Arthur Hayes: Stocks Peak in 2027 Before the AI Bubble Bursts

Anthropic

Across the shows: 60 speakers in other episodes, 48 bullish and 41 bearish calls, past 30 days

• Arthur Hayes · Flop LabsBearish

  Anthropic will likely struggle to pay non-cancelable compute contracts, a key AI bust signpost.

  Hayes names Anthropic or OpenAI missing a payment, or signalling trouble paying for the data compute deals they have signed, as the milestone that will mark the AI bubble cracking, around late 2027. He says these deals are non-cancelable contracts for compute the labs say they want to buy, and that the data centers were financed by others. He argues Anthropic and OpenAI do not make money and are going public to access equity markets. He expects them to issue junk-rated corporate debt to fund compute purchases. If revenue falls short and rates stay high, they cannot finance themselves, and that failure would show the biggest AI compute buyers cannot afford the build-out.

  “Anthropic or OpenAI misses a payment or signals are going to have trouble paying these data compute deals that they've signed. So they have these non cancelable contracts for all this compute that they say they want to buy.”at 2:16

• Arthur Hayes · Flop LabsBearish

  Anthropic will face a cash-funding crunch for compute purchases in late 2027, triggering an AI bust.

  Hayes argues that late 2027 is when it becomes put-up-or-shut-up time for Anthropic and OpenAI, because they must send cash to buy the compute they have committed to. He presents this as his working mental model of how an AI bust unfolds, rather than a certainty, and says he is not trading on it and could be very wrong on timing.

  “it's put up or shut up time for Anthropic and OpenAI in terms of having to send cash to buy this compute. That's my working mental model, but I'm not trading based on that.”at 37:14

OpenAI

Across the shows: 23 speakers in other episodes, 10 bullish and 17 bearish calls, past 30 days

• Arthur Hayes · Flop LabsBearish

  OpenAI will likely struggle to pay non-cancelable compute contracts, a key AI bust signpost.

  Hayes names Anthropic or OpenAI missing a payment, or signalling trouble paying for the data compute deals they have signed, as the milestone that will mark the AI bubble cracking, around late 2027. He says OpenAI has signed non-cancelable compute contracts it says it wants to buy. He argues OpenAI does not make money and is going public to access equity markets, and will likely need junk-rated debt to finance compute purchases. If revenue is insufficient and rates are high, it cannot finance itself, which he sees as the signal that the biggest AI compute buyers cannot pay for the build-out.

  “Anthropic or OpenAI misses a payment or signals are going to have trouble paying these data compute deals that they've signed. So they have these non cancelable contracts for all this compute that they say they want to buy.”at 2:16

• Arthur Hayes · Flop LabsBearish

  OpenAI will face a cash-funding crunch for compute purchases in late 2027, triggering an AI bust.

  Hayes argues that late 2027 is when it becomes put-up-or-shut-up time for Anthropic and OpenAI, because they must send cash to buy the compute they have committed to. He presents this as his working mental model of how an AI bust unfolds, rather than a certainty, and says he is not trading on it and could be very wrong on timing.

  “it's put up or shut up time for Anthropic and OpenAI in terms of having to send cash to buy this compute. That's my working mental model, but I'm not trading based on that.”at 37:14

Bitcoin

Across the shows: 10 speakers in other episodes, 6 bullish and 3 bearish calls, past 30 days

• Arthur Hayes · Flop LabsBullish

  Bitcoin is the most responsive asset to fiat debasement and will outperform in a bailout; Maelstrom is long.

  Hayes argues the best performing assets when a balance sheet is debased are the growing bank reserves or bank lending, the amount of credit in the system. He says Bitcoin has proven to be the most responsive asset to an increase in fiat dollars, yen, euros and yuan. He expects authorities to print money and bail out the AI credit build-out, and that Bitcoin will be the best performer when this happens. He says Maelstrom is long Bitcoin and will wait, since timing these events is hard. He says crypto peaked in October and is down 50% because AI buildup absorbed the credit that would have gone to crypto in 2025 and 2026.

