The "Source" of Massive Capital Expenditures: How High Is the "Hidden Leverage" of Tech Giants?

The balance sheets of tech giants may be far heavier than the book figures suggest.

Goldman Sachs' latest research indicates that off-balance-sheet lease commitments for AI hyperscale cloud providers such as Google, Microsoft, Amazon, and Meta have accumulated to as high as $1.5 trillion, with approximately $1 trillion not yet reflected in financial statements because execution has not yet begun.

Meanwhile, another set of data compiled by Morgan Stanley shows that purchase commitments for the aforementioned companies, plus NVIDIA and Oracle, total nearly $982 billion.

Combined, the scale of potential undisclosed financial obligations approaches $2 trillion. The core risk of these implicit liabilities lies in the fact that if data center investments fail to deliver expected returns, these contractual obligations will gradually come to light, exerting substantive pressure on the liquidity and leverage levels of the relevant companies.

Bond prices for hyperscale cloud providers have already seen a mild pullback, and market attention to their long-term solvency is rising.

Superficial Stability: Net Leverage Ratios Remain Low

From the perspective of traditional credit metrics, the financial position of hyperscale cloud providers remains solid.

Morgan Stanley data shows that the average net leverage ratio for this group is only 0.5x, lower than the 0.8x for the overall technology sector and far below the 1.8x average for US non-financial corporations. More notably, the total cash held by these companies still exceeds their book debt.

This is the logic of the optimists: Core businesses continue to generate substantial cash flow, debt scales are relatively limited, and credit ratings are generally high. Even if returns on data center investments fall short of expectations, equity investors will bear the brunt first, rather than creditors. This judgment holds certain rationality within the traditional financial framework.

However, as financing methods become increasingly diverse, the stable picture presented by these metrics is being eroded by a series of commitments that remain off the balance sheet.

Creative Financing: Starting with Meta's "Beignet" Model

The precedent for this off-balance-sheet financing originated from the structured financing solution Meta designed for its "Hyperion" data center in Louisiana.

Meta and Blue Owl jointly established a venture company, Beignet, with the latter developing and holding the data center assets. Meta holds only a 20% equity stake in Beignet but has committed to leasing Hyperion for at least 20 years.

It was this lease guarantee that enabled Beignet to issue amortizing bonds totaling $27 billion—setting a record for bond issuance.

Although Meta effectively bears the repayment obligation, this debt does not appear on Meta's own balance sheet.

This "best of both worlds" structure promptly prompted imitation by other hyperscale cloud providers, leading to the emergence of various forms of lease arrangements.

Accounting Loopholes: $1 Trillion Hidden in Footnotes

The reason these commitments can remain "invisible" lies in the specific provisions of US Generally Accepted Accounting Principles (US GAAP).

Under GAAP ASC 842, lease payment obligations must be recognized on the balance sheet only after the lease has actually commenced, at which point right-of-use assets are also recognized. Lease commitments that have not yet commenced are disclosed only in the footnotes of financial reports.

Goldman Sachs analysts aggregated approximately $1 trillion in "not yet commenced" lease commitments by systematically reviewing these footnotes. This figure represents a significant increase from the total lease commitments of approximately $200 billion five years ago and is further expanded from Goldman Sachs' estimate of $750 billion last month.

Notably, Fitch and Moody's also do not include not-yet-commenced leases in their rating metrics; only S&P has adopted a more conservative approach, incorporating leases with "substantial debt characteristics" into its considerations.

Goldman Sachs analysts stated bluntly: From a credit perspective, the above treatment may underestimate corporate leverage levels and future liquidity needs, as these obligations will eventually be recognized and contract payments will come due.

Purchase Commitments: Another Near-$1 Trillion Gap

In addition to leases, hyperscale cloud providers have made substantial purchase commitments for computing power, chips, equipment, and electricity.

These commitments also do not appear on the balance sheet as traditional debt, yet they constitute substantive financial obligations that must be fulfilled in the future.

Statistics from Morgan Stanley credit analysts show that as of the end of the first quarter this year, purchase commitments for Google, Microsoft, Amazon, NVIDIA, and Oracle totaled $982 billion.

Added to the $1 trillion in lease commitments not included in financial statements as calculated by Goldman Sachs, the total potential implicit financial obligations approach $2 trillion.

Market Reaction: Bond Prices Show Mild Pressure

These concerns have begun to reflect in the bond market.

Taking Beignet bonds as an example, their yield has risen from a low of 5.65% at issuance last autumn to approximately 6.95% currently, with prices showing a noticeable decline.

However, analysts point out that part of the pressure stems from the passive transmission of the overall rise in US Treasury yields and does not yet constitute a strong signal of credit deterioration.

But the core logic of the market is becoming increasingly clear: whether all these implicit burdens can be absorbed ultimately depends on whether capital expenditures on data centers can deliver expected returns.

Currently, hyperscale cloud providers are still in a cycle of massive expansion, with limited visibility into investment returns. For credit investors, how to incorporate these nearly $2 trillion in potential obligations into valuation frameworks beyond traditional metrics is becoming an unavoidable imperative.