The AI spending boom has become two things at once. It is the trade investors are most sure about, and the place they think a credit accident is most likely to start.

In Bank of America's September Global Fund Manager Survey, 53% named buying global semiconductor stocks the most crowded trade in the market.

A crowded trade is one nearly everyone already owns, which is what makes it fragile when sentiment turns.

In the same survey, 42% named AI infrastructure spending the most likely source of a systemic credit event, up from 38% in August.

Government debt ranked second at 25% and private credit third at 19%.

Both answers describe the same money.

AI Spending Still Has Believers

BofA Global Research said in the survey that "investor conviction on macro boom and fast-paced AI capex is still strong."

The survey ran Sept. 4-10 among 190 managers overseeing $512 billion.

A growing 79% of respondents expect no AI infrastructure spender to announce a spending cut this year, up from 71% in August.

For equipment suppliers, that is a demand forecast.

Nvidia Corp. (NVDA  ) shows why. Its data center revenue reached $89 billion in the second quarter of fiscal 2027, up 117% from a year earlier.

The company also forecast $108 billion in total revenue for the following quarter.

The five largest spenders, Amazon.com Inc. (AMZN  ), Alphabet Inc. (GOOGL  ) (GOOG  ), Meta Platforms Inc. (META  ), Microsoft Corp. (MSFT  ) and Oracle Corp. (ORCL  ), are on track for roughly $700 billion of capital expenditures in 2026.

Analysts at Bank of America and Evercore project the figure passes $1 trillion in 2027.

The Supplier's Revenue Is The Customer's Bill

Nvidia's sales are its customers' outlays. Strong chip demand proves the equipment is wanted. It does not prove each buyer earns a return on it.

Oracle is where that gap is most visible.

In its results released earlier this month, the company reported $28.5 billion of capital spending in a single quarter and negative free cash flow of $5.4 billion, partly covered by a $20 billion at-the-market equity sale.

Full-year capex guidance is $90 billion to $95 billion, and management gave no date for a return to positive free cash flow.

A record net 33% of survey respondents said companies are overinvesting.

"AI hyperscaler capex continues to be seen as the most likely source of a systemic credit event (per 42%)," said analyst Michael Hartnett.

That 42% reading increased from 38% in August, exceeding government debt at 25% and private credit at 19%.

The Credit Market Is Already Separating The Two

On Wednesday, Apollo Chief Economist Torsten Slok added evidence from credit markets: investors are paying more to protect against hyperscaler defaults while bank protection costs remain stable.

Five-year CDS on a basket of Amazon, Google, Microsoft and Oracle has climbed above 100 basis points, the highest in Slok's eight-year series.

The equivalent contracts on JPMorgan Chase & Co., Citigroup Inc., Bank of America Corp., Wells Fargo & Co. and U.S. Bancorp have sat flat near 40 basis points.

A credit default swap, or CDS, is insurance on a company's bonds. A rising premium means lenders want more money to carry the same risk.

"The widening in hyperscaler CDS is not driven by dealer hedging of new issuance," Slok said.

According to Slok, this is driven by a debt-financed spending cycle with rising leverage, negative free cash flow, and an uncertain payback on assets that lose value over time.

"The widening in hyperscaler CDS is not driven by dealer hedging of new issuance," Slok said.

The five hyperscalers sold about $121 billion in U.S. investment-grade bonds in 2025, compared with an average of roughly $28 billion per year from 2020 to 2024, according to Bank of America.

By midsummer 2026, issuance had passed $180 billion.