  “And we've proven that over time. And so Bitcoin is the most responsive asset to an increase in the amount of fiat dollars, yen, euros, yuan in the world. And so me at Maelstrom, we're long Bitcoin.”at 13:48

• Arthur Hayes · Flop LabsBullish

  Bitcoin benefits from a coming government bailout after the AI bust; Hayes holds it unlevered.

  Hayes is long Bitcoin as a trade on a coming bailout, which he expects to come whether he is a year early or a year late. He holds the position with no leverage. He argues governments will not restrain themselves on Bitcoin's price because so few people own it, so the bailout response will not be shaped by Bitcoin's rise.

  “I'm just long with no leverage, ... crypto long the bailout. And I know the bailout's going to come.”at 37:31

CoreWeave

Across the shows: 51 speakers in other episodes, 12 bullish and 46 bearish calls, past 30 days

• Arthur Hayes · Flop LabsBearish

  CoreWeave's junk debt and weak cash flow are old news the market is only now pricing.

  Hayes says the market only recently began to focus on CoreWeave's debt stack, though nothing new happened. He says the company is issuing lots of junk debt and its cash flow situation is poor, and the market is only now recognizing this. CoreWeave stock fell roughly 40% over two weeks, and neoclouds have already shown credit stress. He does not think this means the AI story is over. He says NVIDIA would have to fall a lot for the theme to be over. He does not trade these names and would never short them.

  “Just like no one cared about CoreWeave's debt stack. It wasn't like two weeks ago, all of a sudden CoreWeave did something out of the ordinary. It's just people like, Oh yeah, these guys are issuing lots of junk debt and their cash flow situation is pretty shitty.”at 4:19

Pendle

• Arthur Hayes · Flop LabsBullish

  Cash-flow DeFi tokens like Pendle, which pay cash to holders, should see rising token prices.

  Hayes says Maelstrom is long some cash-flow generating projects such as Athena and Pendle, which are decentralized applications that generate actual cash flows flowing back to token holders. He believes crypto investors outside Bitcoin, Ethereum and the mega caps will look for these projects and reward them with a high and rising token price. He notes that a rising macro liquidity backdrop helps tokens but does not guarantee any particular altcoin rises, which depends on the investor's skill.

  “...We are along some cash flow generating projects like Athena, like Pendle, some of these decentralized applications that generate actual cash flows. Those cash flows flow back to token holders.”at 28:12

US stock market outlook

• Arthur Hayes · Flop Labs

  General equity markets will peak in the first half of 2027.

  Hayes expects general equity markets to peak sometime in the first half of 2027. He argues investors only care about revenue until a company misses a payment. He says that moment will bring a multitrillion-dollar bailout, after which the best performing assets will be crypto. The peak call is tied to the AI compute financing stress he expects to surface.

  “I think the general equity markets peak sometime in the first half of two thousand twenty seven. Everyone cares about revenue now until somebody misses a payment.”at 1:00

• Arthur Hayes · Flop Labs

  General equity markets will peak in the first half of 2027, then fall sharply in late 2027.

  Hayes's working model is that general equity markets peak in the first half of 2027. The downturn comes in late 2027, when the AI labs must fund large compute purchases with cash and that funding need forces a reckoning. He says he is not trading on this timing, is cognizant he could be very wrong on it, and so is not shorting anything, staying long with no leverage.

  “I think the general equity markets peak sometime in first half of twenty twenty seven, and then we get the, Oh, shit moment in late twenty twenty seven”at 36:57

US rates & the Fed

• Arthur Hayes · Flop Labs

  Authorities will print money and bail out AI debt, because letting AI lenders fail is politically unacceptable.

  Hayes argues every politician prefers to bail out rather than allow a recession. He says the authorities will print money and kick the can down the road, as in 2008 and 2020. He does not think the authorities will let Anthropic or OpenAI fail, because of the political fallout. He argues a systemic issue exists in the insurance and private-credit channel, which he says would make a multitrillion-dollar bailout inevitable. He cites the 2023 Bank Term Funding Program as precedent, when Bank of America was functionally insolvent and officials stepped in. He expects a minor wobble first, with officials acting in a way similar to 2023.

