Michael Burry Warns of Systemic Contagion Risk in $573 Billion AI Financing Web

Over the past 12 months, the physical infrastructure of the artificial intelligence boom has been underwritten by a $573 billion web of interconnected debt, leases, and financial guarantees. That figure, mapped across 26 disclosed financings, forms the basis of a new warning from "The Big Short" investor Michael Burry, who argues that the market is ignoring the systemic contagion risk embedded in how the AI expansion is being funded.[1][2]

Burry’s latest commentary, published late Sunday, centers on an analysis from Ares Management’s Fall 2026 Alternative Credit Newsletter. The report details how the $573 billion spans five specific categories of financial instruments: corporate bonds, data-center construction loans, hardware leases, GPU-backed borrowings, and residual-value guarantees.[1][2]

The financing network ties together virtually all the major players in the current AI expansion, including Meta, Oracle, Microsoft, Amazon, Google, Nvidia, OpenAI, and Anthropic. Within this matrix, a single technology company might act as a borrower in one transaction, a tenant in another, and a financial guarantor in a third, creating a highly interdependent capital structure.[2]

The vulnerability, according to Burry and the Ares analysis, is that every node in this network rests on the assumption that AI spending will continue to grow uninterrupted. As the Ares report states, "Every transaction, every obligation, every backstop, every guarantee, and every agreement" in this network relies on sustained revenue growth. If revenue disappoints and corporate boards begin redirecting capital, the financial pressure could spread rapidly through these connected deals.[1][2]

A central mechanism driving this risk is the rapid depreciation of the underlying hardware. Burry highlighted data showing that the residual value of Nvidia's H100 chips has dropped by 51% over three years. Because many of these loans are collateralized by the hardware itself, lenders increasingly require residual-value guarantees from manufacturers and cloud providers to offset the depreciating assets.[2]

According to Burry, manufacturers and cloud service providers offer these safeguards "not because they want to, but because they have no choice." He described the rapid depreciation of the hardware as "very, very real," noting that the financial engineering required to mask it is becoming increasingly strained.[2]

Burry has previously argued that hyperscale vendors are stretching the useful life of these chips on their balance sheets far beyond a realistic two-to-three-year replacement cycle. He estimates that this accounting approach could cause the industry to understate depreciation by roughly $176 billion between 2026 and 2028, potentially overstating earnings for Oracle by 27% and Meta by 21%.[2][3][4]

The ultimate holders of much of this debt are increasingly found in the insurance industry. With U.S. insurers managing roughly $9 trillion in investment assets, their balance sheets have become a primary funding source for digital infrastructure. Burry noted that he has been gathering insurance-industry data for months to map exactly where these asset-backed securities reside.[1][2]

The demand for this infrastructure remains intense, driven by the relentless pace of AI model development. Research groups and tech firms continue to release increasingly complex systems—such as Hugging Face's newly detailed Holo4, designed to power generalist computer-use agents—which require massive, sustained compute power to train and operate.[5]

The immediate test for this financing web will be the upcoming earnings cycles for the hyperscalers and major AI developers. If the capital expenditures required to train the next generation of models begin to outpace the revenue generated by enterprise AI adoption, the structural resilience of this $573 billion network will be tested in real time.[1][3]