Key Notes
- Arthur Hayes warns that potential selling by AI-company insiders could weaken share prices and make infrastructure financing more expensive.
- His share-supply estimates are scenarios, and unlocking existing stock differs from issuing new equity that dilutes ownership.
- Latest Fed data show lower weekly bank assets alongside rising credit and loans, leaving the proposed link to an AI credit contraction unproven.
Arthur Hayes warned that a wave of share sales at SpaceX, Anthropic and OpenAI could put pressure on AI stocks and spread into the debt markets financing the industry’s expansion, with consequences for banks and cryptocurrencies.
In his November 5 essay, D is for Dilution, the BitMEX co-founder argues that employees and early investors seeking to sell shares could overwhelm demand just as AI businesses need more capital for computing infrastructure.
The warning rests on a chain of events: weaker equities undermine lenders’ confidence, borrowing becomes more expensive, and companies struggle to finance construction. Hayes sees a potential financial crisis at the end of that chain. His figures and forecasts describe a bearish scenario, rather than announced insider sales or a demonstrated funding freeze.
The Share Supply Behind the Warning
Hayes estimates that SpaceX’s public float has increased by about 50% since its IPO and that another 8.6 billion shares could unlock by June 2027. The company trades on Nasdaq as SPCX.
Its IPO documents do establish staggered selling restrictions, including a 366-day lockup for Elon Musk and separate release schedules for other holders. An expiration makes stock eligible for sale under the applicable terms; it does not require the owner to sell.
For Anthropic, Hayes projects what he calls 83% dilution by June 2027. Applying that percentage to a hypothetical $1.5 trillion IPO valuation, he arrives at roughly $1.245 trillion of buying needed to absorb all the potential selling at unchanged prices.
That calculation assumes the additional supply reaches the market. A share unlock expands the pool available to trade, while issuing new shares can reduce existing holders’ ownership percentage. The two can create selling pressure through different mechanisms, and neither establishes a fixed cash requirement for the issuer.
Anthropic’s amended prospectus still leaves the IPO price range and offering share count blank. ANT remains its proposed Nasdaq symbol. Prospective investors have pressed for a valuation closer to $1.5 trillion, as we previously reported, but the offering has not been priced.
Why the Debt Market Matters
Hayes’s concern extends to the cost of financing AI infrastructure. A falling stock price can lead lenders to question a borrower’s outlook and demand higher yields, tighter covenants or more cash held in reserve. Those demands would make the next round of investment harder to fund.
The scale of the spending is substantial. Anthropic has disclosed approximately $895.4 billion in aggregate purchase and lease commitments, largely for computing and infrastructure, with spending extending into 2045. About $69.8 billion falls due in the 12 months after September 30, 2026. Some arrangements have termination rights or conditions, so the aggregate is not an immediately payable debt balance.
OpenAI faces another test of investor appetite. The company reportedly reduced its funding target to $20 billion at a $1.25 trillion valuation, as we reported. Hayes expects similar selling pressure after an eventual IPO, although the essay’s assumed offering timetable is his own scenario.
A possible combination of SpaceX and Tesla (NASDAQ: TSLA) adds another variable. The companies have considered an all-stock transaction, as we reported. Without an agreed exchange ratio or final structure, the ownership effect cannot be calculated from the merger idea alone.
Bank Data Give a Mixed Signal
Hayes points to a weekly decline in U.S. banking assets as an early warning, while acknowledging that one observation does not establish a trend. He argues that weaker AI lending and securities financing could eventually restrain bank balance sheets and dollar liquidity.
The Federal Reserve’s October 30 H.8 release provides a useful distinction. Seasonally adjusted total assets at commercial banks fell by approximately $45.9 billion in the week ended October 21, to $26.08 trillion.
Bank credit increased by $28.8 billion over the same week, however, and loans and leases rose by $13.2 billion. A fall in total assets therefore did not coincide with an aggregate decline in lending in those figures. The release also does not isolate banks’ exposure to AI businesses.
For Hayes’s argument to become a broader credit story, investors would need to see persistent financing pressure: more expensive debt, reduced access to loans or bond issuance, and investment plans being curtailed. The weekly banking figures alone do not establish that sequence.
Bitcoin Faces Two Stages in Hayes’s Scenario
Hayes expects tighter dollar liquidity to hurt cryptocurrencies initially. Banks and investors pulling back from risk could reduce the money available for speculative assets even if the original losses occur in AI equities.
He then anticipates a reversal. In his view, a sufficiently severe crisis would force the Federal Reserve and other central banks to expand monetary support, eventually creating more favorable conditions for Bitcoin. Neither the intervention nor its timing is an announced policy response.
The distinction leaves an uncomfortable gap for investors: the liquidity squeeze he fears would arrive before the monetary relief he expects. For now, Hayes says Maelstrom will stay on the sidelines. The next AI offerings will test whether buyers are willing to absorb insider sales while also financing the infrastructure those companies have promised to build.
