Arthur Hayes says the AI buildout may not end with a clean stock-market air pocket, but with a credit event that leaves lenders and leaseholders holding the bag. If that happens, Bitcoin could eventually benefit from the liquidity response, after getting whacked in the first round like every other risk asset.
- Hayes frames AI infrastructure as a credit story, not just an overhyped equity trade.
- Big tech spending is still roaring, which weakens the case that the AI boom is already cracking.
- Bitcoin may suffer in a deleveraging shock first, then gain if central banks ease afterward.
In his essay Situationship, published on Aug. 4, 2026, the BitMEX cofounder argues that AI infrastructure looks less like software hype and more like leveraged real estate. The comparison is blunt, but the logic is simple: data centers, long-term leases, and hardware financed through debt-like structures can turn fragile fast if demand cools or the gear goes stale.
Hayes’s core line is that AI is a “credit story like 2008 and not an earnings story like 2000.” That distinction matters. The dot-com bust was mostly about companies failing to turn revenue growth into profit. A credit crisis is uglier. The damage spreads through financing, refinancing, and obligations that do not just vanish when the music stops.
Think of it this way. A stock bubble can deflate. A credit bubble can drag down lenders, lessors, private credit funds, and anyone else who thought the shiny new asset on the balance sheet would keep earning its keep. That is where Hayes’s warning gets teeth.
He is basically saying AI data centers act a lot like property projects. They are expensive, capital-intensive, and built on assumptions about continued usage, strong cash flow, and equipment that does not become yesterday’s junk the moment a faster chip shows up. If that sounds uncomfortably close to real estate finance, that is because it is.
But there is a real counterpoint here. The public numbers from the hyperscalers still show a live, expanding buildout, not a collapse. Alphabet, Microsoft, and Amazon are still spending heavily, and their cloud businesses are still growing. That does not kill Hayes’s thesis, but it does mean the AI boom is not visibly breaking yet.
Alphabet reported $44.9 billion of capital expenditure in the second quarter and raised its 2026 capital expenditure guidance to $195 billion to $205 billion, from $180 billion to $190 billion. It said about 60% of its technical infrastructure investment went toward servers, while 40% went toward data centers and networking equipment. Google Cloud revenue rose 82% year over year to $24.8 billion, operating income reached $8.8 billion, and backlog climbed to $514 billion.
Amazon’s AWS business added another useful data point. AWS revenue rose 37% to $42.2 billion in the second quarter, while operating income hit $16.6 billion. Amazon said its trailing twelve-month free cash flow moved to an outflow of $7.6 billion, mainly because property and equipment purchases increased by $66.1 billion.
That is not the profile of a sector running out of enthusiasm. It is the profile of a sector still shoveling money into infrastructure like there is no tomorrow. That may be rational if demand keeps up. It may also be the sort of thing people later call “obvious in hindsight.” Markets love that trick.
Where Hayes’s argument gets sharper is in the obligations hiding behind the spending. Alphabet disclosed $85.2 billion of future lease payments not yet started as of June 30. Those leases are scheduled to begin between 2026 and 2031, with contract terms up to 26 years. Alphabet also reported $811 billion of purchase commitments and other contractual obligations, along with $98.2 billion of long-term debt. It issued more than $51 billion of fixed-rate notes in the first half of 2026.
Microsoft disclosed $62.9 billion of finance lease liabilities as of March 31, plus another $196.6 billion of leases, mainly for data centers, that had not yet commenced. Meta reported about $182.88 billion of uncommenced lease obligations and $237.67 billion of noncancelable contractual commitments as of March 31, then added another $24 billion of infrastructure contracts in April.
This is the part that makes the “credit story” framing more than a spicy metaphor. A lot of AI infrastructure is not sitting neatly on a single balance sheet as a simple capex line. It is spread across leases, purchase commitments, joint ventures, and private financing structures. That can work beautifully while demand is strong. It gets ugly when utilization slips, refinancing tightens, or the hardware you just financed starts looking slow and expensive compared with the next generation.
Meta’s financing structures show how creative this can get. Meta and BlackRock announced a venture for a one gigawatt campus in El Paso, Texas, which they said represents more than $10 billion of investment. That followed an earlier Meta venture with Blue Owl Capital for an estimated $27 billion data center campus in Louisiana. Blue Owl funds received an 80% interest, while Meta retained 20%, and part of the outside funding came through debt sold privately to PIMCO and other bond investors.
