Hyperscalers (Microsoft, Meta, Oracle, Amazon) outspend operating cash flows to fund AI infrastructure capex; combined capex eyed above $1 trillion.
The AI infrastructure race is pushing top technology firms—Alphabet, Amazon, Meta, Microsoft, and Oracle—toward combined annual spending exceeding $1 trillion from 2027 onward, according to an ICICI Bank Research report. These five hyperscalers are expected to spend $729 billion on capital expenditure in 2026, rising to $1.069 trillion in 2027 and remaining above $1 trillion through 2030, reaching $1.184 trillion in 2028, $1.172 trillion in 2029, and $1.213 trillion in 2030.
This spending surge is stretching companies' balance sheets far beyond their operating cash flows. Between 2018 and 2021, hyperscaler capex represented roughly one-third of their operating cash flows. That ratio climbed to 45% during 2022–2025 and is projected to reach 90% in 2026 and 104% in 2027—meaning capex will exceed cash generation by 2027. The report attributes this divergence to rapidly advancing capital expenditure as companies ramp up AI investments, while operating cash flow growth lags because AI investments take years to generate meaningful returns.
Large upfront investments in hardware infrastructure, data centers, power generation, GPUs, and networking equipment have created a significant funding gap. To bridge it, hyperscalers are increasingly turning to debt markets. US hyperscalers borrowed $220 billion through debt instruments in 2026, a figure the report expects to rise as AI infrastructure demands grow. Most of this borrowing has been investment-grade, with hyperscalers except Oracle maintaining high-investment-grade ratings. The five firms alone are projected to exceed 5% of the investment-grade index by year-end 2026, while AI-related debt is estimated to account for around 15% of US investment-grade debt overall.
US investment-grade corporate bond issuance reached $1.7 trillion during January–July 2026 and is expected to cross $2 trillion for the full year. The report estimates that AI-related debt could account for close to a third of net new investment-grade supply in the US market.
A particular structural pressure emerges from the maturity profile of this borrowing. Approximately 80% of hyperscalers' bond issuance carries maturities exceeding five years, with 23% classified as "very long-term." Technology companies are seeking multi-year funding to finance their infrastructure projects, creating direct competition with long-term US Treasury bonds. The US Treasury's net issuance of long-term debt in the 20–30 year maturity bucket stood at around $424 billion in 2025. The report estimates total US bond supply in that maturity range at $400–500 billion annually, while AI debt issuance for similar maturities is estimated at approximately $500 billion—potentially exceeding government long-term debt supply.
The report describes this dynamic as "reverse crowding out"—a reversal of the traditional crowding-out phenomenon where government borrowing reduces capital available for private investment. Here, private-sector AI borrowing is instead putting additional pressure on sovereign debt markets for the same pool of long-term capital.
The funding pressure extends beyond the five major hyperscalers. The report estimates total AI debt issuance at $489–570 billion in 2026, compared with $220–250 billion for the core hyperscalers alone. At approximately $500 billion, AI debt supply would represent roughly 120% of the US Treasury's $424 billion net long-term maturity supply, placing downward pressure on US Treasury yields.
Hyperscalers are also diversifying their borrowing into currencies beyond the US dollar—particularly euros and Canadian dollars—which is increasing competition for debt issuance in those markets and pushing up bond yields globally.
The report expects the reverse-crowding-out phenomenon to intensify through 2027 as AI capex remains above $1 trillion and AI debt issuance peaks. However, relief is anticipated as AI investments begin generating revenue and operating cash flows fund a larger share of capital expenditure. AI borrowing could therefore peak in 2027 and moderate thereafter, reducing competition for sovereign debt and easing pressure on bond yields.