MARKET UPDATE: What Breaks First — Bitcoin, AI, or the Bond Market?
- Avi’s trading framework is to add to demonstrated winners; Jonah adds that deal access itself should be scrutinized. “Losers average losers” means doubling down becomes especially dangerous when the information behind a losing trade has changed. Jonah’s venture experience illustrates how better networks brought better deals, while earlier failures had discouraged him from investing.
- Jonah treats CoreWeave’s debt sell-off as a genuine financing warning. He cites 2032 bonds at 79 cents on the dollar, CDS at 16 points upfront, and projected 2026 free-cash-flow burn of $32 billion. Unlike a sentiment scare, a credit squeeze can force borrowers into damaging decisions: “This is a credit-originated problem.”
- The hosts agree that neocloud distress need not derail the broader AI trade; Jonah expects consolidation. His shale analogy is falling output prices meeting exhausted financing, followed by stronger operators buying distressed capacity. He calls it a “microcredit bubble”: creditors absorb losses and effectively subsidize hyperscalers’ acquisitions.
- Jonah favors buying certain AI inputs during sharp sell-offs. He likes copper, NVIDIA, and Intel, while avoiding neocloud equity and debt; defaults could disrupt chip orders before hyperscalers assume them, potentially with weaker supplier pricing. His longer-term demand argument includes an edge-computing market of “$100 billion or $500 billion.”
- Jonah doubts compute futures will become a major exchange-traded market. Rapid obsolescence, fragmented hardware, bespoke delivery, and incumbents’ interest in opaque pricing obstruct an enduring benchmark. He expects bilateral or brokered contracts instead, while conceding, “I could be wildly wrong. Don could be right.”
- Avi reads Bitcoin’s resilience during the Nasdaq sell-off as bullish. His rule is that bad news failing to push an asset lower often precedes an advance. He also remains bullish on ZRO after its move from $2.30 to $2, citing his earlier discussion with Brent and Helena.
- Jonah’s separate Bitcoin concern is an AI-driven cryptographic breakthrough. He worries future open-source models spawning “fifty million sub-agents” could create a six-to-twelve-month threat that Bitcoin developers are unprepared to address. Avi says quantum is the more plausible, though more distant, risk, and that breaking public-key cryptography would put the whole internet at risk.
1. Winning positions deserve capital, but access deserves scrutiny
- Avi’s unconventional-opportunity example is Donut World in Iraq: outsiders identified an underserved market and, he says, now make roughly $10 million annually. His point is calculated risk in overlooked places, especially as AI automates conventional work. “If you’re willing to look in areas where nobody else is willing to look,” opportunities remain. Jonah jokingly translates that into abandoning trading to sell donuts abroad.
- The trading analogue is Paul Tudor Jones’s “Losers average losers.” Avi argues against repeatedly adding to underperformers, especially after the original information changes. Proven winners deserve additional capital: growing revenue from $1 billion to $10 billion can be easier than moving from $100 million to $1 billion because product-market fit and relationships already exist.
- Jonah describes wealth as self-reinforcing: good opportunities arrive more readily once you are in a strong position. He compares Bezos’s venture-investing advantage to a pipeline of strong deals seeking his name. Jonah’s own early angel-investing losses discouraged him just as his network began producing better deals. His prescription is to seek annuities from weakness and swing bigger from strength. He warns that a deal arriving through a Twitter DM or AngelList syndicate has likely already been shown to top people who passed; better opportunities may come through more private channels.
2. CoreWeave’s credit problem requires outside funding to keep arriving
- Avi calls the recent sell-off an overreaction: Nasdaq fell 5%, and Intel, one of his largest positions, dropped from $120 to $90. He considers Intel attractive again but is not adding because he already owns his position. Jonah agrees the broader macro picture is fine, but says credit problems deserve more scrutiny than another tariff or geopolitical panic: liquidity and solvency problems can force damaging behavior.
- Jonah cites CoreWeave’s 2032 bonds trading at 79 cents and CDS requiring 16 points upfront. His model forecasts free-cash-flow burn of $32 billion in 2026, $34 billion the following year, $36 billion the year after, then $29 billion and $31 billion. Those deficits require external debt or equity. The bond deterioration therefore matters directly to the company’s ability to finance its plans; he calls CoreWeave the most creditworthy neocloud.
- Across hyperscalers and neoclouds, Jonah cites $641 billion and $231 billion of 2026 CapEx, respectively, describing total spending as roughly $1 trillion this year and $1.5 trillion next year. Aggregate free-cash-flow deficits reach $165 billion and $334 billion. CoreWeave represents roughly 10% of next year’s hole in his work; the Street estimates its gap at $20 billion, versus his roughly $35 billion.
- Avi locates the financing engine in Google, Facebook, and other hyperscalers. Borrowing at 6%, 8%, or 9% would still leave them investing; 15% and genuine concern about Facebook’s creditworthiness would represent a different situation. He accepts possible neocloud trouble but questions its transmission. Jonah agrees Microsoft remains free-cash-flow positive and expects the largest hyperscalers to manage their funding needs.
