The artificial intelligence infrastructure arms race is pushing the world's largest technology companies into unprecedented financial territory, where operational cash flow alone can no longer sustain the pace of investment. Tech giants are turning to every available tool—leasing structures, corporate debt issuance, share buyback cuts, and even equity offerings—to bridge a funding gap that is reshaping the global capital markets.
The scale of off-balance-sheet commitments tied to this buildout has now surpassed $3.1 trillion, revealing a true financial risk picture much deeper than what standard earnings reports suggest. This shift marks a definitive departure from the era of self-funded growth, forcing a fundamental reassessment of how the industry's future is being financed.
According to a recent report from Morgan Stanley, hyperscale cloud providers are projected to see cash capital expenditures exceed $1.2 trillion by 2027, while total operating cash flow during the same period is estimated at just around $1 trillion. That gap is already materializing, with both Amazon and Google having seen their free cash flow turn negative in the second quarter of 2026, and Meta expected to follow suit in the coming quarter. This inflection point signals that the era of self-sustaining cash generation is over, making external financing the new engine of AI growth. The scale and complexity of this funding push have put the entire market on edge. The same report highlights that when combining hyperscalers with chip giants Nvidia and Broadcom, disclosed off-balance-sheet commitments and guarantees now exceed $3.1 trillion—encompassing leases, purchase obligations, guarantees, and various credit support structures. At the same time, accounting treatments that may understate capital spending and overstate free cash flow are drawing increased investor scrutiny, raising the critical question of whether returns from new AI infrastructure can truly cover the rising cost of capital.
The Breaking Point: When Investment Outpaces Self-Generation
The Morgan Stanley analysis reveals that hyperscalers are now reinvesting more than 40% of their revenue into AI capital expenditures—a level of intensity that outstrips what their operational cash flows can sustainably support. Despite the hefty profit margins in their legacy businesses, these companies can no longer rely solely on internally generated funds to maintain their current investment trajectory. The timeline of this transformation is accelerating. At the end of 2025, hyperscalers began issuing bonds; by 2026, both debt and equity financing have accelerated sharply. Alphabet, for instance, has cut its annual share repurchases from over $60 billion to zero, and issued $50 billion in new stock in the second quarter of 2026. Together, these moves have freed up more than $110 billion for AI infrastructure, but they come at a price—shareholders are starting to feel the dilution. This is a notable reversal for a company that had reduced its share count by 13% over the past decade.
Morgan Stanley's data shows that on-balance-sheet debt and lease liabilities for hyperscalers have risen to a combined $770 billion. Furthermore, hyperscalers' share of the investment-grade non-financial corporate bond market has jumped from just 2% in 2025 to 19% year-to-date in 2026, underscoring how critical debt financing has become to sustaining AI capital expenditures.
The Hidden Debt: Unpacking the $3.1 Trillion Off-Balance-Sheet Risk
Beyond the on-balance-sheet liabilities, an even larger risk lurks in the shadows. The Morgan Stanley report reveals that hyperscalers are enabling suppliers and data center developers to secure financing and begin construction before any payment is due or liability is recorded. They do this through lease guarantees, purchase commitments, and various credit support mechanisms. Specifically, hyperscalers—along with Nvidia—have committed to a staggering $1.1 trillion in future lease payments (on an undiscounted basis) that have not yet commenced. Microsoft accounts for $329 billion of this, Oracle $261 billion, Meta $279 billion, Amazon $137 billion, and Google $85 billion. Meanwhile, combined disclosed purchase commitments from Nvidia, Broadcom, and the hyperscalers have reached $1.7 trillion, with Google alone responsible for $707 billion. Morgan Stanley believes the core logic of these structures is sound: the investment-grade credit backing of these cloud giants allows special purpose vehicles to borrow from private credit markets at lower costs, and then the hyperscalers lease the completed facilities. However, the bank warns that while these contracts hold strategic value until supply and demand normalize, an early market balance could leave companies facing the risk of paying for excess capacity or being forced into burdensome renegotiations.
