Arthur Hayes vend à découvert Bitcoin : l'IA a épuisé la liquidité, le BTC a du mal à dépasser les cent mille

Arthur Hayes liquidation positions HYPE, NEAR, etc., believing that the AI bubble will burst and swallow all risk assets, making it difficult for Bitcoin to rebound to $100,000 in the short term.
(Background summary: Legendary investor Carman warns of AI bubble: his firm Baupost has refused to invest in OpenAI and Anthropic)
(Additional context: Arthur Hayes: AI is the new subprime crisis, market narrative shifting from deflation to war inflation, Bitcoin expected to hit $125,000 by year-end)

Table of contents

Toggle

• Midterm elections: Trump’s AI reversal strategy
• SpaceX IPO: a time bomb with 100x P/S ratio
• AI bubble: liquidity draining crypto markets
• Rate hike expectations: pivot and collapse
• Investment shift: energy stocks outperform Bitcoin

This article is summarized from Kyle Chasse’s crypto analysis on YouTube. Arthur Hayes has liquidated his largest crypto holdings—HYPE, NEAR, Worldcoin, and Zcash—not because of crypto itself, but due to a macro chain of reasoning from oil prices, Iran war, Trump’s midterm strategy, to the bursting of the AI bubble. He believes Trump may reverse his stance and attack the AI industry to salvage midterm election prospects, and once the AI bubble peaks, the crypto market cannot remain unaffected; SpaceX’s IPO with a valuation of $1.8 trillion and a 100x P/S ratio, in his view, is a liquidity time bomb waiting to explode.

Host Kyle Chasse: Arthur, welcome back. Recently, you sold all your holdings in Zcash, HYPE, NEAR, and everyone is criticizing you for exiting the scam, raising prices to dump, etc. Why did you sell everything? What’s really going on?

Arthur Hayes: I just published an article called “Reality Check,” about five thousand words, outlining the points I’ll briefly discuss in this podcast. If you want a deeper understanding of my reasoning, I strongly recommend reading it on my Substack. But essentially, it’s about a reflexive interaction between oil prices and Trump’s midterm campaign rhetoric—he needs to help the Republicans beat the Democrats in November and hold the House and Senate. The current Iran war—whether you like it or not, it’s there, right here, right now.

So Trump and the Iranian Revolutionary Guard need to reach some agreement to end this conflict. Both sides face a real constraint: oil prices determine how angry different regions of the world are at each other. Trump must worry about domestic voters—high oil prices and energy-driven inflation are unpopular.

Iran faces pressure from China and other developing countries—“What are you guys doing? We need this oil, these goods passing through the Hormuz Strait. I know the US attacked you, but find a way to fix it.” So the higher oil prices go, the more eager everyone is to negotiate; when prices fall suddenly, no one wants to reach an agreement anymore. We’ve been swinging back and forth in this tug-of-war for about three months, or as long as the war has been ongoing.

As this process unfolds, we are gradually depleting commercial and national reserves of oil and other hydrocarbons. Any energy analyst’s chart shows similar conclusions—pre-war inventories were ample, so people believed oil and natural gas supplies were oversupplied, leading to relatively low prices.

But now we are consuming these surpluses at an ever faster rate. At some point, we will reach a certain level—I don’t know if it’s tens of billions of barrels—each analyst has their own number and projected date. Once that date passes, the situation will suddenly become very, very bad. The only way to restore market balance is to rapidly push up oil prices.

This is the worst-case scenario—Trump and the Iranian Revolutionary Guard cannot reach an agreement. By October, the Hormuz Strait remains effectively blocked, with only 25-30% of shipping volume passing through, far from enough. A more likely scenario is that in a month or two, some agreement is reached, and shipping through the strait somewhat resumes.

But then everyone needs to rebuild inventories—you must rebuild national reserves, and of course, you will stockpile even more than before—because you’ve just experienced firsthand the consequences of being completely at the mercy of Trump and a group of Iranian generals, who decide whether your country receives goods.

Midterm elections: Trump’s AI reversal strategy

So you think: “I need to stockpile more oil, natural gas, helium—everything needed for a modern economy.” This will increase demand, though it may not push prices to catastrophic levels, but it still means that in three or four months, oil, natural gas, and other commodity prices will be higher than today.