3. Distressed neoclouds could subsidize the next phase of AI expansion
- Jonah’s shale comparison carries the mechanism: producers overstated economics by excluding ancillary costs, expanded supply until oil prices collapsed, and then lost financing simultaneously. Larger operators acquired the wreckage. He expects a similar squeeze in compute, where spot prices currently exceed long-term contracted prices: convergence and falling prices undermine leveraged neoclouds, leaving hyperscalers positioned to buy them. He qualifies the company list—Lambda is “probably fine.”
- Avi’s challenge is whether NVIDIA, Intel, Google, and the Nasdaq actually suffer lasting damage if hyperscalers simply acquire the capacity. Jonah’s researched conclusion is categorical: the “microcredit bubble” will not metastasize into a broader AI-chip or Nasdaq meltdown. Creditors will recover less than par and subsidize hyperscalers’ purchases at “cents on the dollar.” Contract defaults can still create substantial chip-stock volatility before orders transfer to stronger buyers, potentially with weaker supplier pricing.
- The positioning follows that distinction. Jonah wants to avoid neocloud equity and debt, buy unrelated assets during spillover panics, and buy NVIDIA or Intel if they collapse on these fears. Copper remains attractive because it connects the infrastructure; he also likes regional electricity around new data centers, while admitting he does not know how to express that exposure. He would not buy a compute future here.
- His longer-run demand case extends beyond hyperscalers, currently the “marginal barrel.” Jonah expects cheaper tokens to push computing toward the edge, creating another “$100 billion or $500 billion” market; he highlights the CPU intensity of that opportunity and Intel’s positioning. His Jevons-paradox argument is that cheaper compute creates additional demand for compute capacity, with prices staying relatively consistent over time.
4. Compute lacks the stable benchmark that makes commodity futures work
- Jonah respects Don Wilson’s record, particularly establishing Cumberland early in crypto, but rejects treating Wilson’s compute activity as proof of a future multitrillion-dollar derivatives market. Exploration can be a relatively cheap option for DRW. His disagreement comes with an explicit concession: “I could be wildly wrong. Don could be right.” He expects compute futures to appear but doubts they will take off.
- The first obstacle is an enduring benchmark. Jonah says oil’s chemistry has remained the same for tens of millions of years; an H100 contract loses relevance as buyers move to H200s, B200s, Rubins, Feynmans, and newer chips. Meanwhile, AMD, TPUs, Trainium, and custom hyperscaler ASICs could further fragment the market. Jonah distinguishes this from equity indices that successfully change constituents: an evolving stock index does not establish that rapidly changing compute can support a commodity future.
- Physical delivery adds another layer: PCIe versus SXM, InfiniBand versus Ethernet, eight-GPU nodes versus 1,000- or 4,096-GPU clusters, location, uptime, and reserved versus spot capacity all affect what buyers receive. There is no tight, hub-delivered standard comparable to oil. Avi supplies the gemstone analogy: repeated attempts at diamond futures struggled because products were so individual, although lab-grown diamonds are beginning to permit greater standardization.
- Jonah’s strongest objection is incentive alignment: incumbents profit from opaque prices and wide gaps between spot and long-term contracts. His LNG analogy highlights intermediaries such as Vitol, BP, and Shell benefiting from informational advantages. Why would compute incumbents surrender theirs? He expects financialization through ISDA-like bilateral agreements or brokered blocks, potentially settled through a clearing venue. A successful futures market, he says, would still have unusually extensive fragmentation and basis differences.
5. Bitcoin’s price resilience meets a separate cryptography anxiety
- Avi’s closing market signal is straightforward: when bad news stops pushing an asset lower, an advance often follows. Bitcoin held up despite substantial Nasdaq selling and AI fears, which he considers constructive. He notes Hyperliquid nearly reached $100 before retreating and ZRO moved from $2.30 to $2. His bullishness on LayerZero draws on the earlier Brent and Helena interview about developments in computation.
- Having dismissed AI-credit contagion as a reason to fear Bitcoin, Jonah introduces a different concern. He calls quantum risk a “hoax” but worries an open-source model—perhaps generation six or seven—could spawn “fifty million sub-agents” pursuing advances in breaking public-key cryptography. Watching AI solve mathematical problems has changed his confidence. He doubts Bitcoin developers are prepared for a six-to-twelve-month non-quantum threat, despite their engagement with a ten-year quantum horizon.
- Avi says quantum is the more plausible risk, though further away, and that a mathematical break in public-key cryptography would compromise far more than Bitcoin. Jonah challenges Avi’s claim that this would put the whole internet at risk, pointing out that there is no public key for a JPMorgan account; Avi argues that public-key cryptography secures internet interactions. Jonah asks whether he should therefore hold his Bitcoin. Avi reiterates the broader threat and his ranking of the risks; the technical disagreement remains unresolved.
