paybacks are still 2 years. Our unit economics: we’re going to go free cash flow breakeven in 2028.”
Ugh.
“And we’re going to do…” Well, 2028 is a long time from now.
I know.
But the stock is just destroyed, right? They said, “We’re going free cash flow breakeven in 2028. In 2029, we’re going to make $30 billion of free cash flow. So if you just chill until 2029, we’re going to make so much money.” That’s what Oracle said yesterday.
So they’re kind of like… But at the same time, they’re like, “By the way, we’re cutting our investments.” Oracle, which was this giant sort of accelerator, is finally listening to the market. It’s like, “We’re not going to issue more debt and equity. We’re going to slow things down—
Mm-hmm.
—and we’re going to wait for these ROIs to pan out.” So that’s what’s happened. We’re now consolidating. CDS has started to drop on CoreWeave and Oracle. But then the question is, do you get soaked on the equity because CoreWeave still has to finance these deals? Do they have to issue equity into a crappy market? It’s a high-complexity trade.
That’s a pretty beautiful explanation, honestly. I liked your article too. I thought it was very well written. Okay, one of the last things I want to ask you about that I think is happening is—we’re seeing these layoff scares again. It’s funny: Citrini posted “the 2028 Intelligence Crisis” article in February 2026, and in it he talks about how the euphoria was nuts. The S&P flirted with 8,000, the Nasdaq broke 30K, and the first set of layoffs started to hit.
I remember when Jack Dorsey and Block announced their layoffs. I was like, “Oh my God,” and the stock was up 15%. I thought, “Oh, this is it. Everyone that’s struggling is just going to announce layoffs. It’s a free shot. AI—everyone can do it, and it’s going to start this trickle effect.” It didn’t happen. Then the other day, Microsoft offered buyouts to 60,000 people. It’s a monster. I think it’s 25% of the company: 60,000 people. Then there’s Dropbox, and all these floating discussions about maybe now is the time. I think the Microsoft one was the significant one. I’m curious what you think is going to happen on that side of things, paired with your views on SOXL and memory from here.
On layoffs, the indicator I always look at is Accenture’s total headcount.
Yeah.
Because they report it. It’s a KPI. They still have about 800,000 people.
Facts, yeah.
And the stock is doing well.
Yeah.
Microsoft, I think, is a little different. If you think about AI, the industries most affected are people who are writing commodity software. But that’s not a huge percentage of employment. If you have stuff like Cheesecake Factory growing at 15% and cruise lines growing at 20%, obviously people have jobs, right? The idea that there’s mass unemployment is clearly false.
Yes.
The unemployment you need to worry about is closer to China, where the economy goes off a cliff. I’m not worried about AI destroying everyone’s jobs.
Yeah.
The unemployment situation I’m worried about is that there’s not an ROI on AI CapEx.
Yeah.
The data center pulls back, GDP goes negative, Trump attacks Iran, the economy decelerates, and you go from 15% Cheesecake Factory comps to negative 15%. There’s a real economic deceleration. I don’t think Citrini’s argument that Accenture and software are just going to fire everyone is really true.
Yeah.
If anything, as you’re seeing with ServiceNow, part of the reason they’re whiffing is that everyone has to start hiring FDEs. They have to hire—
What does that mean?
—forward-deployed engineers, which is a—
Oh, okay, okay, okay.
—fancier term for, “Okay, we need to transition you from your old software mode to AI mode.”
Got it.
“And now we need to hire all these people to help you with that transition.” That’s why Accenture is going up, actually, right? Accenture is doing these so-called “AI transformations.” That’s actually headcount expansion. An interesting example is Bank of America has a Mythos deployment.
Oh.
They said it’s about $300 million of savings, and then they’re ramping it to $700 million of savings next year.
Oh.
But they had about a $2 billion charge for money laundering, and they have to hire about 6,000 people to go through that and do all the compliance work. That gives you a picture of a big bank. It’s a microcosm. AI is starting to make some headwinds, but—
Yeah.