Arthur Hayes: Following this logic, Trump and his Republican allies’ midterm election (November 2026) has a high chance of ending with the Democrats taking control of the House. Just check Polymarket—odds of Democrats retaking the House are now up to 82%.

Why? Clearly, Trump is being beaten on the issue of inflation. People think inflation is terrible and worsening. In voters’ eyes, it’s the Democrats in the White House, and this damn conflict and war are their fault—so the blame naturally falls on the Republicans. That’s why everyone expects them to lose, and badly.

The problem is, there’s little you can do about inflation—policy has long lag effects, and supply chains are only now starting to digest what happened three or four months ago.

I don’t think Trump can turn the inflation narrative around much. People see and feel it at the gas station—Trump has no magic trick to make inflation disappear—it’s there, every time you fill up. So what other issues could shake up the entire US political spectrum? The answer is AI data centers—regulation, taxes, all of that.

I think the Democrats are finding a great campaign message: no more data centers, taxing AI giants, regulating AI. Because not only the poor will lose jobs, but the rich’s jobs will also be replaced by AI—at least that’s what people fear.

Arthur Hayes: If the opposition party can exploit this fear, they hold two powerful pieces of information: one, that the war Trump is waging causes vicious inflation; and two, that there’s a wave of AI construction backed by Republican politicians. So my theory is, if Trump wants to pull a rabbit out of his hat, the only issue he can flip on is AI.

Taking the Democratic side, he might say: “We need to crack down on data centers, establish a national AI dividend, tax them.” That’s Trump-style rhetoric. He can say a lot of things, but whether he acts after November is another matter.

I think this is their only chance—portraying themselves as the party protecting Americans from AI harm, and then Americans will forget that it’s actually the Republicans who financed all this, because people are forgetful. So I see this as the main risk.

And Trump’s willingness to attack AI depends purely on oil prices, which are a reflexive relationship with the Iran Revolutionary Guard. The longer this war drags on without resolution, the more we accumulate future pressures on commodities that could push prices higher. Trump is more likely to use AI as a tool to try to win the election, at least to help Republicans hold the House.

Obviously, the most destructive aspect of the AI narrative is taxation and regulation. We’ve already seen in Korea—when a politician announced a national AI tax, Cosby’s stock hit the limit down that day.

So I think if this rhetoric starts being promoted by the ruling party, especially openly by Trump, we’ll see the AI bubble peak—at least in the coming months until the election—and that will drag the crypto market down with it. That’s the core argument. I really don’t want to think about this anymore, so last week I completely liquidated my entire portfolio.

Host Kyle Chasse: What are your current most liquid assets—cash or government bonds?

Arthur Hayes: Government bonds and energy stocks.

Host Kyle Chasse: Do you still think energy can hold up if the AI bubble bursts?

Arthur Hayes: We still need oil, and that’s regardless of whether you like it or not. People need oil; it drives civilization. And I’m not saying AI won’t keep growing—problem is, our willingness to pay a forward multiple for that growth will decline, so these assets’ prices will fall.

That doesn’t mean these companies won’t be profitable—they probably will be—but we originally thought they’d be even more profitable, and now they’re not, so we sell those stocks. That’s the logic.

Arthur Hayes: I trade based on gut feeling and intuition, not analysis. I feel we’re in some phase of the AI bubble—I’m not sure which. I listened to Marco Papovich’s podcast over the weekend; he’s a strategist at BCA, with a pretty good YouTube channel called Geopolitical Cousins. I highly recommend subscribing.

He’s expressed many views in the podcast and articles—he makes an important point: when you invest in AI, you’re not investing in profits, but in capital expenditure for data centers.

SpaceX IPO: a time bomb with 100x P/S ratio

I often forget this: you’re investing in the second derivative—trend acceleration or deceleration. If the trend accelerates, you’re willing to pay an infinite multiple of future revenue; if it decelerates, you’re not. And this thing won’t grow as fast as you need.