Full transcript
Again, I usually have an “everything is fine” approach. I usually just believe that nothing ever happens. I’ve been long and strong for many years of a multiyear megatrend rally. I’m concerned about neoclouds creating intra-AI volatility, but for the first time in basically this decade, I’m starting to get worried about Bitcoin. Not because of quantum—I think that’s a hoax. That’s long-cycle FUD that won’t pan out. But I am worried that—
Good morning, good afternoon, good night, wherever you are. I’m coming to you live from Israel. I’m sitting in my hotel room, and—
Yo.
I am ecstatic. Do you know why I’m ecstatic, Jonah?
Why, Avi? Tell me.
Because the market’s looking great. The market’s looking amazing. This is the best part about it, right? We don’t podcast for 2 weeks, everything hits the fan, and I guarantee you this podcast has marked the bottom. We’re going straight up after this. We skipped all the nonsense.
I’m sorry we left you out to dry, guys, but if you remember one thing from the 1000x podcast, it’s that you can never panic-sell and never bet against America. We’re going up. Where was I the last 2 weeks? I was everywhere, by the way. I was in Singapore, and we did a lot of very good interviews there.
We interviewed Arthur Hayes. For those listening to the podcast live, you’re going to get some intelligence here. We interviewed David Choi from USD.AI, and I’m very bullish on CHIP now. That’s a hint. That episode is going to come out at some point. He dropped some very interesting things on my podcast related to how the token is going to accrue value.
I also interviewed Illia from NEAR. I’m less bullish on NEAR, candidly. I really like Illia; he’s a super smart guy. I didn’t really think he answered some questions as well as he could have. Then I interviewed Huff from Pair Protocol. That episode is going to come out next week.
After Singapore, I was in Istanbul, where I sent out a tweet that angered a million Turks. I said Istanbul was the prime example of Islamic conquest, and that people need to stop being so mad about Western civilizational conquest while ignoring Islamic conquest. That got me a lot of followers. Then, promptly after that, I went to Iraq. So—
Are you in the Special Forces, Avi?
I’ll tell you this, Jonah. This is something that not a lot of people know, because I didn’t know it until a week ago. One of my close friends has a close friend who owns the largest donut shop chain in Iraq, called Donut World. I want you to guess how much Donut World makes every week.
I bet they make bank, Avi. Let me guess: $100,000 in net profit every week.
Something like that. They’re making, like, $10 million a year, Jonah. Donut World in Iraq. And you know what? These people aren’t Iraqi. They just came in there and said, “Iraqis need to eat donuts. There isn’t anyone offering donuts to these people.” Now they’re making $10 million a year, and they’re much richer than we’re ever going to be because they’re selling donuts to the right people.
The key is that in order to make money, you have to take calculated risks. If you’re willing to look in areas where nobody else is willing to look, if you’re willing to go sell donuts to the Iraqis, you’re going to make money. You can’t bet on conventional ways of making money, because the world isn’t conventional anymore. With AI, all the conventional stuff is getting automated away. You have to look for the opportunities nobody else is looking for. That’s why I was in Iraq—and a wedding. Mostly the wedding, but—
Yeah, I was there scoping out the donut market. Basically, the way I read this is: just give up. We like to be actionable on the 1000x podcast, where 2 traders get together to talk about trading every week. Our action item for this week is: just don’t trade. Sell donuts in Azerbaijan or Iraq or maybe Kazakhstan. There are a lot of people in Kazakhstan and not a lot of donuts—or maybe there are. You should go to Kazakhstan and find out. That’s kind of what we’re getting at here.
1. The Winners Keep Winning
I’m going to frame that. You know how Paul Tudor Jones has “Losers average losers, winners average winners” on his desk? There’s that famous picture of him at his desk, leaned back in his chair, with a sign that says, “Losers average losers.” What that means—
What does that mean?
Very specifically, if you’re in a trade and it’s losing, it’s underperforming, and you keep doubling down—especially if the information you made the bet on has changed—you’re a loser. You’ve got to bet on the winners, and when the winners win, you double down.
Mm.
When the losers lose, you get out. People make this mistake. The world today—less so in the past, more so today—is a world of winners winning. The first billion is the hardest. Going from $100 million to $1 billion in revenue is extremely difficult. Going from $1 billion to $10 billion in revenue is actually easier. Why? Because you’ve proven your product-market fit. You’ve built the relationships that need to be built.
Once you see an investment working, oftentimes it’s better to double down than to sell and try to buy something that’s underperforming. This is my view on things. That’s what “losers average losers” means: you’ve got to bet on the things that are doing well. I kind of like the idea, in today’s world, of framing it as “sell donuts and rock.”
Somebody just wrote, “Is this all satire?” and I wrote back, “Sort of.” I think the losers— You know what’s interesting? Wealth is a very self-reinforcing state. What even is wealth? I noticed that when I had no money and had a garbage job after Lehman blew up, I was basically fighting for my life at Barclays Capital, which took over Lehman’s bankrupt units and then proceeded to fire 14,000 people in a very short period of time.
