Over $4.1 Trillion: Beware AI’s Off-Balance-Sheet Bill

30% was Intel’s brief intraday loss on 30 October. The shares touched $64.24, erasing nearly $150 billion in market value in the company’s worst single-day performance since the 1970s. Revenue guidance for Q4 came in at $15 billion to $16 billion, leaving a gap of more than $1 billion against analysts’ average forecast of approximately $17 billion. Positive cash flow is not expected this year, CFO David Zinsner explicitly stated. An incident at the Israeli factory is expected to subtract approximately $300 million from Q4 revenue. PC shipments are forecast to shrink by approximately 12% this year and another 10% in 2027. Before the results, investors had priced Intel as an AI beneficiary: from the beginning of the year to 29 October, the shares had at one stage advanced nearly 150%. The report instead showed fabrication plants absorbing cash flow and higher memory prices squeezing the core PC business. Operational errors at a single company account for Intel’s predicament. The $4.11 trillion figure compiled afterwards pointed to a problem with far broader systemic implications.

$4.11 Trillion: The Commitment Breakdown

Three categories dominate the $4.11 trillion total, none of them conventional debt: purchase commitments that cannot be cancelled, signed leases awaiting commencement, and guarantees and credit support for data centre and chip projects.

Meta ended the third quarter with non-cancellable purchase commitments of $455 billion, up sharply from $349 billion. Leases awaiting commencement reached $356 billion from $279 billion, alongside $41.5 billion in residual value guarantees for the Hyperion and El Paso joint ventures. Amazon reported unconditional purchase obligations of $254 billion, almost twice the previous $130 billion, and leases awaiting commencement of $165 billion against $137 billion previously. Commitments expanded by $131 billion over the quarter, or 41%, giving Amazon the highest percentage increase among hyperscalers. Alphabet became the first individual company with more than $1 trillion in off-balance-sheet commitments, taking its total to $1.059 trillion. Purchase commitments increased by $95 billion; leases awaiting commencement reached $131 billion from $85 billion. It also doubled energy equipment guarantees to $16.2 billion and added TPU hardware repurchase guarantees of $6.6 billion. An increase of $35 billion, or 6%, put Microsoft at the restrained end of the expansion. Its purchase commitment data still rely on the end-June Form 10-K disclosures, however. The next annual report will most likely bring an upward revision to the actual figure.

The most aggressive expansion was at SpaceX, where off-balance-sheet commitments reached $51 billion on 26 October. They had stood at $29 billion at the end of the second quarter and $40 billion at the end of the third quarter. The increase was 174%.

The distinction is between borrowing already drawn and spending already pledged. Off-balance-sheet commitments fall into the latter category. Rent becomes payable when a lease is signed, whether servers are fully installed or computing capacity has been sold. A purchase commitment similarly binds the company to payment, even if chip deliveries are late or demand fails to materialise. Financial statement footnotes are the only place these obligations appear, labelled as “commitments”; they are neither interest on the income statement nor liabilities on the balance sheet. Their cash payment requirements are nonetheless as inflexible as those of on-balance-sheet debt.

Approximately $1 trillion in actual debt and lease liabilities sits on the companies’ balance sheets. Measured against those recognised liabilities, off-balance-sheet commitments exceed a ratio of 4 to 1.

90 Days: Nearly $1 Trillion in Additional Commitments

Off-balance-sheet commitments passed through three levels in just over four months: approximately $1.8 trillion at the end of June, $3.12 trillion at the end of August and $4.11 trillion in October. The total thus more than doubled. Restricting the calculation to the five companies that have already filed third-quarter reports still shows a quarterly rise of $577 billion, or 23%, with combined commitments moving from $2.48 trillion to $3.05 trillion.

Growth at this speed sends its own signal. Borrowing on the balance sheet faces limits imposed by credit ratings, interest costs and the bond market’s ability to absorb issuance. Accounting makes growth in off-balance-sheet commitments look much “quieter”. A company facing a downgrade and higher financing costs from additional balance-sheet leverage can instead arrange spending through long-term leases or purchase commitments. The “cost” reflected in its financial statements is then much lower.

For the clearest example, look at Meta’s financing sequence. Even with operating cash flow of $97 billion and Aa3/AA- credit ratings, the company saw its CDS set a record after the earnings release. It then announced an equity fundraising plan of up to $95 billion. Of this, $37.5 billion had already been completed; approximately $60 billion remained awaiting execution through an at-the-market offering programme. Off-balance-sheet commitments are accumulating at an AA- rated company just as the market pushes its CDS higher and equity financing replenishes its capital. The combination indicates that credit pricing has started to reflect off-balance-sheet commitments. Meta’s chosen response is equity, not debt.