He recently posted a chart showing the second derivative of capital expenditure growth—higher numbers mean harder acceleration. We’re already at $800 billion in 2026, and he predicts the second derivative of AI capital expenditure will start decelerating from 2027. You can’t pay 100x sales for SpaceX or any AI company when both profits and capital expenditure are decelerating.

Even if these companies’ revenues are still growing, that’s not the point—what matters is how fast they grow, how much the rate of change is, and your perception of that change. Mathematically, we know that, based on the law of large numbers, in the near future, the growth rate of capital expenditure can’t be as fast as from 2023 to 2026—that’s physically impossible.

So, when will the market discount this future and say, “I’m no longer willing to pay 50, 60, 70 times earnings for these AI stocks or supply chain companies”? When will the market realize that opposition parties worldwide are exploiting this mood—“screw data center inflation, screw AI replacing my job”?

Why are only Elon, Sam Altman, Zuckerberg, and about fifteen others becoming trillionaires, privatizing all human knowledge? And what about my share?

This isn’t unique to the US; the whole world is asking the same question: if AI is trained on human interaction data, and they legally or illegally use all this public and private data, why can they enjoy all the profits? For those with enough assets involved in these equity stories, it’s a legitimate question.

Eventually, the market will feel a backlash. There’s always conflict between capital and labor—voluntary or forced—and at some point, some agreement will be reached. If you hold those assets when that happens, you’ll probably get crushed. These thoughts keep circling in my mind. Then I sat down, tried to figure out what exactly happened, and spent a morning liquidating.

Host Kyle Chasse: How do you see the market moving from now until the end of the year?

Arthur Hayes: To answer that, I’ve been pondering another question: why hasn’t Bitcoin risen higher since November 2022? I’ve repeatedly said on your show and elsewhere: it all depends on liquidity. If future liquidity increases, Bitcoin should go up.

But now it seems that’s wrong. Because if we count from the commercialization of ChatGPT on November 30, 2022, until now, Bitcoin has indeed risen, but AI stocks like Nvidia have risen much more. When did Bitcoin peak? Last October, at $125,000. So all the liquidity created during this period—my model estimates it’s in the trillions of dollars—why hasn’t Bitcoin reached $500,000 or $1 million? Why has it underperformed AI?

I usually don’t look at where the money flows—just say “more money, so Bitcoin should rise”—that’s a lazy way of thinking. It’s worked before, but not this time. So I went back, re-examined my model, asked myself: what am I missing? The answer is, we all believe AI might be one of the most transformative technologies in history, undergoing massive capital expenditure—trillions of dollars.

But how much debt has AI swallowed during this period? Has AI actually squeezed out all other risk assets, pre-absorbing excess liquidity?

On a macro level, I usually don’t use M2 because I think it’s too coarse and not detailed enough, but let’s use it as an example.

Since ChatGPT, US M2 has increased by at least $1.5 trillion. I also asked reliable sources at Perplexity AI: how much debt has been issued to AI and related companies? An estimate is about $1.5 trillion, with $1.3 trillion issued between 2025 and 2026.

In other words, although the AI frenzy ignited at the end of 2022, the debt pump in capital markets is actually heavily skewed toward the later stage, only really ramping up recently.

My theory is, Bitcoin’s rebound from lows is due to the creation of substantial liquidity, and AI has not significantly drained it before 2025. Bitcoin has room to benefit from this wave of liquidity.

From 2022 to mid-2025, factors like reverse repos declining and others favor it. But if you look at the capital expenditure and loan charts of AI companies, the real volume picks up in 2025, especially 2026. And that’s exactly when we see Bitcoin struggling—peaking last October, now down 50-60%. So if all liquidity is flowing into AI, and that trend shows no signs of stopping.

If an AI bubble correction or burst occurs, investors won’t suddenly pour a huge amount of money into Bitcoin—they’ll sell AI, and also sell Bitcoin, sell everything. When the bubble bursts, all assets correlate at 1. Everything falls together until the dust settles, then some assets start to outperform.

Therefore, if I believe that in the next six months, due to rising oil prices and US political factors, the AI complex will undergo a major correction, Bitcoin can’t escape unscathed.

AI bubble: liquidity draining crypto markets

It should perform better after correction, but you must first go through that decline. That’s why I currently see no environment very favorable for Bitcoin and other cryptos.