  “So print the monies, kick the can down the road to the next administration or whoever else runs your country, and that is the bailout.”at 7:28

• Arthur Hayes · Flop Labs

  Governments and central banks will likely print money and levitate financial assets rather than let markets reset.

  Hayes says the most likely outcome is that authorities print money and levitate financial assets, so he is positioned for that rather than a reset. He rejects the traditional 60/40 portfolio in this environment, saying 'sixty forty is bullshit' and that the choice is simply the best asset to hold while money is printed. He leaves only a small probability for a reset that does not look like a bailout.

  “knowing that the most likely outcome is they're going to print money and they're going to levitate financial assets. So we just pick the best one.”at 35:42

AI bubble & tech valuations

• Arthur Hayes · Flop Labs

  AI is a bubble: data center buildout is capital intensive with unclear returns on capex.

  Hayes says the AI data center buildout is extremely capital intensive and it is unclear whether it will generate real returns on that capex. He agrees with Ken Griffin that AI is in a bubble but says that does not mean shorting AI stocks. He expects the bubble to be recognized as cracked in late 2027. His evidence: a SemiAnalysis estimate of 50 GW of US data centers in 2027, roughly double 2026, which he says means a lot of compute must be bought. He notes the largest capex as a share of GDP since the railroads, which he calls a massive malinvestment. He argues the whole world is in the same trade, with US, China and Japan either building data centers or supplying commodities and intermediate goods to them. The crack trigger he watches is Anthropic or OpenAI failing to pay for non-cancelable compute contracts.

  “The data center build out is insanely capital intensive and unclear whether or not it's going to generate any real returns for all of that CapEx.”at 1:47

Private credit

• Arthur Hayes · Flop Labs

  PE firms use captive insurers to offload AI data-center debt, creating systemic risk akin to AIG.

  Hayes says a Nick Namath essay describes a scheme PE firms such as Apollo and KKR run with captive insurance companies. A lot of the so-called AAA and investment-grade debt tied to SPVs that finance data centers ends up on insurers' balance sheets. Policyholders' premiums are invested in this debt, and when it is re-rated to where it should trade, the PE sponsors have put in no equity because the strategy is to offload risk to captive insurers. He calls this systemic, comparing it to AIG, and says politicians cannot let policyholders take the loss, which makes a multitrillion-dollar bailout inevitable.

  “scam that PE firms like Apollo and KKR are running with these captive insurance companies. ... a lot of the so called AAA investment grade rated debt associated with these SPVs to finance these data centers ends up on insurance companies' balance sheets.”at 9:14

Gold & precious metals

• Arthur Hayes · Flop Labs

  Gold will rise on central bank buying and geopolitics, but by less than Bitcoin in a bailout.

  Hayes says he is long a lot of gold and gold miners. He expects gold to rise for reasons beyond the bailout, namely central bank buying and geopolitical issues. In a bailout of the type seen before, however, he thinks Bitcoin responds more cleanly and by more than gold. He frames both as ways to protect against currency debasement.

  “But I do think that Bitcoin responds more cleanly to a bailout of the type that we've seen in the past. Whereas gold will go up a lot. I guess I don't think it'll go as much as Bitcoin.”at 14:55

AI infrastructure & data centers

• Arthur Hayes · Flop Labs

  Data center overbuild will flood the market and drive compute prices sharply lower.

  Hayes says the world has already overbuilt compute. He cites analysts' expectation that the US will build and deliver double the amount of data centers, and says China is building faster because it is cheaper. He argues the price of compute will plummet from this oversupply. He sees cheap compute as the raw food for an agentic future, since agents need cheap compute to do more complex tasks and use more of them. He says financial markets are sending the wrong signal to build data centers, which will oversupply compute and enable the agent economy.

  “we've already overbuilt compute. As many analysts say, we're going to build double the amount of data centers, deliver double the amount of data centers, at least in The United States. ... So the price of compute is going to plummet.”at 16:23

Open-source AI models

• Arthur Hayes · Flop Labs

  Governments should run self-hosted open-weight models, not closed models from US labs.

