The balance sheets of technology giants may be far more burdened than their reported figures suggest.
Goldman Sachs' latest research reveals that off-balance-sheet lease commitments for AI hyperscalers, including Alphabet, Microsoft, Amazon, and Meta Platforms, Inc., have accumulated to a staggering $1.5 trillion. Approximately $1 trillion of this amount is not reflected on financial statements because the leases have not yet commenced.
Simultaneously, data from Morgan Stanley indicates that purchase commitments for these companies, along with NVIDIA and Oracle, total nearly $982 billion.
Combined, these potential undisclosed financial obligations approach $2 trillion. The core risk of these hidden liabilities is that if data center investments fail to deliver expected returns, these contractual obligations will gradually surface, exerting substantial pressure on the liquidity and leverage levels of the affected companies. Bond prices for these hyperscalers have already experienced a modest pullback, and market attention on their long-term solvency is intensifying.
Superficial Stability: Net Leverage Remains Low
From the perspective of traditional credit metrics, the financial health of hyperscalers still appears solid. Morgan Stanley data shows that the average net leverage ratio for this group is only 0.5 times, lower than the 0.8 times for the broader tech sector and far below the 1.8 times average for U.S. non-financial companies. More notably, the total cash held by these companies still exceeds their on-book debt. This underpins the optimistic view: core businesses generate substantial cash flow, debt levels are relatively limited, and credit ratings are generally high. Even if data center returns fall short, equity investors would bear the brunt before creditors. This assessment holds some validity under traditional financial frameworks. However, as financing methods diversify, the robust picture presented by these metrics is being eroded by a series of commitments that exist outside the balance sheet.
Creative Financing: The Meta "Beignet" Model
A precedent for this off-balance-sheet financing stems from Meta Platforms, Inc.'s structured financing plan for its "Hyperion" data center in Louisiana. The company formed a joint venture, Beignet, with Blue Owl, which developed and owns the data center asset. Meta holds only a 20% equity stake in Beignet but has committed to leasing Hyperion for at least 20 years. This lease guarantee enabled Beignet to issue amortizing bonds totaling a record $27 billion. Although Meta effectively bears the repayment obligation, this debt does not appear on its own balance sheet. This "win-win" structure has prompted other hyperscalers to follow suit, with various customizable lease arrangements emerging.
Accounting Loophole: $1 Trillion Hidden in Footnotes
The ability to conceal these commitments stems from specific U.S. GAAP rules. Under GAAP ASC 842, lease payment obligations must be recognized on the balance sheet only after the lease has actually commenced, at which point a right-of-use asset is simultaneously recorded. Lease commitments that have not yet begun are only disclosed in the footnotes of financial reports. Goldman Sachs analysts compiled this $1 trillion in "not-yet-started" lease commitments by meticulously reviewing these footnotes. This figure is a sharp increase from total lease commitments of about $200 billion five years ago and has expanded further from Goldman's estimate of $750 billion last month. Notably, Fitch and Moody's also exclude not-yet-started leases from their rating metrics, while only S&P takes a more conservative approach by including leases with "substantial debt characteristics." Goldman Sachs analysts state that, from a credit perspective, this treatment may underestimate a company's leverage and future liquidity needs, as these obligations will eventually be recognized and contract payments will come due.
Purchase Commitments: Another Nearly Trillion-Dollar Gap
Beyond leases, hyperscalers have made significant purchase commitments for computing power, chips, equipment, and electricity. These commitments do not appear as traditional debt on the balance sheet but constitute substantial financial obligations that must be fulfilled in the future. According to Morgan Stanley credit analysts, as of the end of the first quarter, the combined purchase commitments of Alphabet, Microsoft, Amazon, NVIDIA, and Oracle totaled $982 billion. Adding this to the $1 trillion in off-balance-sheet lease commitments identified by Goldman Sachs brings the total potential hidden financial obligations to nearly $2 trillion.
Market Reaction: Bonds Show Moderate Pressure
These concerns are already reflected in the bond market. For instance, yields on Beignet bonds have risen from a low of 5.65% at issuance last fall to around 6.95% currently, with prices declining noticeably. However, analysts note that some of the pressure is passively transmitted from the rise in overall U.S. Treasury yields, not yet constituting a strong signal of credit deterioration. But the market's core logic is becoming increasingly clear: whether all these hidden burdens can be absorbed ultimately depends on whether the capital expenditures on data centers can deliver on expected returns. Currently, hyperscalers remain in a massive expansion cycle with limited visibility on investment returns. For credit investors, incorporating these nearly $2 trillion in potential obligations into their valuation framework, beyond traditional metrics, is becoming an unavoidable challenge.
Comments