Morgan Stanley reports that AI capex now exceeds cash flow, with tech giants' off-balance-sheet commitments surpassing $3.1 trillion.

Artificial intelligence investment is decisively shifting from an era of self-funded cash generation to one driven by external financing. According to Morgan Stanley’s latest research report, hyperscaler cash capital expenditure is projected to exceed $1.2 trillion by 2027, while total operating cash flow over the same period will amount to only around $1 trillion. This structural funding gap has compelled technology giants to mobilize leases, bonds, equity financing, and customer prepayments at unprecedented scales, pushing cumulative off-balance-sheet commitments and guarantees beyond $3.1 trillion.

The report arrives just ahead of Nvidia’s fiscal 2027 second-quarter earnings announcement, which market participants increasingly view as a system-level stress test rather than a single-company fundamental review. For years, Microsoft, Alphabet, Amazon, and Meta primarily funded data center construction and GPU procurement through robust internal cash flows. However, as AI investment scales intensify, internal liquidity can no longer fully cover capital expenditures. Morgan Stanley estimates that hyperscalers are currently reinvesting more than 40% of revenue into AI capex. Even with high-margin legacy businesses in search, advertising, software, and cloud computing, this intensity strains operating cash flow alone. Consequently, Amazon and Alphabet reported negative free cash flow in the second quarter of 2026, with Meta expected to follow suit next quarter.

The underlying dynamic is straightforward: the AI race forces technology companies to simultaneously construct data centers and power infrastructure while securing advance supply agreements for GPUs, memory, and networking equipment. Because facilities typically take years to build, companies must lock in hardware deliveries long before completion to avoid idle capacity, generating a growing pipeline of future cash obligations. Regulatory filings reveal that hyperscalers’ undiscounted future commitments have already surpassed $2.7 trillion—roughly equivalent to three years of current operating cash flow. When including Nvidia, Broadcom, and various guarantee structures, total off-balance-sheet commitments and guarantees exceed $3.1 trillion.

Morgan Stanley categorizes this new funding architecture into four distinct layers. The first relies on operating leases and special purpose vehicles (SPVs). Rather than issuing direct corporate debt, tech firms now utilize SPVs to raise capital from private credit markets, signing long-term leases while providing investment-grade credit support for project financing. This structure defers immediate construction costs and temporarily keeps associated debt off corporate balance sheets. To date, hyperscalers have committed to approximately $1.091 trillion in lease payments that have not yet commenced: Microsoft accounts for roughly $329 billion, Meta approximately $279 billion, Oracle about $261 billion, Amazon around $137 billion, and Alphabet approximately $85 billion. Concurrently, advance purchase commitments disclosed by hyperscalers, Nvidia, and Broadcom have reached roughly $1.7 trillion.

The second layer involves traditional bonds and on-balance-sheet debt. Major tech companies began meaningfully increasing debt issuance in late 2025, accelerating through 2026. Combined short-term and long-term debt, alongside operating and finance lease liabilities, now total approximately $770 billion across the group. Their footprint in U.S. fixed income is expanding rapidly; hyperscaler bond issuance accounted for just 2% of non-financial investment-grade corporate bond supply in 2025, but climbed to 19% year-to-date in 2026. As a result, AI investment returns will face increasingly stringent cost-of-capital constraints.

The third layer consists of reduced share buybacks and direct equity issuance. Alphabet provides the clearest example: after historically repurchasing more than $60 billion in stock annually, the company suspended buybacks entirely and issued approximately $50 billion in equity during the second quarter of 2026. Together, these actions freed more than $110 billion in cash for AI infrastructure development. This pivot also means that after reducing its outstanding share count by roughly 13% from decade-high levels, Alphabet shareholders are once again confronting dilution. Microsoft remains the only covered hyperscaler to maintain both its buyback program and debt discipline.

The fourth layer encompasses customer prepayments and semiconductor manufacturer credit support. Oracle recently disclosed approximately $4.6 billion in customer prepayments, classified accounting-wise as deferred revenue. When clients pay more than a year ahead of service delivery, these arrangements functionally operate as financing instruments. More symbolically, chipmakers like Nvidia and Broadcom have begun backing chip-leasing SPVs that issue debt to purchase GPUs and lease them to AI labs lacking investment-grade ratings. Manufacturers provide residual value support, agreeing to cover shortfalls if lessees default and secondary chip sales fail to repay creditors.

Nvidia’s deepening involvement in infrastructure financing has drawn particular scrutiny. Earlier this month, the company announced partnerships with Goldman Sachs, BlackRock, and Apollo Global Management to deploy $500 billion toward global AI computing infrastructure. Nvidia also committed up to $105 billion to underwrite a data center campus in Ohio slated for OpenAI leasing. Market observers warn these structures risk circular financing: Nvidia funds AI developers, who may subsequently use those proceeds to purchase Nvidia hardware, creating a closed loop of capital and revenue within the ecosystem.

Sean Burke, senior equity analyst at Janus Henderson, which holds a substantial Nvidia position, stated: “They need to discuss these two large partnerships or agreements in great detail and, to some extent, alleviate market concerns about circular financing.” Nvidia CEO Jensen Huang has consistently defended the strategy, arguing that AI companies remain in a rapid expansion phase and that direct capital and infrastructure support is necessary to alleviate computing capacity shortages. Yet investors remain focused on whether these financing arrangements and guarantees effectively isolate corporate balance sheets from systemic leverage risks.