Beware! AI’s Off-Balance-Sheet Bill Has Exceeded $4.1 Trillion

On October 30, Intel’s intraday decline briefly widened to 30%, with shares falling as low as $64.24 and nearly $150 billion in market value wiped out, marking its worst single-day performance since the 1970s. Its Q4 revenue guidance of $15 billion to $16 billion was more than $1 billion below analysts’ average forecast of approximately $17 billion; CFO David Zinsner explicitly stated that the company does not expect to achieve positive cash flow this year. An incident at its Israeli factory is expected to reduce Q4 revenue by approximately $300 million, while PC shipments are projected to decline by approximately 12% this year and another 10% in 2027. From the start of the year through October 29, Intel had at one point gained nearly 150%, with the market pricing it as an AI beneficiary. Its earnings report revealed a different reality: fabrication plants were consuming its cash flow, while its core PC business was being squeezed by rising memory prices. Intel’s predicament reflects one company’s operational missteps. Then, the $4.11 trillion figure was compiled, pointing to a problem far more systemic than the operational failures of a single company.

The Composition of the $4.11 Trillion

This $4.11 trillion is not debt in the traditional sense. It mainly comprises three categories: non-cancellable purchase commitments, signed leases that have not yet commenced, and guarantees and credit support provided for data centre and chip projects.

By category, Meta’s non-cancellable purchase commitments surged from $349 billion to $455 billion in the third quarter, leases that had not yet commenced increased from $279 billion to $356 billion, and residual value guarantees for the Hyperion and El Paso joint ventures totalled $41.5 billion. Amazon’s unconditional purchase obligations nearly doubled from $130 billion to $254 billion, while leases that had not yet commenced rose from $137 billion to $165 billion. Its commitments increased by $131 billion in a single quarter, a rise of 41%, the largest percentage increase among hyperscalers. Alphabet’s purchase commitments rose by $95 billion, leases that had not yet commenced increased from $85 billion to $131 billion, energy equipment guarantees doubled to $16.2 billion, and it added $6.6 billion in TPU hardware repurchase guarantees. Its total off-balance-sheet commitments reached $1.059 trillion, making it the first individual company to exceed $1 trillion. Microsoft added $35 billion, an increase of 6%, showing the most restrained expansion. However, its purchase commitment figures still reflect the disclosures in its end-June Form 10-K, and the actual figure will most likely be revised upward in its next annual report.

SpaceX recorded the most aggressive growth: its off-balance-sheet commitments rose from $29 billion at the end of the second quarter to $40 billion at the end of the third quarter, then increased further to $51 billion on October 26, a rise of 174%.

Taken together, these categories make the nature of off-balance-sheet commitments clear: they are not money already borrowed, but money already committed to be spent. Once a lease is signed, rent must be paid regardless of whether the servers are fully installed or the computing capacity has been sold. Once a purchase commitment is made, payment must be settled regardless of whether chips arrive on schedule or demand materialises. These commitments do not appear as interest on the income statement or as liabilities on the balance sheet; they exist only as “commitments” in the financial statement footnotes. Yet the rigidity of their cash outflows is no less than that of on-balance-sheet debt.

The companies’ actual debt and lease liabilities recorded on their balance sheets total approximately $1 trillion. The ratio of off-balance-sheet commitments to on-balance-sheet liabilities exceeds 4 to 1.

Nearly $1 Trillion Added in 90 Days

From approximately $1.8 trillion at the end of June to $3.12 trillion at the end of August, and then $4.11 trillion in October, off-balance-sheet commitments more than doubled in just over four months. For the five companies that have already filed their third-quarter reports alone, combined off-balance-sheet commitments rose from $2.48 trillion to $3.05 trillion, a quarterly increase of $577 billion, or 23%.

The pace of growth is itself a signal. The expansion of on-balance-sheet debt is constrained by credit ratings, the bond market’s capacity to absorb issuance and interest costs; the expansion of off-balance-sheet commitments is much “quieter” in accounting terms. When a company finds that adding leverage to its balance sheet would trigger a rating downgrade and raise financing costs, structuring spending as long-term leases or purchase commitments carries a much lower “cost” in its financial statements.

Meta illustrates this path most clearly. A company with $97 billion in operating cash flow and credit ratings of Aa3/AA- saw its CDS reach an all-time high following its earnings release, then announced plans to raise up to $95 billion in equity, of which $37.5 billion had already been completed, with another approximately $60 billion in an at-the-market offering programme awaiting execution. An AA- rated company is doing three things simultaneously: accumulating off-balance-sheet commitments, seeing the market push its CDS higher, and replenishing capital through equity financing. These three developments occurring together indicate that the market has begun incorporating off-balance-sheet commitments into credit pricing, while the company has chosen to respond with equity rather than debt.