—relative to the other crap that’s going on in the economy, it’s actually not that big.
Yeah.
Which is incredibly problematic, right? If you’re wondering what the value is, Anthropic would be super bullish if there were mass unemployment. We’d all be buying these labs because they’d be the only equity. It’s the only way you can escape the permanent underclass—
Yeah, just buy Anthropic.
No, man. It seems like things might just kind of be hanging out. Spend is accelerating, and people aren’t seeing an ROI, I think. So I think Citrini is kind of wrong.
Yeah.
DoorDash isn’t going away. I’m not on Claude Code ordering DoorDash.
No, you’re not.
Muse is exciting. Agentic shopping is here to stay, but there’s a world where that just decreases Google’s margins. The winner might just be the consumer economy. I think it’s less exciting. The geopolitical stuff is more worrying.
As for SOXL, I think an underrated catalyst is that in January, Huawei and DeepSeek are launching a training chip, and not a lot of people are talking about it.
Mm.
Obviously, in China, tech stocks are just so destroyed that I think people are depressed. They’re like, “Ah, we’ll see what happens.”
We’ll see. Yeah, fuck it.
Huawei is private, so you can’t even buy it, and—
Yeah.
—DeepSeek is private. The only 2 stocks you’d want to own in China, you can’t even buy. It’s completely over there.
Yeah.
But yeah, that’s coming out in January. If it works, I think memory… What you’re seeing is that Vera Rubin upgrades are going to be very, very exciting, assuming inference demand holds.
Oh.
But Nvidia is pretty expensive. You see the Korean equity index trading at 6 times earnings, and it has crazy growth—higher growth than in the U.S., because people think it’s cyclical. If you get into a market where people start treating chips or Nvidia as cyclical, then, oh my God, it could drop a lot. The downside of tech up here is very, very high, whereas stuff like China is already bombed out.
Do I think the risk-reward of buying semiconductors here is good, given what we’re seeing with AI investment ROI? No, I think it’s terrible.
I think it’s terrible. Okay, I think the perfect way to conclude is with something I didn’t really ask you about, but it was the center thesis of a lot of your posting over the last couple weeks. It had me pretty fired up: your long coins thesis. Have you wavered from that? I don’t think HYPE was really included. We talked about HYPE, but what about the rest: BTC, the privacy stuff, Monero, Zcash, CHIP, whatever? Where do you stand on the long coins pair trade? Today is October 29, 2026.
Yeah, I said I don’t want to own ETH and SOL.
Yeah.
I said I prefer HYPE and BNB because basically they’re generating fees. Up here, HYPE is obviously less compelling because it’s already run. I think BNB is still reasonable. Generally, you just want to own the cheapest fee-generating assets you can on-chain.
I think the core Monero thesis is: what do I own that’s actually going to go up as a result of DeFi hacks, which are going triple digit?
Yeah.
…non-cyclical, triple-digit year-on-year growth. It clearly has a beneficiary. So it’s like, what do I want to own? And the Wagyu bridge on Hyperliquid can get people into Monero who probably weren’t going to touch Monero before, right? The fact that you can trade it on Hype, bridge in and out, and get spot XMR via Hyperliquid with no KYC is an incremental demand driver. We’ll see if it stays open, right?
Yeah, yeah.
But that was the Monero trade. With BTC, I like the fact that we’re going at 6.5% a year in deficit spending. Incremental debt in the U.S. is completely insane, and people in TradFi are starting to get worried about fiscal situations.
Look at the guys involved: Druckenmiller and PTJ are both fiscal guys who love dooming about the fiscal situation. We already tried to take Saylor out of the market. And if you’re bearish about the AI narrative, think about all the quantum FUD we had. For example, look at Korea. At one point, Korean stocks were trading 36 times the volume of Upbit. Memory was doing all of the volume in Korea.
Wow.
