Moody's Ratings says the largest artificial intelligence infrastructure buyers are replacing a low-capital software model with one that requires factories of servers, chips and data centers. CNBC reported the assessment on July 24, 2026, as the ratings firm projected that annual capital expenditure across six tracked companies would approach $1 trillion next year.
The warning is not that the biggest technology groups are running out of money. It is that Amazon, Alphabet, Meta, Microsoft, Oracle and CoreWeave are taking on new obligations faster than AI revenue has had time to mature, changing how creditors must measure balance-sheet strength.
Capital Spending Is Moving Toward $1 Trillion
Moody's expects the six companies to spend about $785 billion on capital assets in 2026 and roughly $1 trillion in 2027. That money supports physical capacity: buildings, computing hardware, power connections and the equipment required to train and operate increasingly large AI systems.
The shift is significant because software businesses traditionally expanded without rebuilding their physical base at the same pace as sales. Data centers reverse part of that advantage. Cash must leave before a facility begins producing revenue, and equipment can become less competitive while companies are still recovering the cost of installing it. That puts greater weight on utilization, energy costs and the useful life of each generation of chips. Capacity that earns less than expected can weaken free cash flow even when demand continues to grow.
Direct debt across the group has reached approximately $460 billion, Moody's said. Equity is also part of the funding mix. Alphabet announced a stock sale worth almost $85 billion in June, showing that even a company with substantial operating cash is willing to issue ownership stakes to support the scale of its infrastructure plan.
Leases Add Obligations Beyond Reported Borrowing
Long-term data center leases create the largest single pool of commitments in the report. Moody's counted about $1.2 trillion across the six companies. Over $820 billion of that total relates to leases that have not begun because the facilities are still under construction. Much of that obligation therefore represents capacity that cannot yet produce revenue but will require payments after the sites are delivered.
Those contracts may sit outside conventional debt totals, but they still commit future cash. Moody's treats them as debt-equivalent liabilities because companies will owe rent once the sites are delivered. The obligation therefore becomes economically important even when it is not presented alongside bonds and loans on the balance sheet.
The timing creates the central credit tension. Construction commitments are being fixed now, while demand, pricing and utilization will be tested over many years. A company can reduce immediate borrowing by leasing capacity, but it cannot easily escape the payment schedule if AI customers use less computing power than expected.
Strong Balance Sheets Do Not Remove Company-Level Gaps
Moody's does not expect imminent investment-grade rating trouble for Microsoft, Alphabet, Amazon or Meta. It said those companies remain among the most financially resilient corporations globally, supported by expanding cloud operations and large existing businesses that generate cash outside the newest AI products.
The pressure is more immediate for Oracle. Its Baa2 rating carries a negative outlook and stands two steps above speculative grade. A narrower rating cushion means that higher debt, large lease commitments or weaker returns can matter sooner, even if the same spending supports a broader industry expansion.
CoreWeave starts from a still different position. The specialized AI cloud operator is rated Ba3 in the high-yield market and uses complex private borrowing to finance fleets of graphics processors. It has less room than the largest platforms to absorb a long period between purchasing capacity and earning an adequate return from it.
These differences prevent a single conclusion about all six companies. The same data center can represent manageable expansion for a cash-rich platform and a concentrated refinancing risk for a lower-rated provider. Credit analysis therefore turns on who owns the obligation, how it was financed and which customers ultimately pay for the computing capacity.
AI Returns Must Arrive Before the Financing Hardens
Moody's also identified a circular structure in the AI market. Large cloud providers invest in private AI developers, including OpenAI and Anthropic, while those developers spend heavily on computing services supplied by the same investors. The relationships can strengthen demand, but they also tie suppliers to a common set of customers and expectations. If a small number of laboratories slows purchases or struggles to finance promised orders, several infrastructure providers could face weaker utilization at the same time.
There are important offsets. AI computing demand remains strong, cloud revenue continues to grow and the largest providers hold long-term customer contracts collectively worth hundreds of billions of dollars. Those agreements can make future revenue more predictable and help support credit quality while capital spending remains elevated.
Even with those protections, Moody's said the financial question is moving from access to capital toward the productivity of what that capital builds:
“Investors will increasingly focus on these companies’ ability to realize an adequate return on investment.”
The credit outcome will depend on whether revenue catches the obligations before they become fixed costs without matching demand. Direct borrowing can be refinanced, equity can spread risk among shareholders and leases can shift where financing obligations are recognized, but none of those structures can replace cash generated by occupied data centers. The spending surge becomes durable only if customers pay enough, for long enough, to cover both the technology and the financing behind it.