At Amazon, cash flow provides a second view of the pressure. Free cash flow over the trailing twelve months fell to negative $42.4 billion from positive $14.8 billion a year earlier, while quarterly capital expenditure reached $77.26 billion. Over that single quarter, off-balance-sheet commitments grew by $131 billion, the largest percentage increase among hyperscalers. Capital expenditure and off-balance-sheet commitments accelerating together is not a coincidence: they are two accounting expressions of the same AI infrastructure arms race.

$4.11 Trillion Tests Rating Methodologies

Leverage calculations differ across rating agencies, Goldman Sachs credit strategist Amanda Lynam highlighted. Adjustments for off-balance-sheet leases are made by S&P, but not by Moody’s or Fitch. A company’s reported leverage can consequently vary according to the framework applied. With commitments of a few hundred billion dollars, the methodological gap has limited influence. At $4.11 trillion, more than four times on-balance-sheet liabilities, it can change the credit assessment.

Meta’s outlook was revised on 30 October by S&P from “stable” to “negative”, with the AA- rating left intact. The explanation directly identified AI capital expenditure as pushing free cash flow below zero. It also singled out future lease obligations, which had climbed to $355.7 billion. Recognised debt at Meta is not high. Including the $355.7 billion of lease obligations in adjustments under S&P’s methodology, however, changes the course of its leverage.

Implicit guarantees present a separate concern, according to Mark Habib, head of digital infrastructure at S&P Global. Hyperscalers may have significant undisclosed AI-related debt tied to guarantees for data centre construction or smaller AI companies. He called these arrangements “interconnected financing structures”. US technology giants have extended up to $300 billion in guarantees for AI data centres and chips over the past year. Rather than sitting in one place, these may appear across line items as residual value guarantees, joint venture commitments and credit support. One example is Meta’s $41.5 billion in residual value guarantees for Hyperion and El Paso; a different category comprises Alphabet’s energy equipment guarantees and TPU hardware repurchase guarantees.

Opaque AI financing compounds financial vulnerabilities, the Bank for International Settlements has previously warned, identifying circular financing as the most prominent risk. A further warning has come from S&P: revenue inflation caused by circular financing is among the key sources of greater uncertainty in the AI industry.

$4.11 Trillion: The Market’s Risk Assessment

Being “hidden” is not what makes off-balance-sheet liabilities risky. The problem is fixed cash payment requirements matched against uncertain revenue. Pressure travels through the following chain: cash receipts miss expectations → project cash is inadequate → collateral loses value and financing conditions tighten → parent companies or guarantors supply extra funding → construction and procurement are scaled back.

Cash receipts initiate the process. A project earns revenue only if computing capacity can be rented out at sufficiently high rates. Lower rental rates leave cash inflows unable to meet lease payments and purchase commitments, so project companies feel the strain first. Falling collateral values and tighter financing follow, requiring parent companies or guarantors to inject funds. This is where off-balance-sheet commitments become on-balance-sheet cash outflows. A parent whose own cash flow is contracting must then raise more debt or equity to supply that additional funding.

A threshold separates the two states of this chain. With normal cash receipts, an off-balance-sheet commitment remains a footnote entry. A project cash shortfall, however, can send pressure back to the parent much faster than expected because those commitments are inflexible. At the company level, Intel’s loss of 30% of its market value in a single day on 30 October offers an example. 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 main inputs for mainstream valuation models. Signs of a change are now emerging: a record post-earnings CDS level at Meta, an outlook downgrade by S&P, and public scrutiny of inconsistent rating methodologies from Goldman Sachs. Off-balance-sheet commitments are starting to enter credit markets, rating agencies’ assessments and sell-side research as a variable in their own right.

A systemic repricing may hinge less on the absolute total than on how commitment growth compares with operating cash flow growth. Future off-balance-sheet spending remains manageable while operating cash flow covers on-balance-sheet obligations and a portion of off-balance-sheet commitments. If cash flow growth lags commitment growth, fixed payment requirements turn a “long-term arrangement” into “near-term financing pressure”.

One bill has already fallen due: Intel’s 30% drop on 30 October. The $4.11 trillion represents the bill still to come. Pricing the first took the market one day. The time needed for the second depends on when the first transmission chain progresses from “cash receipts falling short of expectations” to “parent companies providing additional funding”.