And clearly, my positions in NEAR, HYPE, Worldcoin, and Zcash have been very profitable—I sold at profit. I want to pocket the gains from these trades and sit on the sidelines. These assets might continue to rise, but at least in my model, I feel uncomfortable with the risks at this point—those foreseeable unknown risks and how they might evolve—that’s why I exited.

Host Kyle Chasse: There’s another thing I’ve been thinking about—the S&P 500 is rising, but most stocks are actually falling; the index is being dragged by a few tech giants. More importantly, we’re about to see new IPOs from OpenAI, Anthropic, and SpaceX, which could add over $4 trillion in new market cap. Do you think these will drain liquidity from everyone for a while? How do you see these IPOs unfolding?

Arthur Hayes: I think these are hard to perform well because market expectations aren’t that they just trade normally—they expect these IPOs to surge 50%, have crazy gains, to show the market still believes in AI, that they picked the right star company, and that it will keep soaring.

SpaceX’s valuation around $1.8 trillion will make it the seventh-largest company globally. To go beyond that, it would need to surpass Amazon. If you read its S-1, you’ll see SpaceX’s valuation is close to 100x P/S. That’s utterly absurd—it will be the seventh-largest company in the world, yet it has proved nothing.

Yes, it’s a great idea—space data centers, policy issues around ground data centers, I agree that logic makes sense. I follow a Substack called Semi Analysis, which does deep research on semiconductors and AI.

They wrote an article comparing the costs of space vs. ground data centers, concluding that operating data centers in space costs four times more than on land. Not only that, but there aren’t enough chips to realize Elon’s vision, and building data centers on land is still feasible. It’s not as easy as you think, but it’s four times cheaper, so you’ll choose land until you simply can’t anymore.

According to the most optimistic estimates, space data centers won’t truly match land-based costs until at least ten years from now.

So the current reality is: the entire internet capital is willing to pay 100x sales for a product that’s four times more expensive, with rockets that might explode, and won’t make real money for at least ten years. That’s a classic crypto scam—low float, fully diluted garbage coins. 4-5% circulating supply, rising to nearly 25% by September, insiders selling from July to October, and it’s the seventh-largest company in the world, yet it’s achieved nothing in this data center argument. Except for satellite internet and similar businesses, which are doing well, but that’s not why you buy SpaceX.

So I think this is unlikely to meet market expectations. I’m not saying it will fall, but even if it rises 10%, the market reaction will be “that’s not enough, I want 50%, 60%, 70% gains.” That will make investors question: as insiders sell more and more, should I still buy into Anthropic or OpenAI’s IPO in September?

Pricing it so high creates an almost impossible-to-beat scenario. If it’s a $100 billion company, it can double or triple, and people will say AI is still valid, SpaceX’s surge is because they chose a lower valuation IPO.

But now it’s about maximum extraction—$1.8 trillion. To perform better than Nvidia? Very, very hard. Sorry Elon, you can’t beat Jensen. I don’t even know who the current CEO of Amazon is, but you can’t beat these companies. They have real revenue, are operational, and have proven their claims. SpaceX’s current valuation is still mostly on paper.

Maybe time will prove it, but would you really push a $1.8 trillion stock up another 50%? That’s extremely difficult. That’s why I think this will be a major event that severely shakes people’s faith in the AI narrative—simply because of its size, it’s almost impossible to go up.

Host Kyle Chasse: So you think these IPOs will cause liquidity to flow out? Will there be a massive capital shift? Or will it be like Elon—whoever sells first, sells best?

Arthur Hayes: It’s the first. People will get excited, pulling liquidity from other assets.

If SpaceX underperforms, I think Anthropic and OpenAI will face huge pressure to lower their IPO prices. If these two AI giants are forced to cut valuations or reduce fundraising before listing, it sets a very dangerous precedent—like the market is cutting its own throat, softly admitting that the AI bubble was too big, and lowering expectations proactively.

Suddenly, expectations fall—people hesitate: why did they lower prices after SpaceX? Why reduce issuance size? All these changes could cool investor enthusiasm. So capital might be pulled from other parts of the market, or everyone might just cool down on the AI bull story and gradually exit, triggering a chain of price declines.