Accounting Fog: Lowballing CapEx, Inflating Free Cash Flow
The report specifically points out that prevailing accounting methods make it difficult for investors to accurately gauge the true financial burden of AI infrastructure construction. Microsoft provides the most striking example. Last quarter, the company announced it was extending the useful life of its data centers from 15 to 25 years. This change caused a significant reclassification of many data center leases from finance leases to operating leases. Because Microsoft includes finance lease capital expenditures but excludes operating leases when calculating free cash flow, this accounting estimate change artificially makes its capital expenditure look lower and its free cash flow look higher—even though the economic substance hasn't changed, and the leases still function like debt-financed data center construction. The SPV structure also creates an accounting blind spot. Both Meta and Google have disclosed that their data center SPVs are not consolidated during the construction period, on the grounds that neither believes it is the primary beneficiary—meaning they feel they lack the power to direct the SPV's most significant economic activities. Morgan Stanley cautions that this judgment is not static; as the likelihood of providing residual value support rises, the consolidation conclusion could change at any time, potentially impacting reported leverage ratios and overall balance sheet presentation.
Innovative Financing: Customer Deposits and Chipmakers Lending Their Balance Sheets
As traditional funding channels become saturated, more creative structures are emerging. Oracle has pioneered the use of customer prepayments as a source of capital expenditure funding. In its most recent quarter, Oracle disclosed receiving $4.6 billion in customer prepayments earmarked for capital spending. While these prepayments are recorded as deferred revenue on the balance sheet, their nature is closer to debt financing. Payments are received more than a year before revenue is recognized, and Oracle is required to recognize interest expense based on its incremental borrowing rate. With Oracle's 10-year and 30-year bond yields at approximately 6.9% and 7.9% respectively, the interest cost on customer prepayments is likely comparable. Chipmakers are also stepping in to provide direct financing to their customers through SPVs. Both Broadcom and Nvidia have announced the creation of chip-financing SPVs. These vehicles would issue bonds to purchase chips, which are then leased to unrated AI labs. Both companies are providing residual value guarantees to cover scenarios where a lease defaults and chip resale proceeds are insufficient to repay bondholders. Broadcom's announced chip-leasing facility can support capital expenditures for up to 20 gigawatts of computing power, and Nvidia has similar structures underway. These arrangements enable unrated AI labs to secure chips at funding costs close to those of investment-grade suppliers. Morgan Stanley notes that when chipmakers sell to these SPVs, they may need to split revenue between the chip sale and the residual value guarantee. The guarantee liability should be measured at fair value at the time of sale, and if the guarantee is never exercised, the associated gains are typically not recorded as operating revenue.
What to Watch: The Core Questions Ahead
Morgan Stanley ultimately boils the issue down to a core principle: as each new round of commitments drives up financing costs, the return on new AI infrastructure investment must be sufficient to cover the rising cost of capital. If not, the entire premise of the buildout will need to be reconsidered. The bank identifies four key areas to monitor closely. First, the comparability of free cash flow figures has been severely compromised by differences in accounting treatments, making it essential to look through lease classifications and SPV structures to restore the real picture of capital expenditures. Second, continued reductions in share buybacks among hyperscalers and the potential for further share issuance will have a material dilutive impact on earnings per share. Third, hyperscalers' share of supply in the investment-grade bond market has surged from 2% to 19%, and the pressure for credit spreads to widen is building. Fourth, the vast off-balance-sheet commitments of $3.1 trillion, should they trigger consolidation requirements or face a supply-demand reversal, would deliver a major shock to the reported leverage of the companies involved. This massive expansion of AI infrastructure financing has already moved well beyond what tech balance sheets can comfortably absorb, and it is reshaping the capital structure of the entire technology industry at an unprecedented speed.