Morgan Stanley: AI's $1.6 Trillion Capex Will Be Highly Dependent on Credit Markets by 2027
nashnova research
Morgan Stanley estimates that hyperscalers and emerging cloud providers will see their combined free cash flow plunge to negative $340 billion by 2027, forcing AI infrastructure financing fully into the bond market and putting material supply pressure on the investment-grade credit universe.
Where does the money come from — why must AI building be funded by debt?
Hyperscalers (Amazon, Alphabet, Microsoft, Meta) and emerging cloud providers are pouring capital into AI infrastructure at a combined scale of roughly $1.6 trillion.
Morgan Stanley calculates their aggregate free cash flow (operating cash flow minus capex) will flip from positive in 2025 to negative in 2026, then plunge to roughly negative $340 billion in 2027.
This means → their own earnings can no longer cover the bill. The gap must be filled by issuing bonds — AI infrastructure financing shifts entirely to credit markets by 2027.
How big is the hole — what does negative $550 billion look like?
Factor in operating-lease payments (recurring costs for leased equipment and data centers), and the 2027 funding gap approaches negative $550 billion.
In plain terms = these companies cannot cover even their rent from earnings — the real hole is far larger than the headline cash-flow figure suggests.
2028 improves somewhat but remains deep in deficit — this is not a one-year squeeze but a multi-year financing marathon.
How much have they already borrowed — $255 billion and counting?
Hyperscalers have already issued roughly $255 billion in unsecured bonds this year, covering about 30% of their projected 2026 capex.
Amazon alone issued $98 billion (up from $15 billion last year). Alphabet issued $81 billion (up from $38 billion). SpaceX went from zero to $25 billion, at a weighted coupon of 5.9%.
This reflects an AI arms race already pushing bond supply higher in real dollars — Amazon's issuance grew more than sixfold in a single year.
Borrowing is getting more expensive — where is the rate pressure coming from?
Hyperscaler bond yields have risen 100 to 180 basis points since the start of the year. Oracle's 10-year spread widened the most — 180 bps total, of which 76 bps came from credit-spread expansion.
For AA-rated issuers like Amazon and Alphabet, roughly 70–80% of the yield increase comes from rising U.S. Treasury rates themselves.
In plain terms = borrowing costs are climbing on two fronts: the baseline government rate is higher (macro environment), and the market is also charging more for these companies' own credit risk (company-level repricing).
Stress test — who can take it and who is in danger?
Morgan Stanley stress-tested two issuer tiers under a scenario of zero operating-income growth from 2026 to 2027.
High-quality issuers (Amazon, Meta, Alphabet, Microsoft): issuing $400 billion at 7% would drop their interest-coverage ratio from 30.7× to 11.9× — This means → a steep decline, but still roughly in line with the investment-grade index median. Uncomfortable, but survivable.
Lower-quality issuers (Oracle, SpaceX): issuing $120 billion at 8.5% would slash coverage from 4.0× to 1.8× — a razor-thin safety margin, one step from an inability to service debt.
Can the market absorb this much debt — is capacity the same as willingness?
The U.S. investment-grade bond market is roughly $10 trillion in size, with year-to-date net inflows exceeding $100 billion, and it "almost never fully shuts down."
Yet Goldman Sachs' credit desk warned this week: the market looks calm on the surface, but once the AI-capex debt-issuance machine restarts in roughly two months, that calm will be tested.
Morgan Stanley's conclusion is sharp: capacity to absorb and willingness to absorb are two different things. 2027 will be the year that tests the gap — In plain terms = the market *can* buy, but the price it demands may be ugly.
市场有风险，内容仅供研究参考，不构成投资建议。