Host Kyle Chasse: Going back to Trump’s anti-AI narrative—some might argue that all those AI giants are actually helping him get elected, or at least influencing him. We know he’s had many private dinners and discussions with them—they’re his big donors and supporters, and he’s publicly supported and praised AI.

Rate hike expectations: pivot and collapse

I haven’t counted how many people publicly oppose AI, but I know most people don’t have warm, fuzzy feelings about it, so this could be a smart pivot. But I’ve never heard anyone say it explicitly—that’s a bold prediction. How confident are you? Any signs that make you believe he might go this route?

Arthur Hayes: I still use Perplexity AI—I asked it: Polymarket says Republicans will lose; is there a path to victory? People must understand the underlying political logic: why does Trump desperately want to keep the House in midterms?

It’s not about some noble ideology; it’s purely for political self-preservation. If the House is taken by Democrats, he and his family will be served mountains of congressional subpoenas every day for two years. Democrats can keep chasing him, harassing him. He’ll have no chance to create any real legacy of a “second Trump term.” That’s why he wants to win.

I also believe Trump has no ideological stance—he only cares about winning. During the 2020 pandemic, he provided the largest fiscal transfer to Americans since the New Deal—sending checks, everyone got one. No income threshold, lots of fraud, rich and poor alike received checks. So I don’t think he’ll turn to naked populism—catering directly to public opinion—and public opinion is anti-AI.

AI has triggered negative emotions among both Republican and Democratic voters. So I asked AI: assuming these seat projections are correct, ignoring those that are basically safe due to redistricting, they still need to win some seats to keep the House.

Then I asked: in all competitive districts within the margin of error, are there any local legislations banning or restricting data center construction? Search all these districts. The result: if Trump turns against AI, he could flip enough seats to give Republicans the House, because these districts already show local residents don’t want data centers built in their neighborhoods, and some have already taken local action.

And again, all this is just rhetoric—Trump doesn’t need to do anything. He can call Jensen and those AI giants: “Listen, I’ll crack down hard on you in the next four months, don’t panic. Come November, none of this will happen.” That’s how he operates.

He attacks them, stock prices fall, some people lose money. Look at his tariffs—his hedge fund buddies lost billions when he tried to rewrite US trade infrastructure substantively. He finally pulled back at a critical moment, but at least he proved he’s willing to try.

So I don’t see why, if his political strategists see that an anti-AI stance can bring enough votes—even just as rhetoric—they wouldn’t do it? The only victims are the stock market, and only wealthy people lose money there. You don’t even have to do anything—just talk. No bills will pass.

After November, everything goes back to “we must win the AI race and beat China.”

So I think, given the inflation narrative is already set in stone and unchangeable, this is a viable path for the Republicans to win. I don’t care if oil prices fall another 50%; gasoline might drop a bit, but supply chain issues are already transmitting higher prices on supermarket shelves by October, and Trump is almost powerless to stop it.

Host Kyle Chasse: Let’s talk about Waller. I know there’s not much certainty yet, since his first FOMC meeting after taking office is next week. Based on his previous statements and the upcoming midterms, what do you think his policy stance will be?

Arthur Hayes: I don’t remember the specifics of his recent speeches, but one narrative is: you can see through the commodity inflation during wartime, then believe that AI productivity miracles will bring growth without inflation, so you can cut rates. I think that’s the narrative the market wants to believe about Waller.

The unfortunate reality is that oil prices are higher and won’t come down soon; the two-year Treasury yield is currently about 60 basis points above the effective federal funds rate. The market is telling the Fed: you need to hike. That’s the signal the market is sending—whether they do it or not, I don’t know.

I also think Trump might privately ease his obsession with rate cuts, because if he wants to do something about living costs, the last thing he should do is have the Fed start cutting when inflation is at 3.5-4%. If he really cares about affordability, cutting rates would lead to a disastrous midterm defeat.

So I think, considering market positioning, Waller will find it very difficult to cut rates. The baseline is to stay put, and it all depends on the language—whether hawkish or dovish. If hawkish, it signals inflation pressures are building, and the Fed might need to act; markets will price that in: “They will hike at some point.”

