Analysis
Major technology companies are carrying roughly $3 trillion in off-balance-sheet AI-related financial obligations, nearly double a July estimate of $1.65 trillion and triple their combined reported on-balance-sheet debt, Fortune reported. The obligations largely consist of long-term purchase commitments for GPUs and servers, alongside multi-year data-center lease agreements -- structures that represent real financial commitments but do not appear as conventional debt on a company's balance sheet.
The growth rate is the real story
The growth rate is the more alarming figure than the level itself: hidden AI-related debt at US tech giants has expanded roughly eightfold in four years, reaching $1.65 trillion by the July estimate before this latest figure nearly doubled that within a matter of weeks. That pace of growth in an off-balance-sheet financing category is difficult for outside analysts, credit-rating agencies, or even the Federal Reserve itself to track in real time, since the obligations are scattered across purchase agreements and lease structures with varying disclosure requirements rather than centralized in a single reportable debt figure.
“## The Fed is publicly split Fed officials are notably split on how worried to be.”
The Fed is publicly split
Fed officials are notably split on how worried to be. New York Fed President John Williams has said he does not see the AI buildout as a bubble, while San Francisco Fed chief Mary Daly has called the sheer growth rate and scale of the spending 'potentially very worrisome.' That kind of public disagreement among senior Fed officials is itself informative -- it suggests the central bank does not have a settled internal view on how to model the systemic risk this financing structure represents, which is precisely the concern Fortune's reporting highlights in its framing that the Fed doesn't fully know who is financing the boom.
- Microsoft, Meta, Alphabet, Amazon -- the hyperscalers carrying the bulk of the reported off-balance-sheet AI obligations through purchase commitments and leases
- Private credit funds -- increasingly key financiers of AI-related assets, per separate reporting on the sector
- The Bank for International Settlements -- has warned that an AI spending pullback could hit credit markets comparably to the 2008 financial crisis, given how much of the financing runs through interlocking, opaque deal structures
The structural parallel worth sitting with is that 2008's systemic risk was amplified less by the size of any single mortgage exposure than by how interconnected and opaque the securitized instruments built on top of those mortgages were. A $3 trillion AI financing web spread across purchase commitments, leases and private credit vehicles carries a similar opacity risk, even though the underlying assets -- GPUs and data centers -- have very different fundamentals than subprime mortgages.
The counterweight is that GPUs and data centers are real, productive assets generating revenue today, unlike many of the synthetic instruments at the center of 2008 -- this is closer to conventional infrastructure financing at unusual scale and speed than to a purely speculative bubble. Whether that distinction holds depends entirely on whether AI revenue growth continues to justify the capex, which is the same question Alibaba's stock just got marked down 10% on this week.
The next hard deadline is the Fed's December meeting, where staff have signaled they intend to present a first standardized estimate of aggregate hyperscaler purchase-and-lease exposure -- until then, the $3 trillion figure is Fortune's synthesis of scattered disclosures, not an official central-bank number.