XRP Swell was standing-room-only recently. You’re like, “That seems stupid,” but it is what it is, right? Coins have bottomed out, and the AI trade is topping. What if there’s a world where we stop talking about Claude breaking, developing a quantum computer, and people just say, “Okay, well, Saylor is not going bankrupt. PTJ says the debt dynamic is unsustainable. Our only option is monetary”? I think the other subtle thing people may not be thinking about is that crypto has already adapted to this whole cyber-threat environment.
I love this take. I love this take.
We literally just had to do it, and everything else hasn’t, right? If we start to see decelerating hacks in crypto and accelerating hacks in TradFi, you have GDP decelerating, more monetary policy action, and fiscal expansion to bail out the economy, which drives BTC up. Then you have the worst behind us in terms of hacks. I just think we’re at an inflection point where the AI trade was sucking liquidity out of crypto.
Yep.
It’s starting to inflect negatively. Crypto itself is doing fine. Hash power is up on Bitcoin.
It is.
I’m freaked out about stablecoins. I really don’t like the stablecoin supply chart.
Like, it’s not going up?
Yeah. Tether and Circle aren’t getting incremental volumes.
Which is bizarre, I think.
Yeah, Circle’s story on earnings is like, “Hey, guys, the velocity is going up. You just need fewer stablecoins in this.”
It’s like, stop.
No, man, we need to see the number go up. It’s so weird.
We need to see the number go up.
That’s why I’d get a lot more excited about ETH or SOL if we had materially increasing stablecoin velocity. Right now, the story is that Hyperliquid, Lighter, and Binance make all the money from perp trading. BTC wins the store-of-value narrative, maybe along with some privacy coins.
Yeah.
For ETH and SOL to pump, stablecoin supply needs to go up.
Yeah.
If it doesn’t go up, then you have a Democratic administration and a story about regulatory uncertainty, and that’s just bad. I think the final thing I’d say on crypto is that, with Hyperliquid and ICE, there’s a world where in the next 2 years we get crypto entrenched in traditional financial systems, making it much harder to unplug.
Mm-hmm.
The flip side is that I’m a little bit worried about a blue wave. In that case, Elizabeth Warren is going to be the banking chair.
Oh my God.
The entire thing is going to be about World Liberty Financial and Trump family stuff.
God.
That clearly isn’t going to be good. Why did he pardon CZ?
Elizabeth Warren, give me a break.
It’s like, “Ah, he seems like a good guy.” Elizabeth Warren as the banking chair is going to look crappy. But I think people are going to start to get afraid. The Brazil thing, the left-wing populist thing, and Elizabeth Warren—it’ll make people say, “Yo, I need to get my money out of this place.” That’s why I like the Zcash, Monero, and Bitcoin store-of-value run.
Yeah.
You’re like, “Okay, yeah, the left-wing populists probably aren’t going to love this Hyperliquid integration, but they are going to be super stoked about stealing your money,” right? So, yeah.
That blue wave is flying. Brazil, New York—obviously, 2028 is looming. It’s budding. You can feel it in the air.
My last take is that I think the Republicans are… There’s this Pope thing. I think we’re really early on this narrative. Palmer Luckey has genetically engineered war dogs.
Insane.
People are genetically modifying human brains to run LLMs. There’s all kinds of dystopian sci-fi nightmares. But the real optics are about playing God, and the Pope is already against it. I think we’re going to see this shift where religious people around the world stop liking artificial intelligence.
Abortion isn’t the main thing anymore. Now it’s, “Okay, you’re literally modifying the human genome and playing God,” and that’s worse than abortion. I think that’s what people might not be thinking about in terms of the politics of the situation. Everyone’s like, “Oh, yeah, human embryo alteration, installing Neuralink in babies’ brains—that’s going to fly well with Christians.” No, probably not. I think that’s still contrarian and sort of inevitable, and it probably puts a damper on the right wing in the U.S. and globally for a bit.
Yeah. The fucking war dogs thing is nuts. Good job, Alexander. You’re the best, man.
Thank you.
It’s always fucking box office. The articles are great, by the way. You just going to keep firing them? I love them.
Yeah, yeah. Still firing. All right. Talk soon, man.
Dude, have a good one. Peace.
Bye.