Moody's Missed on Earnings and Lost 40% of Its High-Yield Fees: AI Credit Became the Ratings Engine, and the Same Credit Is Already Cracking

Moody's reported third-quarter 2026 adjusted diluted EPS of $4.14 on Oct. 21, against consensus of $4.25 to $4.31 depending on the tracker, on revenue of $2.02B. GAAP diluted EPS was $4.24, up 18%, and the adjusted operating margin held at 52.6%. The miss came from the composition of the ratings book, not from credit demand: shareholders of a record earnings base got a flat revenue line.

The damage sat inside Moody's Investors Service, where revenue fell 1% to $1,087M even though the segment's own investment-grade line grew 18%. Moody's Analytics did what it does: revenue of $936M, up 3%, with annual recurring revenue of $3.69B, up 8%. The beat streak mattered because it was embedded in a shares stock trading at roughly 28x the midpoint of reaffirmed full-year guidance.

Corporate Finance revenue fell 8% to $530M while bank-loan revenue fell 27% and high-yield revenue dropped 40%. Investment-grade revenue rose 18% on 16% more issuance. That is not a ratings-market problem: total rated issuance volume across Moody's books fell only 2%, even as aggregate debt issuance dropped 6% year over year in the third quarter. The fees collapsed on mix while the market itself held up.

Recurring revenue reached about 64% of company revenue this quarter, up from about 61% a year earlier, and 34% of Moody's Investors Service revenue against 31% a year ago. The migration is real but early: ratings transaction revenue is still 66% of that segment. Surveillance and analytics are the ballast while new-issue fees remain the swing factor in every quarter that follows.

Full-year guidance was reaffirmed, not raised: adjusted diluted EPS of $16.50 to $17.00 and high-single-digit organic revenue growth. Nine-month adjusted EPS of $13.15 leaves $3.35 to $3.85 for the fourth quarter, a midpoint near $3.60 against $3.64 a year earlier. The street's $16.98 full-year estimate therefore requires the very top of the guided range. Management is signalling a flat fourth quarter, not a recovery, and the shares fell to $475.35, down 1.2%.

The five largest US hyperscalers spent roughly $575B of capital expenditure across their four most recently reported quarters, and J.P. Morgan estimates 2026 hyperscaler capex at $697B. Some of that is funded with bonds, which is where Moody's fees come from — and it is where the credit risk now sits. Moody's own ratings research in February 2026 put committed future data-center leases at the top five US hyperscalers at $969B, with $662B not yet commenced and sitting off balance sheet, roughly 113% of their combined adjusted debt.

The credit Moody's now earns fees on is repricing fast. CoreWeave, rated A3 by Moody's on a securitization in March 2026, saw its April-issued bonds yield near 15% by Oct. 15, with the cost of insuring its debt through credit default swaps near 1,000 basis points and the market implying roughly a 63% chance of default over five years, up from about 30% a year earlier. Total debt stands near $50B.

The Federal Reserve Bank of Chicago put average bank exposure to AI-adjacent industries at about 0.8% of total assets, which sounds harmless. The composition matters more: outstanding exposure averages roughly 9% of tier-1 capital while committed exposure runs near 25%. Committed exposure runs more than double the drawn balance, and commitments become loans before a borrower misses, so a stress scenario starts at the committed number, not the drawn one.

On Oct. 13, 2026, London Stock Exchange Group confirmed talks with Permira to buy credit-data platform Octus for about $3.5B, calling the deal immediately accretive to its Data & Analytics revenue and highlighting Octus's leading position in illiquid and private credit. That price is the signal: as credit migrates toward AI capital expenditure and private credit, the scarce asset is no longer the rating, it is the data and workflow layer that prices it.

The peer comparison is the uncomfortable part. In the second quarter of 2026, S&P Global Ratings grew revenue 17% to a record $1,339M with transaction revenue up 25% to $746M, and its own data showed US investment-grade issuance up 42% and US high yield up 25%. One quarter later, Moody's ratings revenue was down 1% and its transaction revenue down 5%, with high-yield revenue down 40%. S&P Global's ratings engine grew 17% into a record quarter while Moody's turned negative one quarter later, which points at pipeline and mix rather than at the issuance cycle alone.

Over days to quarters, the next three months test whether this was a Moody's problem or a ratings-cycle problem. Over one to three years, the question is whether AI capital expenditure becomes a durable fee pool or a concentration risk that lands on the same balance sheets Moody's is rating.

Fact: Moody's missed, high-yield ratings revenue fell 40%, investment-grade revenue rose 18% on AI-related issuance, and recurring revenue plus analytics carried a 1% revenue quarter. Inference: Moody's fee base has become a derivative of AI capital expenditure, with fewer, larger, better-credit issuers buying fewer but bigger mandates, which raises revenue quality and lowers its correlation with the credit cycle. Speculation: if AI capex slows in 2027, the agency loses its only growing ratings line just as the lenders financing the same build-out start booking losses, and surveillance and analytics become the shock absorber rather than the growth engine.

The practical consequence for portfolios is a split. On the fee side, the winners are the holders of recurring credit and private-credit data: analytics ARR of $3.69B growing 8%, 95% retention and a consolidation wave pricing credit data at $3.5B a deal. On the risk side, the exposure sits in AI-adjacent bank commitments near 25% of tier-1 capital, business development company software books near $115B, and AI infrastructure bonds yielding 15%. The same AI capital expenditure that created Moody's growth pool is now the collateral behind the credit losses, and the October print showed both sides for the first time.