AI Hyperscaler Off-Balance-Sheet Liabilities Rise to $3.1 Trillion
nashnova research
A new Morgan Stanley report finds that off-balance-sheet liabilities across Amazon, Microsoft, Google, Meta, Nvidia, and Broadcom have hit $3.1 trillion — up $1.3 trillion in roughly three months — signaling that the true funding pressure behind AI infrastructure far exceeds what public filings show.
Where does a $3.1 trillion "hidden tab" come from?
Morgan Stanley analyst Todd Castagno tallied leases, debt, equity commitments, and purchase obligations across six companies — Amazon, Microsoft, Google, Meta, plus Nvidia and Broadcom — and found off-balance-sheet liabilities totaling over $3.1 trillion.
This means → none of this debt shows up on a balance sheet, yet the companies are already locked into paying it out. Roughly three months ago the figure was $1.8 trillion — the pace of increase is striking.
In plain terms = the liabilities you see in earnings reports are the tip of the iceberg; a fresh $1.3 trillion mass just surfaced beneath the waterline.
Can these companies actually afford what they are spending?
The report projects that by 2027, hyperscaler capital-expenditure cash outlays will exceed $1.2 trillion, while combined operating cash flow over the same period is estimated at roughly $1 trillion.
This means → capex has already outstripped their ability to self-fund, leaving a gap of at least $200 billion — and the report warns capex will keep climbing while cash-flow growth narrows.
To plug the gap, hyperscalers are pulling every lever at once: issuing public-market bonds, suspending share buybacks, and raising equity — each of which dilutes or redirects value away from existing shareholders.
How do these off-balance-sheet arrangements actually work?
Hyperscalers use special-purpose vehicles — SPVs, standalone legal entities set up for a single project so the financing stays off the parent company's books — to fund data-center construction.
Specific mechanisms include guaranteeing debt raised by SPVs backed by private credit firms, and signing long-term purchase commitments for GPUs and networking gear to ensure compute capacity is ready when a data center goes live.
In plain terms = the companies tuck their borrowing, building, and equipment obligations inside shell entities, keeping their own financial statements looking clean — but the cash still has to be paid eventually.
What does this mean for investors?
Castagno states plainly in the report: "Every new arrangement requires a commitment of future cash, and every upward revision to capex requires an incremental source of capital."
This reflects a core risk: publicly disclosed capex figures understate the real funding pressure, while free cash flow is overstated because operating leases are shifted off-balance-sheet via SPVs.
The report warns that returns on new AI infrastructure must cover rising capital costs or the outcome will be "catastrophic" — and there is no sign yet that AI revenue growth is anywhere near the inflection point needed to cover this scale of spending.
市场有风险，内容仅供研究参考，不构成投资建议。