Moody’s warns debt-funded AI spending is eroding cash flow at Alphabet, Amazon and Microsoft
Moody’s and other agencies are raising alarms as AI infrastructure costs leave major tech firms drawing heavily on bond markets and burning cash to stay ahead.

One number in Alphabet’s quarterly report breaks the pattern. While revenue reached an all-time high, free cash flow turned negative and capital expenditure doubled to $44.9 billion for the quarter, causing a net drain of cash.
FinanceBuzz reported on August 18 that Alphabet spent at twice the rate it collected, and Moody’s Ratings put Alphabet, Amazon, Microsoft, Meta, Oracle, and CoreWeave on notice, warning that massive AI spending is eroding free cash flow and adding risk to companies previously considered untouchable, as Forbes wrote on July 23. Apple is not in the room.
At stake is whether the AI infrastructure buildout can pay for itself before lenders lose patience. Alphabet and Amazon are buying share in enterprise AI markets by building first and hoping revenue will follow, while Oracle’s aim is to become the central compute landlord. Cryptonomist reported on August 17 that Oracle’s spending hit $55.66 billion for fiscal 2026, overshooting its $50 billion target, with guidance as high as $95 billion for next year.
Meta, aiming to catch up in GPUs, raised $30 billion in bonds in October 2025 and separately financed a $12 billion Texas data center. CoreWeave, lacking an ad or cloud cushion, is exposed through and through. The agencies want to see these debts serviced if AI growth stalls, but none of the companies’ plans budget for that outcome.
Moody’s set off the alarm, but the real pressure is older: depreciation is already hitting the books while much of the hoped-for AI revenue remains in the future. Silicon Analysts calculated on July 10 that the US hyperscalers spent approximately $434 billion on property and equipment in the four quarters through March 2026, booked against about $149 billion of depreciation. That means real cash goes out years before revenue arrives.
Remio, citing Moody’s, reported in August 2026 that this year’s combined capital expenditures from the six watched companies may total $785 billion.
Any company can live with thin margins, but none survives for long spending ahead of collections while servicing new debt.
Even bullish analysts admit the pace is new. Cryptobriefing wrote on June 12 that Microsoft, Alphabet, Amazon, Meta, and Oracle sold $159 billion in bonds by mid-June, up 47 percent over all of 2025, and Yahoo Finance reported on July 18 that related investment-grade issuance hit a record $182 billion by mid-July.
Benzinga quoted Goldman Sachs in August 2026 calculating that hyperscaler leverage ratios doubled from 0.9 to 1.8 times in just six months, while the bond market’s appetite for new supply dropped from $75 billion to $25 billion. Investors are charging more to lend: Cryptonomist’s August 17 note cited widening bond spreads for Meta, Alphabet, and Amazon as buyers raise yield demands.
The risk ricochets down the line. S&P Global, Cryptonomist reported on August 17, cut Oracle’s rating to BBB-, just above junk, in July 2026, and Oracle’s five-year credit default swaps spiked. The prime names lose cheap financing as borrowing costs rise, forcing companies to pause projects—what penciled out at 4 percent interest no longer works at 6 percent.
The effect stretches to government finance: Reuters (via AOL) on August 21 and Mogazmasr on August 18 both reported that 30-year US Treasury yields reached a 20-year high near 5.27 percent, the cloud builders’ spree cited alongside federal deficits. The big technology borrowers now compete with the Treasury for long-term money, and both lose to the yield.
WorldCom and Global Crossing made the parallel mistake twenty-five years ago, borrowing tens of billions to lay fiber expecting future use to cover the cost. When revenue fell short, the creditors dictated their fate, despite the network’s eventual usefulness. The difference today is that Alphabet, Amazon, and the rest are stable, cash-generative giants with on-demand networks, not startups banking on projections. Yet the opposite example holds too: Comcast and cable operators used debt to build networks in the 2000s, carried that leverage through the financial crisis, and emerged in control. Debt buries the slow and the unlucky; it rewards the patient and necessary.
Google Cloud’s 82 percent growth, reported by 24/7 Wall St. on August 17, shows gains are real, but the scale of investment eclipses operating cash in sheer volume. Barclays expects hyperscaler capital expenditures to take up 85 to 90 percent of total operating cash flow between 2026 and 2028, according to a cited August 2026 report. A further weight is off-book: Moody’s counted $662 billion in signed but not yet started data-center leases in February 2026, commitments exceeding all disclosed debt.
If the bet fails, the architects of the build don’t pay; their rewards are equity-based and vest long before the depreciation runs out. Instead, the costs hit bondholders, index funds holding technology credit, and towns tied to unfinished data centers that shutter once capital dries up. Meanwhile, Nvidia and chipmakers get paid up front, bankers earn on every bond transaction, and utilities profit from multi-decade power contracts however things play out.
The judgment is exact. Investors spent two years pricing AI-related risk into Nvidia’s earnings but barely accounted for the stress on technology balance sheets.
With Moody’s moving the debate to credit filings, the terms are shifting: companies funding their own AI growth choose their pace, but those relying on bonds now have their speed set by lenders, who tighten terms each week the revenue lags.