I had to re-interview for my job every Friday, constantly on edge. That’s a story for another time. That was crazy. I had no money in my checking account, and my job was pretty borderline. There was no way to escape that. The only way out was through. There wasn’t some secret hack or opportunity being presented to me that I could react to that would give me an escape hatch.
I was in a losing position, and I was averaging losing outcomes. I ground away and did back-office work for a very long time, answering the phones and such to survive. My pitch to the people who re-interviewed me every Friday was, “Why would you fire me? I’m the cheapest person here. When you fire the more senior guys, who are kind of duplicative with each other, who’s going to be younger, more tech-savvy, and better at figuring out their spreadsheets than me? Obviously, it would be ridiculous.”
They’d say, “Yep, no, we agree. See you next Friday.” That was my only option. Now, after slowly accumulating some stuff and some money, good stuff gets pitched to me. I can react, and these opportunities come my way, whereas they wouldn’t have come the way of 22-year-old me.
I totally agree with you that when you’re in a winning position—whether you’re a wrestler on the mat, a trader in the market, or a professional in some good business system—valuable stuff just lands on your plate. That’s why Bezos is probably one of the world’s best venture investors, not because he’s out there actively grinding on venture, but because his pipeline fills up with pure gold. Everybody wants his name on their cap table.
All he has to do is tell his assistant to select smart-seeming stuff from the cream of the cream and put dough into it, and it’ll probably earn 50% to 250% per year. I totally agree with that position. I think the takeaway as a trader is: when you’re in a position of weakness, try to find annuities. When you’re in a position of strength, that’s when you should start swinging big.
I made this mistake in venture, too. When I was in my 20s, I’d angel invest in stuff my stupid retard friends would show me when they started companies that were kind of bad ideas, and all of those went to zero. Now that I have smarter, more well-connected, and more successful friends, I get shown better deals.
Of course, I fumbled and stopped angel investing for a while because of my failures in my 20s, getting shown bad deals, so I missed out on a bunch of great deals. Now I’m like, “Oh, I see.” It’s just a representation of where you are in life.
So basically, I think the big takeaway—I’m back to angel investing again—is that if it’s coming to you through a Twitter DM or an AngelList syndicate, it’s been shown to all the top people, and they’ve already passed, which is why the round is still open. If you’re somebody with a name, it’s more private, and you’re getting a look at something. That’s when it’s more likely to win. So, yeah: winners average winners, losers average losers. I really like that. It’s kind of meta-convexity to trading opportunities. A really interesting framework, Avi. I like that one a lot.
It says, “Avi’s Wi-Fi is sponsored by Bin Laden.”
That is the best comment we’ve ever had. Avi’s Wi-Fi is sponsored by Bin Laden.
Yeah, no, it is blowing up. Hopefully, we can get through this and see what’s going on. Long story short, the entire market overreacted recently to this mini debt crisis in the AI market. You had a 5% sell-off in the Nasdaq. Bitcoin came off a lot. Intel, one of my biggest positions—this has hurt a lot. It went from $120 to $90. I think it’s a good buy here again, but I’m not buying because I already have my position.
I think that right now we’re climbing the wall of worry again. These are the kinds of opportunities where people are worried about so many different things in the market, but the reality is that a lot of these things are coming to a resolution soon.
I don’t think so. I think it’s way more nuanced than that.
Go for it.
Normally, I’m the guy to say, “Buy the dip. It’s all going to be fine.” This time, yes, buy the dip; it’s going to be fine for now. But I looked into this. CoreWeave’s bonds—something real is going on here, Avi, and I think we’ve got to talk about it. Some very smart people I work with and I dug deep into this, because this is starting to turn into something serious. This isn’t a panic over tariffs, some weird macro glitch, or people freaking out about geopolitics, where you can do what we’ve talked about on previous podcasts and analyze it with common sense.
This is the only kind of credit crisis I ever take seriously, Avi. Credit problems are important because when people run out of liquidity and solvency, they’re forced to do things in the market that are, at best, suboptimal and, at worst, catastrophic. This is a credit-originated problem, so I’m paying more attention to the potential downside and taking it more seriously than I have since 2020, the last time we had a real credit problem. We had a mini credit problem in crypto in 2022, but at the societal scale, you haven’t had a credit problem since COVID shut things down. Obviously, the government two months later literally said, “We’re turning on the cash machine,” and that was the best buying opportunity of a lifetime. Anybody who learned their lesson in 2009 would know that. I did. That was one of the best runs I’ve ever had in my life.
Here, another credit problem is emerging. This time, it’s not that people can’t pay back their debt because they can’t go to work during a pandemic. It’s CoreWeave. There’s a very important section of the economy—data centers, hyperscalers, AI, neoclouds, this whole group of companies. You put computer chips, power, land, electricity, water, shells, employees, and all kinds of CapEx related to data centers into them, and out comes compute and income from that compute. CoreWeave basically does that. The CEO of CoreWeave has gone on many different forms of media saying that you plug the inputs I just mentioned into the box, and out comes compute and income from renting out the compute.