Amazon offers another perspective. Quarterly capital expenditure reached $77.26 billion, while trailing twelve-month free cash flow swung from positive $14.8 billion a year earlier to negative $42.4 billion. Its off-balance-sheet commitments increased by $131 billion in a single quarter, the largest percentage increase among hyperscalers. The simultaneous acceleration of capital expenditure and off-balance-sheet commitments in the same quarter is no coincidence—both are accounting reflections of the same AI infrastructure arms race.

Diverging Rating Methodologies

Goldman Sachs credit strategist Amanda Lynam pointed out that S&P adjusts leverage for off-balance-sheet leases, while Moody’s and Fitch do not. The same company can therefore show different leverage levels under different frameworks. When off-balance-sheet commitments amount to a few hundred billion dollars, methodological differences have a limited impact; when they reach $4.11 trillion, equivalent to more than four times on-balance-sheet liabilities, those differences are sufficient to alter credit assessments.

On October 30, S&P lowered Meta’s outlook from “stable” to “negative” while maintaining its AA- rating. Its rationale explicitly pointed to AI capital expenditure pushing free cash flow into negative territory, and it specifically noted that Meta’s future lease obligations had increased to $355.7 billion. Meta’s recorded debt is not high, but once the $355.7 billion in lease obligations is included in adjustments under S&P’s methodology, its leverage trajectory changes.

Separately, Mark Habib, S&P Global’s head of digital infrastructure, noted that hyperscalers may have substantial undisclosed AI-related debt arising from implicit guarantees for data centre construction or smaller AI companies—what he described as “interconnected financing structures”. Over the past year, US technology giants have provided up to $300 billion in guarantees for AI data centres and chips. These guarantees may be spread across different line items in the form of residual value guarantees, joint venture commitments and credit support—Meta’s $41.5 billion in residual value guarantees for Hyperion and El Paso is one example, while Alphabet’s energy equipment guarantees and TPU hardware repurchase guarantees represent another category.

The Bank for International Settlements has previously warned that the opacity of AI financing exacerbates financial vulnerabilities, with circular financing posing the most prominent risk. S&P has also warned that revenue inflation driven by circular financing is one of the key factors increasing uncertainty in the AI industry.

How Does the Market View This?

The risk of off-balance-sheet liabilities lies not in their being “hidden”, but in the rigidity of cash outflows and the uncertainty of the corresponding revenue. The transmission chain is: cash receipts fall short of expectations → project cash becomes insufficient → collateral values decline and financing conditions tighten → parent companies or guarantors provide additional funding → construction and procurement are cut back.

The starting point is cash receipts. Project-level revenue depends on whether computing capacity can be rented out at sufficiently high rates. If rental rates fall, project cash inflows become insufficient to cover leases and purchase commitments, putting project companies under pressure first. Collateral values then decline, financing tightens, and parent companies or guarantors are required to provide additional funding. At that point, off-balance-sheet commitments turn into cash outflows on the balance sheet. If the parent company’s own cash flow is also shrinking, providing additional funding means further debt or equity financing.

The key feature of this chain is that it has a threshold. When cash receipts are normal, off-balance-sheet commitments are merely a line in the footnotes; once a project develops a cash shortfall, the rigidity of those commitments can transmit pressure back to the parent company far faster than expected. Intel losing 30% of its market value in a single day on October 30 is a micro-level example. The macro-level version is this: when more than one project company experiences a shortfall at the same time, part of the $4.11 trillion could rapidly turn into actual financing needs.

Mainstream valuation models still rely primarily on on-balance-sheet metrics as their core inputs. But that framework is beginning to shift: Meta’s CDS reached an all-time high after its earnings release, S&P lowered its outlook, and Goldman Sachs publicly highlighted inconsistent rating methodologies. Credit markets, rating agencies and sell-side research have all begun treating off-balance-sheet commitments as an independent variable.

The trigger for a systemic repricing may not be the scale itself, but the gap between the growth of off-balance-sheet commitments and the growth of operating cash flow. When operating cash flow can cover on-balance-sheet obligations and part of off-balance-sheet commitments, off-balance-sheet liabilities represent manageable future expenditure. Once operating cash flow growth falls behind commitment growth, that rigidity shifts from a “long-term arrangement” into “near-term financing pressure”.

Intel’s 30% decline on October 30 is the bill that has already arrived. The $4.11 trillion is the portion that has yet to arrive. The market took one day to price the former; how long it takes to price the latter depends on when the first transmission chain moves from “cash receipts falling short of expectations” to “parent companies providing additional funding”.