AI’s Off-Balance-Sheet Bill Surpasses $4.1 Trillion: A Cause for Caution

Q4 revenue guidance of $15 billion to $16 billion left Intel more than $1 billion short of analysts’ average forecast of approximately $17 billion. On 30 October, its shares touched $64.24 as the intraday loss briefly reached 30%, wiping out nearly $150 billion in market value in its worst single-day performance since the 1970s. CFO David Zinsner explicitly said the company did not expect positive cash flow this year. An incident at its Israeli factory is expected to cost approximately $300 million in Q4 revenue; PC shipments, meanwhile, are projected to contract by approximately 12% this year and a further 10% in 2027. Investors had previously treated Intel as a beneficiary of AI, with its shares gaining nearly 150% at one stage between the start of the year and 29 October. The earnings report challenged that view. Fabrication plants were draining cash flow, while rising memory prices were putting pressure on the core PC business. Those difficulties stem from Intel’s own operational missteps. The subsequent compilation of the $4.11 trillion figure, however, revealed a much more systemic problem than one company’s operational failures.

Off-Balance-Sheet Obligations Reach $4.11 Trillion

Three principal categories account for the $4.11 trillion: purchase commitments that cannot be cancelled, leases signed but yet to begin, and guarantees and credit support for data centre and chip projects. None constitutes debt in the conventional sense.

The third-quarter breakdown shows Meta increasing its non-cancellable purchase commitments from $349 billion to $455 billion and its leases yet to begin from $279 billion to $356 billion. Its residual value guarantees for the Hyperion and El Paso joint ventures came to $41.5 billion. For Amazon, leases yet to begin rose from $137 billion to $165 billion, alongside a near-doubling of unconditional purchase obligations from $130 billion to $254 billion. The resulting quarterly increase in commitments was $131 billion, or 41%, the steepest percentage rise among hyperscalers. At Alphabet, an additional $95 billion in purchase commitments was accompanied by an increase in leases yet to begin from $85 billion to $131 billion. Energy equipment guarantees doubled to $16.2 billion, and TPU hardware repurchase guarantees added a further $6.6 billion. Alphabet consequently became the first individual company to pass $1 trillion in total off-balance-sheet commitments, which reached $1.059 trillion. The smallest expansion was Microsoft’s: $35 billion, or 6%. That comparison comes with a qualification, however. Microsoft’s purchase commitment figures remain based on its end-June Form 10-K disclosures, and its next annual report will most likely revise the actual figure upwards.

The fastest expansion belonged to SpaceX. From $29 billion at the end of the second quarter, its off-balance-sheet commitments climbed to $40 billion at the end of the third, before reaching $51 billion on 26 October, an increase of 174%.

What unites these categories is an obligation to spend, rather than borrowing already undertaken. A signed lease requires rent to be paid even if servers are not fully installed or computing capacity has not been sold. A purchase commitment likewise requires payment whether chips arrive on time or demand materialises. In the accounts, these obligations appear only as “commitments” in the financial statement footnotes, not as interest in the income statement or liabilities on the balance sheet. The associated cash outflows are nevertheless no less rigid than those required to service on-balance-sheet debt.

By comparison, approximately $1 trillion of actual debt and lease liabilities is recognised on the companies’ balance sheets. Their off-balance-sheet commitments are more than 4 to 1 times those on-balance-sheet liabilities.

Commitments Swell by Nearly $1 Trillion in 90 Days

Off-balance-sheet commitments reached $4.11 trillion in October, having stood at $3.12 trillion at the end of August and approximately $1.8 trillion at the end of June. That amounts to more than a doubling in just over four months. Among just the five companies whose third-quarter reports have already been filed, the total climbed to $3.05 trillion from $2.48 trillion: an increase of $577 billion, or 23%, over the quarter.

How quickly these obligations are growing matters in its own right. Credit ratings, interest costs and the bond market’s capacity to take up new issuance all constrain on-balance-sheet borrowing. Accounting treatment allows off-balance-sheet commitments to grow much more “quieter” in appearance. Where extra balance-sheet leverage would prompt a downgrade and increase funding costs, expenditure structured as long-term leases or purchase commitments imposes a considerably lower “cost” on the financial statements.

Consider Meta, where $97 billion in operating cash flow and Aa3/AA- credit ratings did not prevent CDS from reaching a record high after the earnings release. The company then set out plans to raise up to $95 billion in equity. It had already completed $37.5 billion, leaving approximately $60 billion to be executed under an at-the-market offering programme. In other words, an AA- rated company is adding off-balance-sheet commitments while the market drives its CDS higher, even as it raises equity to replenish capital. Taken together, these developments suggest that credit pricing has begun to incorporate off-balance-sheet commitments, and that Meta has elected to meet the pressure with equity rather than debt.