These bonds have gone into distressed territory. Their CDS trades 16 points up front, last I checked. They’re 2032 bonds—basically six-year bonds—and they trade at 79 cents on the dollar. This is a real credit problem. We looked into CoreWeave’s numbers. Let me see if I can screenshot this and add it to what you guys are seeing, because this is really important, Avi. This is something we can’t just fudge.
I have a very strong take on this, which I’ll let you finish with CoreWeave, but I think the macro picture is a lot rosier than you’re painting right now.
Yeah, the macro picture is fine, but I’m starting with CoreWeave, and then I’m going to zoom out. You’re right—we agree. We modeled CoreWeave. In 2026, they’re going to incinerate $32 billion of free cash.
Can you zoom in on that a little?
Yeah, sure. Can you see this? This is their free cash flow.
Yeah.
Okay?
Okay.
In 2026, this year, they’re going to incinerate $32 billion of free cash flow. Next year, they’re going to incinerate $34 billion of free cash flow. The year after that, $36 billion. The year after that, $29 billion. They’re going to incinerate $31 billion. All of this money has to come from public markets, so people have to keep extending CoreWeave debt or CoreWeave has to sell equity. This money comes from somewhere, and it’s not coming from their business.
The fact that their bonds just got shanghaied like this is extremely relevant. I’m going to share my screen again. Brad, can you throw it back up there? This represents their ability to raise the money for the deficit you’re seeing here, Avi. This is critical. Now, is CoreWeave going to tank the entire economy? Where we’re going with this is no, but it is going to create a big problem.
UBS just came out with numbers similar to our model and literally initiated CoreWeave with a buy rating. This reminds me so much of shale. Wall Street is so heavily invested in making sure Anthropic, OpenAI, and all the neocloud companies in these IPO pipelines come in and bring those fees that there’s basically a smoke-filled back room of equity researchers publishing the numbers I’m showing you now. The money has to be funded by somebody. The bond chart is there. They can’t fund it anymore. The only reason they’re slapping a buy rating on it is vibes, which is really, you know, pay the bank 100 fees.
Well, okay, look. To be completely fair, all equity analysts do all day is slap buy ratings on companies for vibes. That’s kind of their whole job, which is why nobody takes the ratings seriously. Nobody on Wall Street will ever take an equity analyst recommendation seriously. The reason people read reports is for the information they contain. You take in that information so you don’t have to go out and get it yourself. These reports are really a fact-finding mission. Any serious investor almost always throws out the end conclusions.
I wanted to take a step back and say that I agree CoreWeave might have an issue, and I even agree the debt of these neoclouds might have an issue. But the real engine of financing is the hyperscalers. The real engine of financing is Google. The real engine is Facebook. That’s where the money is coming from. Whether their bonds trade at 6%, 8%, or 9%, they’re making that investment.
If we get up to 15% on these bonds, if Wall Street really falls off and says, “I’m not financing any of this stuff anymore,” and if I actually believe that the credit risk of Facebook is now high, that’s a totally different dynamic. The argument you could make is that CoreWeave is a canary in the coal mine, and that it’s telling you that at some point the hyperscalers will have an issue. But for right now, I don’t necessarily see that correlation, because CoreWeave itself is kind of a shitcoin.
It is a serious asset, but it is kind of a shitcoin.
2. The Neocloud Shale Bust
Bro. Bro. Okay, let me take this a step further. CoreWeave, in my opinion, has a problem. Their CEO says, “Okay, we have all these subsidiaries into which you plug inputs that we’re spending all this money on, and out of which comes compute. Oh, and by the way, the deficit between the amount of money that goes in and the amount of money that comes out is roughly $30 to $35 billion per year. So we’re going to go ahead and raise that.” CoreWeave is the most creditworthy of all the neoclouds.
Brad, if you can throw my screen back up there again. This is the hyperscalers and the neoclouds. This is their AI CapEx: the amount of money that goes into the box, meaning what they spend on chips, power, land, water, and whatever else. These are street estimates: $641 billion of hyperscaler spend in 2026 and $231 billion of neocloud spend. You can see the neoclouds here: CoreWeave, Nebius, and so on.
Cumulatively, that’s about $1 trillion of CapEx this year and about $1.5 trillion next year. But that’s against a free cash flow backdrop of negative $165 billion this year and negative $334 billion next year. So CoreWeave’s $35 billion deficit is about 10% of the total free cash flow deficit next year. And you can see that—
Go into the clouds again.
Sorry, I can’t hear you, Avi. You’ve got to try not using your AirPods. Something needs to change. I can’t hear anything you’re saying.
I said, go into the neoclouds again. Yeah.
Street estimates show a little less free cash flow burn than our estimates. But basically, Microsoft is doing just fine. They’re sitting pretty. Amazon is going to have to raise $46 billion next year if the Street is right. Alphabet is going to have to raise $53 billion. Meta is going to have to raise $37 billion. Oracle is going to have to raise $62 billion. SpaceX has to plug a $109 billion gap. For CoreWeave, the Street says $20 billion; we say $35 billion. Nebius is at $22 billion.