And the biggest risk to the bubble is rising rates—higher capital costs always push people away from the market in some way.

So I think the chance of rate cuts is very low; most likely, they’ll stay put, and it all depends on the wording.

The Fed’s limited options to support the bubble are evident—oil has already pushed the two-year yield above the effective federal funds rate; soaring oil prices and widening spreads have driven yields higher. I see no room for Waller to cut rates now. If the expectation of rate cuts is a pillar supporting your optimism about the AI bubble and its continuation, you should question that assumption.

Investment shift: energy stocks outperform Bitcoin

Host Kyle Chasse: So from now until midterms, do you think there will be any events that could give the market a short-term relief rally? Not manipulation, but real narratives or ongoing events that might bring some bounce before year-end?

Arthur Hayes: Maybe people believe MicroStrategy will keep buying, which could reignite some bullish sentiment, but I see no signs of money printing, and even if there is, it’s directly flowing into AI construction. So I see no major positive catalyst to pull crypto out of this slump, or at least to outperform AI.

Because if we go back to a high-growth, low-inflation perfect economy—what would you buy? Nvidia or Bitcoin? Of course, you’d pick Nvidia, Samsung, right? Because they’ve gone up 50x in two years. Would you buy Bitcoin? Of course not. That’s the problem—AI has performed too well.

If the environment remains the same, and these assets keep doing well, why would you choose crypto? You’d just keep betting on capital expenditure growing at 100% annually, continuously buying these companies. Do you think that’s sustainable?

And that’s exactly the current market belief.

But if I were an institutional investor, and my client said, “The Nasdaq’s up 50%, why only 10% for you?”—“Because I’m hedging, I bought volatility or something.” The client would say, “Why give my money to a fund that only rose 10%, instead of one that went up 50?” That’s the logic of everyone trying to maximize returns—everyone’s asking, “Why aren’t you participating?” That’s the problem.

Host Kyle Chasse: When will you consider re-entering? What could convince you to come back?

Arthur Hayes: If by fall, oil prices are moderate, no big surge, and Trump doesn’t turn against the AI giants, I might re-enter the market, see where value lies.

But all that hinges on a very strict precondition—the next few months, SpaceX, Anthropic, and OpenAI’s epic IPOs must open strongly, even blow out the biggest, most explosive IPOs in history, to match the scale of these offerings. When reality doesn’t match expectations, we’re in trouble.

Host Kyle Chasse: Is there a way to tell when the crypto market might enter the next bull phase?

Arthur Hayes: We need to see more money printing, and the money printed shouldn’t all flow into AI. When will that happen? I don’t know, but I don’t think it’s happening now. You’ve often said that the only way the government gets itself into trouble is by printing money—that’s inevitable.

Is there a way to estimate this timeline, or a catalyst? If the AI bubble really bursts, some financial institutions fail, then you’ll get a bailout.

When? I don’t know. But that’s the moment crypto can outperform—when AI is credit-broken, not that it ceases to exist, but it stops soaring like before, and investors need to trade other assets. I hope that other asset is crypto, and liquidity will flow back into crypto.

I firmly believe the answer is always money printing; the only question is timing. Bitcoin has been the best-performing asset in the past 15 years. But unfortunately, many people didn’t buy at $0—they bought at higher prices. If you entered during the ETF era, on average, you’re at a loss. Everything depends on the path and timing of entry. Just because you entered six months ago doesn’t mean Bitcoin should go up for you. That’s a painful lesson many need to learn.

Host Kyle Chasse: Final quick questions. First, will Bitcoin be above or below $100,000 by year-end?

Arthur Hayes: Below.

Host Kyle Chasse: When will the altcoin season arrive?

Arthur Hayes: We just went through an altcoin season—just four assets. People made a lot of money on HYPE and some others, so I think it’s over, maybe it will come again, but I don’t know.

Host Kyle Chasse: Do you think you’ll buy HYPE back before year-end?

Arthur Hayes: Yes.

Host Kyle Chasse: If you put $1 million into assets today—Bitcoin, HYPE, short-term bonds, gold—what would you choose?

Arthur Hayes: ExxonMobil.