AI Boom Pushes Tech Giants Into Record Debt and Hidden Obligations
The AI buildout is no longer just a story of innovation, according to new Goldman Sachs and Morgan Stanley analyses circulated by the Tel Aviv Stock Exchange. It has become one of the biggest capital-spending waves in modern history, as hyperscale companies race to build data centers, buy advanced chips, and lock in long-term power, real estate, and computing deals before profits are proven.
Goldman Sachs estimates hyperscalers, including Microsoft, Alphabet, Meta, Amazon, Oracle and other cloud players, could spend about $1.1 trillion on capex in 2027, with a bullish scenario reaching $1.4 trillion. Morgan Stanley, which was much more cautious in late 2025, has now largely converged on Goldman’s baseline. The report says AI investments could absorb nearly all operating cash flow at some major tech firms by the end of 2026, forcing a shift away from self-funding.
That is already visible in debt markets. AI-related companies raised about $236 billion in debt from the start of the year through the end of May, up 357% from a year earlier, and 2026 issuance is projected at roughly $570 billion. Combined gross leverage among leading hyperscalers has doubled in just two quarters, from 0.9 to 1.8 debt-to-EBITDA. Credit spreads have widened, and some bonds now trade closer to an A profile than the AA levels that characterized these names for years.
The banks also estimate about $1.8 trillion in off-balance-sheet obligations, including $982 billion in future purchase commitments and more than $822 billion in lease and rental contracts not yet effective. Because many projects are still under construction, depreciation has not yet hit earnings, but Goldman and Morgan Stanley warn of a coming “depreciation wall.” They say accumulated depreciation for Alphabet, Meta, Microsoft and Oracle alone could exceed $520 billion within three years.
The article says the real risk is the gap between rising capex and slower-growing revenue and free cash flow forecasts. It also notes that hyperscalers have more than $2 trillion in remaining performance obligations, much of it concentrated among a small group of counterparties. The piece does not claim a bubble is inevitable, but says AI has become a giant global financing experiment whose success depends on future profits rising fast enough to justify the spending.