That is not the old-school “we built a server room and paid cash” model. It is layered finance. It shifts risk around, which is exactly why people like it until the bill comes due. When leverage is buried in leases and private debt, the stress can stay invisible longer than it should.
Hayes’s comparison to 2008 is therefore less about saying AI is doomed and more about saying the financing stack matters more than the pitch deck. A real-estate-style asset financed with debt and long obligations can become a problem even if the underlying technology is useful. And AI, for all the ridiculous valuations and breathless PowerPoints, is clearly useful.
There is also a good reason the worst-case may not happen. These hyperscalers are not fragile little startups praying for the next funding round. They have massive cash generation, diversified businesses, and deep access to capital. If AI demand keeps translating into cloud revenue and enterprise usage, the financing burden may stay manageable. Big spending alone is not a death sentence. Sometimes it is just big spending.
Hayes, though, is looking further ahead. He argues the slowdown in AI capital spending may start in the second half of 2027 and become clearer in 2028. That is a forecast, not evidence. For now, the better-supported reading is that the buildout is still running hot. But hot cycles have a habit of convincing people that debt, maturity, and asset quality suddenly do not matter. They still do.
The Bitcoin angle is where this gets especially interesting. Hayes sees BTC as a possible beneficiary if an AI slowdown eventually pushes policymakers toward easier money and more liquidity. In plain English: if the system gets stressed and central banks respond by loosening conditions, Bitcoin could benefit later.
But there is a catch, and it is a big one. In the first wave of a credit shock, Bitcoin may get sold along with everything else people can liquidate quickly. Investors do not always dump the asset they dislike most. They dump the asset they can sell fastest. That means BTC can fall hard before any later policy response kicks in.
That part tends to get lost in the usual “crisis equals bullish Bitcoin” cheerleading. Reality is uglier. First comes forced selling. Then comes the central bank reaction. Then, maybe, Bitcoin catches the liquidity wave. The market rarely hands out the pretty version on the first try.
The Federal Reserve backdrop matters here too. On July 29, the Fed held its federal funds target range at 3.5% to 3.75% by a 9 to 3 vote. It did not roll out an AI rescue program, emergency lending facility, or fresh asset purchase plan. Its July monetary policy report also said Treasury bill purchases since early January totaled nearly $250 billion, including about $160 billion of reserve management purchases. The next scheduled meeting is set for Sept. 15 and Sept. 16.
That leaves the market in an awkward middle ground. Rates are not plunging. The Fed is not stepping in to backstop AI-related excess. And the biggest tech firms are still pouring money into servers, chips, leases, and data centers as if the payback window will stay generous forever.
Hayes’s longer-term Bitcoin view stays aggressively bullish, with a possible trading range between $60, 000 and $70, 000, downside near $50, 000, and an eventual move toward $1 million. That is classic Hayes: bold, memorable, and impossible to ignore.
The cleaner takeaway is more restrained. If the AI boom turns into a financing mess, Bitcoin may ultimately benefit from the liquidity that follows. But BTC is not immune to the crash itself. In a real deleveraging event, it can get hit first and asked questions later.
Key takeaways and quick answers
-
Is the AI bubble already bursting?
Not from the spending and cloud numbers cited here. Alphabet, Microsoft, and Amazon are still spending heavily, and revenue growth remains strong. -
Why does Hayes call this a credit story?
Because the risk, in his view, is not just overvalued stocks. It is debt, leases, and financing structures that could crack if cash flow or refinancing conditions weaken. -
Could Bitcoin fall before it rises?
Yes. In a credit shock, investors often sell liquid assets first to raise cash, and Bitcoin can get caught in that wave before any later liquidity support helps it. -
What weakens the bearish AI case?
Strong cloud revenue, growing backlog, profitable infrastructure use, and big cash-rich companies that can keep funding the buildout without a financing blowup. -
Why does this matter for Bitcoin holders?
Because BTC is often treated as a macro liquidity trade. If AI stress leads to easing, Bitcoin could benefit later, but only after surviving the initial scramble for cash.
Further reading
A few useful side roads if you want to track the macro plumbing, the AI capex binge, and Hayes’s broader Bitcoin takes.
