The simple version

Moody's Ratings, one of the major credit rating agencies, published a research note this week warning that the enormous spending on artificial intelligence infrastructure is beginning to threaten the credit quality of six large technology companies: Microsoft, Amazon, Alphabet, Meta, Oracle, and the specialized provider CoreWeave.

This is a different kind of warning from a falling stock. A stock price is a bet on how much a company will be worth. A credit rating is a judgment about whether a company can comfortably pay its debts. When a ratings agency gets nervous, it is not saying the business is worth less. It is saying the finances are getting tighter, and that is worth understanding on its own terms.

The numbers

• Moody's identified six companies whose credit quality it says is threatened by AI spending: Microsoft, Amazon, Alphabet, Meta, Oracle, and CoreWeave (Moody's Ratings research note, week of July 20, 2026, as reported by CNBC and Forbes)
• Moody's projects those six will spend about $785 billion on capital projects in 2026, rising toward about $1 trillion in 2027 (Moody's)
• That is roughly $2.15 billion a day across the six companies (our calculation from the Moody's figure)
• Direct debt across the six has reached about $460 billion (Moody's)
• Off-balance-sheet data-center lease commitments have reached about $1.2 trillion, of which more than $820 billion is for leases that have not started yet because the data centers are still being built (Moody's)
• Those lease commitments are about 2.6 times the direct debt, so most of the obligation sits off the balance sheet (our calculation from the Moody's figures)
• Moody's described the shift from an asset-light to an asset-heavy model as requiring "unprecedented levels of investment and capital raising" (Moody's)
• Moody's noted that Microsoft, Alphabet, Amazon, and Meta retain among the strongest corporate balance sheets in the world, and their investment-grade ratings are not under imminent threat (Moody's)
• The weaker links are Oracle, rated Baa2 with a negative outlook, two notches above high-yield status, and CoreWeave, rated Ba3, already in high-yield territory (Moody's)
• For scale, and counting different company sets: a Financial Times compilation put four hyperscalers' 2026 spending at about $725 billion, a Morgan Stanley estimate put five of them at about $805 billion, and this Moody's figure of $785 billion covers its six-company set (as reported)

What a credit rating actually is

A credit rating is a grade for how likely a borrower is to pay back what it owes. Companies borrow money by issuing bonds, which are loans from investors, and rating agencies like Moody's assign letter grades that tell those investors how safe the loan is.

The grades run from the top, where a company is considered extremely likely to pay, down through investment grade, and then across a line into what is formally called high-yield and informally called junk. That line matters enormously, because many large investors, such as pension funds, are only allowed to hold investment-grade bonds. Cross the line and a company loses access to a huge pool of lenders, and its borrowing costs jump.

That is why Moody's names the specific grades. Oracle at Baa2 with a negative outlook is investment grade but only two notches from that line, with the outlook signaling the direction of travel. CoreWeave at Ba3 is already across it, which is part of why it relies on complex private debt to finance its chips. The four giants sit far above the line, which is Moody's own caveat: they are not in danger, they are moving in a direction worth flagging.

Why AI spending shows up as credit risk

For two decades, these companies made money in one of the most profitable ways in business history. Software costs almost nothing to copy, so once it is built, each new customer is nearly pure profit. That produced wide margins, large cash balances, and almost no need to borrow. Moody's calls this the asset-light model.

AI breaks that model. Instead of copying software, these companies now have to build enormous physical things: data centers full of expensive chips, drawing enormous amounts of power. That is asset-heavy, the kind of capital intensity normally associated with utilities and factories, not software firms. Building it costs so much that even the most cash-rich companies on earth are turning to debt, to selling new stock, and to off-balance-sheet arrangements to fund it.

That last category is the one Moody's flags hardest, and it is worth understanding. The off-balance-sheet lease commitments total about $1.2 trillion, roughly two and a half times the $460 billion in direct debt. These are obligations to pay for data-center capacity that do not always appear as debt on the main balance sheet, which means the headline debt figure can understate the true scale of the commitments.

More than $820 billion of that $1.2 trillion is for leases that have not even started, because the data centers are still under construction. That is the part worth sitting with. Most of the obligation is not for capacity these companies are using today. It is for capacity they have committed to pay for once it exists.

So the credit warning is really a warning about a business-model change. Companies built to need no debt are taking on a lot of it, quickly, to fund a buildout whose payoff is uncertain. That does not make them risky in the way a struggling company is risky. It makes them different from what they were, and a rating agency's job is to notice when the ground shifts.

The Real Cost lens on borrowing to build

The mechanism here is one any household understands, scaled up enormously.

• A family that pays cash for everything has no lender to answer to and total flexibility. A family that borrows heavily to build something gains the building but takes on fixed payments and less room to maneuver
• The borrowing is not automatically bad: a mortgage on a home you will use for decades can be entirely sensible. The question is always whether the thing you built earns or saves enough to justify the payments
• That is exactly Moody's question about the data centers: the spending buys real capacity, and whether it pays off depends on AI revenue arriving fast enough to cover the growing obligations
• The off-balance-sheet piece has a household echo too: a debt you have committed to but tucked out of sight is still a debt, and pretending otherwise only changes when the reckoning arrives, not whether

None of this is a prediction that anything goes wrong. It is the reason a credit rating exists at all: to look past the profit headline at the obligations underneath, and to say plainly when they are growing faster than the comfort that covers them.

What this means

The credit angle is a useful second lens on the AI spending story the stock market has been telling. Stocks reacted to shrinking cash flow with falling prices; a rating agency reacted to growing debt and off-balance-sheet lease commitments with a warning about credit quality. Both are watching the same spending, from different seats.

For a household, the transferable lesson is the value of the off-balance-sheet idea. When you assess anyone's finances, including your own, the obligations that do not show up on the main statement, the leases, the commitments, the things tucked away, are exactly the ones worth dragging into the light. A rating agency does that professionally. It is a good habit to borrow.

What this is NOT

This is not a prediction that any company's credit rating will be cut or that any company will struggle to pay its debts; Moody's itself says the largest four are not under imminent threat. This is not advice to buy, sell, or hold any stock or bond, or any security or fund. This is not a claim that AI spending is a mistake, which this article does not judge; it reports what a ratings agency observed. The figures are Moody's, cited to Moody's, and ratings and outlooks change. This is not investment advice of any kind.

Sources

• Moody's Ratings research note on AI infrastructure spending and hyperscaler credit quality, week of July 20, 2026, as reported by CNBC: https://www.cnbc.com/2026/07/24/moodys-ai-spending-credit-quality-amazon-meta-alphabet.html
• Moody's Ratings research note, as reported by Forbes: https://www.forbes.com/sites/eriksherman/2026/07/23/big-ai-data-center-owners-are-massively-expanding-their-debt/
• Moody's Ratings: https://www.moodys.com/
• Alphabet Inc., Form 8-K, Q2 2026, SEC EDGAR: https://www.sec.gov/Archives/edgar/data/0001652044/000165204426000066/googexhibit991q22026.htm