The debt burden of the AI sector is growing rapidly, and in some places the strain is already noticeable. However, there are no signs of a systemic credit crisis yet, and bitcoin reacts more to the stock market as a whole than to the debts of AI companies as such. In September, the volume of AI debt financing fell to $23 billion — almost half the August level and about 80% below the June peak.
It is precisely this shift from growth to contraction that warrants close attention. A decline in tech company valuations is not a crisis in itself: stocks get cheaper due to revised forecasts or doubts about returns on investment. A crisis begins later — when companies find it harder to service debt, refinancing becomes more expensive, and lenders cut limits.
Where AI sector debts are already faltering
Borrowed money in AI infrastructure has become abundant. The five largest cloud data center operators — Meta, Alphabet, Amazon, Microsoft and Oracle — accounted for more than 15% of new investment-grade dollar bond issuance since the start of the year by early May. For comparison, at the end of 2025 they accounted for only about 3% of outstanding debt in this segment. By early September, global AI borrowing since the start of the year had exceeded $450 billion — more than in all of 2025.
The market absorbed these issues calmly until summer. Most major players have low debt burdens and ratings no lower than AA-. The exception is Oracle. In July, S&P downgraded the company to BBB-, while Moody's rates it at Baa2 with a negative outlook. Oracle's debt stands at about $130 billion, and free cash flow for the fiscal year ended in May 2026 turned out negative — about minus $24 billion. Oracle's five-year CDS, i.e., the cost of default insurance on the company, rose in July to about 2 percentage points — the highest in almost 18 years.
Stress is also visible at the level of individual projects. About $18 billion in bank loans finance the construction of the Project Jupiter data center in New Mexico, which Oracle leases. Banks failed to sell these loans to investors, and in September they were quoted in private markets at 89–91 cents on the dollar. Pressure intensified after a force majeure notice Oracle sent to the developer due to power supply delays. If the parties recognize it, the company will be able to defer the start of full rental payments for up to three years.
The share of private credit is also increasingly noticeable. Private lending deals involving AI companies reached $59 billion in 2025 — almost seven times more than a year earlier. The Bank for International Settlements estimates outstanding private loans to companies in AI, big data and cloud technologies at more than $200 billion. The borrower is often not a tech giant but a data center owner or a project structure, which makes it harder to understand where exactly the risk is concentrated.
At the same time, the broad market remains calm so far. In the Fed's July survey, 50 of 56 banks left lending standards for large and medium-sized companies unchanged. The spread on US high-yield bonds rose from 2.60 percentage points at the end of August to 3.11 percentage points by October 12, with the main jump coming at the end of September. This is a signal of deteriorating conditions, but by historical standards the level remains low. Vulnerabilities may be accumulating where such indicators see less well — in private credit and project finance.
What this means for bitcoin
Bitcoin's link to AI stocks has been weak and unstable this year. Since the start of the year, chipmakers have risen 55%, while bitcoin has lost about 8%. The 30-day correlation of daily returns ranged from +0.04 to +0.54. When the seven largest US tech companies fell 4.8% on July 23, bitcoin declined by less than 1%. And in June, money flowed out of bitcoin ETFs amid a rally in AI stocks.
The opposite also happens. On October 14, after data showed US inflation accelerating to 3.7%, heavily indebted AI companies lost noticeably more than the market: the Nasdaq fell 1.9%, Oracle — 8.8%, CoreWeave — 15.5%, and CoreWeave bonds traded below 80 cents on the dollar. Bitcoin lost 1.8% in a day — almost as much as the Nasdaq.
Academic data confirms that cryptocurrencies react to credit risk asymmetrically. When spreads widen, they depreciate noticeably, but when spreads narrow, they do not recover. However, bitcoin, unlike Ethereum, Solana and Cardano, reacts to credit stress mainly through the stock market rather than directly.
A separate risk channel is miners. Many of them are converting their sites to AI workloads, since renting out computing power yields about $1.5 million in annual profit per megawatt versus about $0.5 million for mining. Contracts worth more than $100 billion have been signed, but revenue is still only about $1.1 billion a year. Cipher Digital sold 1,619 bitcoins for $123.4 million in the first half of the year at a loss of $47.7 million, and its interest expenses in the second quarter were 2.7 times its mining revenue. If construction drags on, selling bitcoin could become a way for several large miners to raise cash.
We can talk about a systemic impact of AI sector debts on bitcoin if three signs converge: the high-yield spread holds above the October 1 peak; AI companies lose access to refinancing; and bitcoin starts falling noticeably more than the Nasdaq on days when spreads widen. So far, there is not enough data to say the strain has become systemic.
Cryptalist analysis: The debt expansion of the AI sector is forming a new macro risk factor that is not yet transmitted directly to bitcoin but could amplify volatility through the stock market and miners' balance sheets. The key indicator for institutional investors is the dynamics of credit spreads and AI companies' access to refinancing: their deterioration will be an early signal of risk repricing in crypto assets as well.