At Amazon, the same issue is visible through cash flow and spending. Trailing twelve-month free cash flow turned negative at $42.4 billion, against positive $14.8 billion a year earlier, as quarterly capital expenditure reached $77.26 billion. Over a single quarter, off-balance-sheet commitments also grew by $131 billion, the largest percentage increase among hyperscalers. These two accelerations in the same quarter are not coincidental: capital expenditure and off-balance-sheet commitments are separate accounting expressions of the same AI infrastructure arms race.

Rating Frameworks Produce Different Leverage Assessments

According to Goldman Sachs credit strategist Amanda Lynam, S&P incorporates off-balance-sheet leases into its leverage adjustments; Moody’s and Fitch do not. Leverage for a given company therefore depends on which framework is used. At a few hundred billion dollars of off-balance-sheet commitments, the effect of those methodological differences is limited. At $4.11 trillion, more than four times on-balance-sheet liabilities, the differences become large enough to change credit assessments.

Meta retained its AA- rating on 30 October, but S&P changed the outlook from “stable” to “negative”. The agency explicitly attributed the move to AI capital expenditure driving free cash flow below zero, drawing particular attention to future lease obligations that had risen to $355.7 billion. Although Meta’s recognised debt is relatively low, including the $355.7 billion of lease obligations through adjustments under S&P’s approach changes the trajectory of its leverage.

A further concern came from Mark Habib, who leads digital infrastructure at S&P Global. He said hyperscalers may carry substantial undisclosed AI-related debt through implicit guarantees for data centre construction or smaller AI companies, describing these arrangements as “interconnected financing structures”. Guarantees extended by US technology giants for AI data centres and chips have amounted to up to $300 billion over the past year. Rather than appearing together, they may be distributed among line items covering residual value guarantees, joint venture commitments and credit support. Meta’s $41.5 billion of residual value guarantees for Hyperion and El Paso illustrates one form; Alphabet’s energy equipment guarantees and TPU hardware repurchase guarantees illustrate another.

Earlier warnings from the Bank for International Settlements identified opaque AI financing as an aggravating factor in financial vulnerabilities, with circular financing the most prominent risk. S&P has separately identified revenue inflation caused by circular financing as one of the main sources of growing uncertainty in the AI industry.

Markets Begin to Price Off-Balance-Sheet Risk

It is the mismatch between inflexible cash outflows and uncertain revenue, rather than the fact that liabilities are “hidden”, that creates the risk. The transmission sequence runs as follows: cash receipts undershoot expectations → project cash falls short → collateral values weaken and financing conditions tighten → parent companies or guarantors supply extra funding → construction and procurement are scaled back.

Cash receipts set that sequence in motion. A project’s revenue rests on its ability to rent out computing capacity at sufficiently high rates. Lower rental rates leave inflows unable to meet lease payments and purchase commitments, so project companies come under pressure first. Falling collateral values and tighter financing follow, requiring parent companies or guarantors to supply more funding. That is the point at which commitments held off the balance sheet become cash outflows on it. A parent whose own cash flow is also contracting must then finance the additional support through further borrowing or equity issuance.

Crucially, this process has a threshold. So long as receipts remain normal, off-balance-sheet commitments amount to a footnote entry. Once a project encounters a cash shortfall, however, their inflexibility can pass the pressure back to the parent much faster than anticipated. Intel’s loss of 30% of its market value in a single day on 30 October provides the micro-level illustration. At the macro level, simultaneous shortfalls at several project companies could quickly convert part of the $4.11 trillion into actual funding requirements.

On-balance-sheet measures remain the principal inputs for mainstream valuation models, although that approach is starting to change. Following Meta’s earnings release, its CDS hit a record high; S&P cut the outlook, while Goldman Sachs publicly drew attention to differences between rating methodologies. Off-balance-sheet commitments are thus beginning to feature as a variable in their own right across credit markets, rating agencies and sell-side research.

A systemic repricing may be triggered less by the absolute amount than by the divergence between growth in operating cash flow and growth in off-balance-sheet commitments. These liabilities remain manageable future expenditure while operating cash flow covers on-balance-sheet obligations and some off-balance-sheet commitments. If cash flow growth lags behind commitment growth, however, inflexible expenditure ceases to be a “long-term arrangement” and becomes “near-term financing pressure”.

The bill represented by Intel’s 30% fall on 30 October has already come due; the $4.11 trillion represents the bill still to come. Pricing the first took the market one day. The time needed to price the second will depend on when the first transmission chain progresses from “cash receipts falling short of expectations” to “parent companies providing additional funding”.