  Hayes responds to a question on a US ban on Chinese models. He thinks the most a US ban would do is stop US departments using Chinese AI, forcing them onto Anthropic or OpenAI. He argues even the US government makes no sense in using a closed-source model. It should host its own bare-metal data centers and install an open-weight model. He says it makes no sense for the military to use a closed-source model, and asks whether the US government would trust Dario, Sam and Elon with its data.

  “it makes no sense to use a closed source model. What you should do is ... install your own open weight model. It makes no sense for military to use a closed source model.”at 23:21

Credit markets

• Arthur Hayes · Flop Labs

  AI-linked corporate debt is crowding out sovereign bond issuance, including Google's Australian dollar deal.

  Hayes argues the credit that central banks created is going into funding the AI buildup. Sovereign yields show sovereigns competing with AI debt. Hyperscalers do not care what interest rate they pay, because they expect their revenue to cover it. Governments other than the US cannot compete this way. He cites France and Australia as having trouble issuing debt in their own currency, and says Google's massive Australian dollar offering crowded out the Australian sovereign market. He frames this as a problem for governments, since investors would rather lend to build AI than to build bombs.

  “...the credit that was created by the banking system via central banks is going into funding AI buildup. And obviously we see that today with sovereign yields. Sovereigns are competing with AI debt, and the hyperscalers don't care what interest rate they pay.”at 24:18

US–China AI & chip race

• Arthur Hayes · Flop Labs

  Beijing sees AI as the silver bullet to outgrow debt and deflation, backing Chinese AI firms.

  Hayes says China is as all-in on AI as the US and is spending as much money on it. He says Xi Jinping has been vocal that AI and AI-adjacent industries are how China can outgrow its debt and deflation problems. Beijing backs DeepSeek, Moonshot's Kimi, Z.ai and similar firms. Billions of yuan are going into data centers, though fewer than in the US because building in China is much cheaper. He says the Politburo views advanced technology as its silver bullet to cut through deflation.

  “Xi Jinping has been very vocal about the fact that he believes AI and AI adjacent industries are the way in which China can outgrow its debt and deflation issues.”at 29:13

AI: overall outlook

• Arthur Hayes · Flop Labs

  AI is extremely useful, but today's frontier AI companies are overpriced and he would not invest in them.

  Hayes says AI is extremely useful for human civilizational advancement, comparing it to railroads, which are still useful even though railroad stocks peaked at around 50% of index market capitalization in the late nineteenth century and never returned to that level. He says the technology is valuable, but the price paid for it is wrong. He would not invest in the companies at the vanguard of AI today. He leaves open that another company created from the oversupply of compute could prove very useful and be a good investment.

  “I think it'd [be] extremely useful. The technology, just like we still use railroads today... Now, I just don't think we're paying the right price for it today. And I wouldn't invest in these companies that are at the vanguard of AI today.”at 30:53

Quantum computing

• Arthur Hayes · Flop Labs

  Q-Day threat to crypto is theoretical and less important to returns than other risks.

  Hayes says there is a lot of FUD about Q-Day, when quantum computers would break crypto, but there is much disagreement on whether it is a physics or engineering problem and many smart people are working on it. He says he sees a risk he cannot quantify and that there are far higher-probability threats to investment returns than this theoretical scenario, which he can worry about today.

  “I think there's way higher probability things that could destroy your investment returns than this theoretical Q Day.”at 32:30

US Treasuries & yields

• Arthur Hayes · Flop Labs

  Bonds at a 5% 10-year yield will underperform the coming debasement, so they are the consensus trap.

  Hayes says bonds are a great investment now that the ten-year yields 5%, and he is not saying investors will lose money. He argues the real question, if a reflationary event is coming, is whether an investor protects capital by making the most money possible. Holding bonds at 5% while Bitcoin rises 100% is underperformance against the debasement to come. He calls underperformance the biggest consensus trade because people think avoiding losses is enough.

  “bonds are a great investment now that the ten year's at 5%. ... It's not about not losing money. It's underperforming the debasement that's to come.”at 33:30