Here’s what’s going to happen, Avi. I don’t think the big hyperscalers like Microsoft, Amazon, and Alphabet are going to have a problem doing what they need to do. Microsoft is actually free-cash-flow positive. That company is bulletproof. They’re Teflon. But the neoclouds are going to collapse, just like the little mom-and-pop shale producers—the neooil producers of the 2010s—collapsed. It’s going to be the same story.
Back then, the shale producers were saying, “Hey, we’re producing lots of oil. Oil prices are high. Everything’s great. We’re free-cash-flow positive.” Then suddenly oil prices started to tank because they produced so much oil. All the economics they were reporting to raise the hundreds of billions of dollars they cumulatively needed from the Street were based on fudged math. It was basically wellhead cost versus open market, like water board and crude cost. It didn’t include all the costs of bringing it to market, all sorts of other ancillary input costs like labor and getting people to pump these wells. All of those costs exploded.
The exact same thing is happening with the neoclouds now. Think of the neoclouds as the shale oil producers of the Bakken, Eagle Ford, Texas, and all the different fields in America in the 2010s. They went upside down on their free cash flow. Wall Street got tapped out, and then all that extra shale came to market and crushed the price of oil. They got double-dildoed at the very bottom, where oil prices were super low and Wall Street got tapped out at the same time. Correlation went to one. Then the big boys—the Exxons, Chevrons, BPs, and Shells of that market—just slurped up all those companies.
Here’s what’s going to happen with AI. CoreWeave is a canary in the coal mine, but it’s not something that should scare you out of holding your broader-cycle bags. Every time there’s a panic over this, you should buy dips in non-AI-related assets, because it doesn’t matter for those assets. But within the AI space, compute prices are going to tank. Right now, spot trades way above long-term agreements for compute. That will converge. Then all these little companies will start to explode, and you’ll see Microsoft, Meta, and Amazon start slurping up CoreWeave, Lambda, and whoever else.
Maybe not Lambda. Lambda’s probably fine. But all these little neoclouds—maybe AIREN, I don’t know—are basically going to get acquired by the big boys, the big hyperscalers like Microsoft and Oracle, right as compute prices crash.
That doesn’t seem bad for the hyperscalers at all. And it doesn’t seem bad for anybody except for the neoclouds that have taken on too much debt. In fact, you can make—
That’s my point, Avi. That’s my point.
I know, but you’re painting the AI market with a broad brush. The AI market contains a lot of different companies. For example, are NVIDIA, Intel, or other chip producers going to go down? The only way they really go down is if we don’t get those continued build-outs from the hyperscalers. If debt drags down the neoclouds and the hyperscalers just buy them up, then we’re on track.
Yeah.
There’s no crack in the market, right? That’s what people are really worried about. Nobody’s sitting here saying, “I’m really nervous because I think CoreWeave and Nebius will go down 8%.” The question is, will the Qs go down? Will Intel go down? Will Google go down? Will nothing go down? That’s the core of the issue.
But that’s what I’m getting at. That’s why I really wanted to do this deep dive with you, Avi. Rather than just lazily saying, “Buy the dip, it’s going to be fine,” I wanted to show that I actually did the work this time. I went in, and my conclusion is that this is a microcredit bubble that’s going to result in industry consolidation on the chip-demand, the chip-consumer side.
I’ve now concluded 100% that it’s not going to metastasize into a broader AI chip stock meltdown that will then metastasize into a meltdown of the Qs. The AI trade is very much intact. Basically, what’s happening is the creditors of these companies aren’t going to recover 100 cents on the dollar, and they’re going to end up subsidizing the hyperscalers’ AI build on their own dime. The hyperscalers will pick up these companies for cents on the dollar, the way Chevron and Exxon bought up shale producers for cents on the dollar and then the oil price recovered.
I think chipmakers like Intel and NVIDIA will experience incredible volatility throughout this consolidation, because there will be times when people default on their contracts to buy Intel or NVIDIA chips. But then those contracts will get novated over to Microsoft and Meta, who will resume buying—potentially not with the same pricing power NVIDIA has today, but close.
Also, this doesn’t include what I sincerely believe will happen: I don’t think hyperscaler demand is going to be the only demand for these chips. Right now, it’s the marginal barrel, so to speak. It’s the marginal source of demand. But in the future, I fully expect compute to go to the edge, especially as token pricing goes down. Then there’ll be a whole new $100 billion or $500 billion market for edge compute.
Intel’s already talking about it with the next-generation chips they’re putting out. Edge computing is very CPU-heavy. There will be a second wave of demand, but I think this is going to create a lot of intra-sector volatility. That will result in dip buys for NVIDIA and dip buys for Intel. Stay away from neocloud equity and debt, and don’t worry one bit about your Qs or your Bitcoin. Do you kind of get what I’m saying?
Because you were quite bullish on copper for some time, does this change your short-term outlook?
Copper is just like semiconductors, right? It’s an input to this whole thing. You need the sand that thinks. NVIDIA makes that—well, TSMC makes it. NVIDIA FTPs files to TSMC. That’s what NVIDIA does, and TSMC makes the sand think. So that’s an input. Then you need copper to connect it all together. You need real estate. You need power.
