Credit rating agencies like Moody's and S&P are increasingly wary of the massive investments hyperscalers are making in AI infrastructure, which are straining their financial health. Moody's projects capital expenditures for physical assets like data centers to reach $785 billion in 2026 and nearly $1 trillion by 2027. This spending binge is leading to significant debt accumulation, with direct debt across the six largest hyperscalers — Amazon, Meta, Alphabet, Microsoft, Oracle, and CoreWeave — already at approximately $460 billion, according to Moody's.
Oracle, in particular, has drawn sharp attention. S&P Global Ratings downgraded Oracle's credit rating to BBB- in July, just one notch above junk status, citing surging capital expenditure, negative free cash flow, and customer concentration. Moody's maintains a Baa2 rating with a negative outlook, expressing concerns that Oracle's leverage could temporarily approach 5x EBITDA. Oracle's fiscal year ending May 2026 revealed negative free cash flows, with the bulk of revenue spent on capital expenditure and roughly $260 billion in data center leases, some starting next year. The company expects capital expenditures of up to $95 billion in fiscal 2027.
Beyond direct debt, rating agencies are alarmed by the lack of transparency regarding off-balance-sheet liabilities. Moody's estimates that outstanding data center leases signed by the six largest hyperscalers have surged to $1.2 trillion, with $820 billion yet to be fully recognized on their balance sheets. These liabilities, which are almost three times higher than outstanding direct debt, are increasing at a much faster pace. Moody's also highlighted the risk of "circular financing" and potential credit risks from concentrated customer contracts, particularly with AI frontier labs like OpenAI and Anthropic, whose ability to meet obligations could impact hyperscalers.
Investor sentiment reflects these concerns. A recent $12.547 billion bond offering by Meta Platforms for a data center joint venture in El Paso, Texas, attracted only $20 billion in orders, a lukewarm reception for a deal of its size, indicating growing investor anxiety about hyperscaler debt. Despite these worries and rising financing costs, particularly at the longer end, analysts like Ravi Shukla of Wells Fargo do not anticipate an abrupt funding shutout, though they expect wider relative spreads and larger new issue concessions for hyperscalers. Most of these companies still hold strong investment-grade ratings, providing some headroom, but the opacity of collateral behind new debt and reliance on brand reputation rather than asset valuation are growing concerns.
Moody's warns that the widespread use of residual value guarantees to shift data center liabilities off the balance sheet risks a "material" deterioration in hyperscalers' credit profiles. The agency also pointed out a $2.1 trillion backlog of contracts signed by Amazon, Microsoft, Alphabet, and Oracle, with OpenAI and Anthropic representing a large share, introducing new dimensions of concentration and credit risks. Should the projected demand for frontier models fail to materialize, these customers might struggle to meet their obligations, impacting the hyperscalers. The bond market is already pricing in higher risk for Oracle, with its share price halving since June to $129 and bond yields surging to 7% to 8%, compared to 2.5% to 3.5% for Alphabet and Amazon bonds.