I’d certainly be long if I could. I don’t know how to do this, but I’d be long regional electricity wherever data centers are getting stood up. I love copper. I think it’s something to dip-buy. I also love dip-buying NVIDIA and Intel if they really collapse, because people haven’t done the work I just did. People are like, “Oh, look at CoreWeave bonds. This whole AI thing is unwinding.” So they’ll panic-sell Intel and NVIDIA on that. You should be there to provide liquidity to those morons.
It’s really just a neocloud-specific problem, in my opinion. And I think the broader Jevons paradox—cheaper compute creating more demand for compute capacity, creating more compute demand, and prices staying kind of consistent for compute—will play out.
I also wouldn’t buy compute here if people had a compute future, which I don’t think is going to happen. I wouldn’t buy that future.
3. Why Compute Futures Fail
You said you saw that your former boss, Don Wilson, came out and said that we’re going to get compute futures pretty soon.
With all due respect in the world for Don Wilson, we will get compute futures, but I don’t think they’re going to take off. I respectfully disagree with Don on that one quite strongly.
That’s fair.
I have a whole bunch of reasons for it, too. I think people who are investing their careers trying to make fetch happen on the compute trading front should stop trying to make fetch happen. It’s not going to happen. I can talk you through it. I’ve been talking for a while, but I can tell you why I think it won’t happen.
I mean, we’re in Utah, and my audio is the worst thing that’s ever happened.
Your audio’s pretty bad. Yeah. Here, how about this? Why don’t I monologue—
Okay.
—on why compute futures aren’t going to happen, why I disagree with Don Wilson, and why you should try to log off and hotspot with your phone? That’s the best I can think of.
So here’s why. Don Wilson is a man I greatly respect and think is an incredible trader. He seems interested in compute trading. I don’t know what his exact views are, so I can’t represent them. But he’s behaving like the type of trader who believes compute derivatives are going to be a big market. I don’t think so.
I will concede that Don has been incredibly right about a lot of markets before they took off. He was early to set up Cumberland when Bitcoin was kind of a joke, and he was dead right that crypto would become a huge market. He was super well-positioned before anyone else took it seriously. He’s doing the same thing with compute, which is generating a lot of buzz online because he’s so respected and well known for being early to big things.
Another thing he does well is buy options at good prices and farm optionality. I’d guess that if compute trading doesn’t happen, or doesn’t become something that transforms society at scale, with multi-trillions of dollars of open interest on various exchanges that DRW is connected to, Don and DRW will be just fine. I’d bet my entire life savings on that. So I think it’s relatively cheap to explore this. But the fact that Don is exploring it isn’t a signal, as many online are saying, that the rest of us should obtain some degree of conviction that compute is going to become a multi-trillion-dollar open-interest market like many commodities.
Here’s why I don’t think so. This is just my opinion; I could be wildly wrong. Don could be right. I’ve disagreed with him in the past, and he’s been right way more times than I have, because he’s the man. But there are a couple of reasons.
First, let’s look at successful commodities like oil and gas. These are commodities with enduring benchmarks. A barrel of oil is, by definition, the same today as it was 65 million years ago. The barrel doesn’t change; it’s the same chemistry. A benchmark has to be enduring. It can’t mutate every year or every six months.
There are benchmarks that kind of mutate, like the S&P 500 and CDX investment-grade or CDX high-yield credit benchmarks. But an equities benchmark is very different from a commodity future. Most people don’t care about CDX. I wouldn’t call that a successful market; I’d call it a niche. The S&P is super successful, but an equity index, again, isn’t a commodity future.
Due to obsolescence, there’s literally no enduring benchmark for compute. A contract on H100s will lose relevance as the world rolls to H200s, B200s, Rubins, Feynmans, and whatever NVIDIA is cooking up next.
Second, the chip universe is already fragmented, and it’s going to fragment a hundred times more. NVIDIA won’t be the only GPU game in town for long. You’ll have AMD, TPUs if they decide to socialize them, Trainium, and custom hyperscaler ASICs. That market is just going to go in a million different directions. There’s going to be a Cambrian explosion of different types of compute, which makes it harder to have a futures market.
You could do something super convoluted, like Platts does with grades of North Sea oil. But those are very, very similar things that endure for tens of millions of years. With these chips, you have to keep rotating them.
The other thing is that the physical is just too bespoke. What underlies an oil future is a solid benchmark with physical delivery to a hub. You just need that, period. The physical in compute is too bespoke. You have PCIe versus SXM, InfiniBand versus Ethernet, 8-GPU nodes versus 1,000-GPU or 4,096-GPU clusters, region, uptime, and reserved versus spot. You don’t have that sort of physical delivery at a hub of something that’s basically—not exactly the same, but within a very tight band of tolerance—the same.
Furthermore, this is the big one: incumbents benefit from opacity in compute. The price of their long-term contracts is so far from spot. This is what has murdered so many attempted futures benchmarks in the commodity space, especially LNG. LNG is a big enough and consistent enough commodity that you’d think there would be a good futures market for it, but there isn’t. Why? Because you need buy-in from the people who stand to benefit most from opacity, like Vitol, BP, and Shell. When spot is all over the place versus the forward, nobody benefits more than a middleman with visibility into something that’s opaque to everybody else. The exact same is true with compute, and I don’t understand why those people would suddenly open the kimono and let it all happen.
I do think compute will financialize eventually. But I think it’s going to be OTC contracts, like what you have with interest rates, where there’s an ISDA standard and people can trade compute bilaterally, or have brokers arrange a block trade, then settle it on something like ClearPort. I just don’t see it. Everybody’s talking so breathlessly about compute futures, but I think they’re never going to happen. I could be totally wrong. If they do happen, they’ll definitely be the most unique, weird, fragmented market, with thousands of different bases, in the history of commodity futures.
That was a great point. I think the underlying point that comes across is that there are far too many different types of compute to create any sort of cohesive market. That makes a lot of sense. It’s a corollary to my time looking at the gemstone market. People have attempted over and over to create diamond futures, and it’s just impossible because the market is so fractured. It’s even more fractured in the diamond market because everyone is so deeply individual. That’s changing a little bit now because you can create standards with lab-grown diamonds.
But any market that has, first, incumbents that set pricing in the first place, and second, a massive amount of differentiated products, is very hard to create futures on. At the same time, Don is a smart guy. I need to dig into his arguments more. We can have a little debate between you and Don about what’s going to happen with the futures market.
4. Bitcoin Refuses To Break
I’m so short on internet right now. We’re going to upload this, clean up the audio, and figure it out. Before I leave, I want you guys to know that generally, whenever an asset faces bad news and doesn’t go down further, we’re going to see it move up first. I’m talking about Bitcoin. Bitcoin did not break despite the Nasdaq selling off tremendously and the fears in the AI market, and we’re pushing back toward the highs. I think this is generally a pretty good place, especially for the assets that are currently winning.
Hyperliquid almost touched $100, then came back down. ZRO touched $2.30, then came down to $2. I’m taking outlays on all these assets. Especially after the interview with Brent and Helena, which, if you haven’t listened to, you should listen to. I think there was a phenomenal outline of what exactly is happening behind the scenes in the world of computation, and specifically LayerZero is part of it. I’m quite bullish on that asset.
Okay. I haven’t done much work on LayerZero, but I’d like to. We should definitely do another session when you’re on a better connection.
5. The Cryptography Risk Emerges
I usually have an everything-is-fine approach. I usually just believe that nothing ever happens. I’ve been long and strong for many years of a multi-year mega-trend rally. I’m concerned about neoclouds creating intra-AI volatility, but for the very first time in basically this decade, I’m starting to get worried about Bitcoin.
Not because of quantum—I think that’s a hoax. That’s long-cycle FUD that won’t pan out. But I am worried that maybe it’s not Fable 5, but generation 6 or 7: an open-source version that’s powerful enough to spawn 50 million subagents, each enabled with new innovations in cracking public-key cryptography. I’m starting to get a little worried about that. Watching AI solve math problems, both inside and outside of cryptography, makes me concerned for the first time. I think the developer community in Bitcoin is engaged enough to get ahead of the 10-year quantum risk, but not engaged enough to get ahead of a 6- to 12-month non-quantum public-key cryptography risk.
Jon, it’s far more likely that quantum is an issue than that we actually crack the entire mathematical formula for public-key cryptography. If we do, Bitcoin is going to be the last thing to worry about. Everything is up for grabs.
No.
Everything.
No.
Literally the entire internet.
No, dude. There’s no public key for your JPMorgan account.
Yes.
There is no public key for that.
Dude, the idea of one-way cryptography literally secures every interaction. There’s a public key every time you connect to Wi-Fi, every time you connect to the internet. When you send data back and forth, it’s secured by public-key cryptography.
Hmm.
That’s all. All of that data is totally up for grabs at that point. It’s way worse than just Bitcoin, and it’s not something that I think… It’s sort of like worrying about doomsday, in my personal—
So you’re saying I shouldn’t vomit out all of my Bitcoin, basically? I should hang on to it?
I’m saying I think quantum is a real risk. Quantum is the bigger risk, but it’s further out. The idea of finding a mathematical solution to breaking public-key cryptography has a far greater impact than just Bitcoin. It impacts the whole internet. In that case, it’s like worrying about logging into your bank account.
Okay.
Worry more about that before you worry about Bitcoin. Anyway, I think that’s as good a place as any. Maybe on the next pod, we can dive into that a little bit more.
Yeah, great talking to you, Avi. Until next time. Thanks, man. Nothing said on the ThousandX podcast is a recommendation to buy or sell any investments or products. This podcast is for informational purposes only, and the views expressed by anyone on the show are solely their opinions, not financial advice or necessarily the views of 1kx Media. Our hosts, guests, and the 1kx team may hold positions in the companies, funds, or projects discussed.