The Fuse of the AI Bubble
It may look like a cash-rich, steady performer—a safe asset—but its books show $27.8 billion to $33 billion in accounts receivable: chips have shipped, but payment hasn’t come in, and some major customers are even using the same GPUs as collateral for loans. Whether a highly leveraged infrastructure company makes its debt payments on time could determine Nvidia’s share price and ripple through retirement savings and employment.
CoreWeave’s Debt Bomb
Twelve hours before default, CoreWeave’s operations center in Livingston, New Jersey, was still running late Sunday night. CoreWeave went public that same year, reaching a valuation of tens of billions of dollars, but its debt-to-equity ratio was 5.27 last quarter—meaning it had more than $5 in debt for every $1 of equity. The company, originally called Atlantic Crypto, was founded in 2017 by three commodities traders and initially used GPUs to mine Ethereum. After Ethereum abandoned mining in 2019, the company pivoted to renting out GPU clusters, power, and cooling facilities, becoming a backbone of generative AI infrastructure and completing one of the largest U.S. tech IPOs since 2021.
GPU Collateral Loops and Customer Concentration
The AI infrastructure industry widely relies on highly leveraged cycles: borrowing billions of dollars to buy Nvidia GPUs, renting them out to AI companies, then using those same GPUs as collateral to borrow more. GPUs are both the product and the security for the debt, and the model depends on GPUs retaining their value and rental income continuing to flow. The high prices of H100s during peak scarcity were built into loan agreements, leverage ratios, and financial models. Nvidia's revenue is also highly concentrated, with two anonymous customers accounting for 39% of quarterly revenue. The market speculates that they are original design manufacturers or systems integrators such as Quanta and Dell, with chips changing hands through intermediaries, meaning Nvidia may not know the end users. The chips often ultimately end up in data centers like CoreWeave.
The AI Revenue Gap
Sequoia Capital has identified an approximately $600 billion gap between AI industry infrastructure spending and actual revenue. AI capital expenditure alone is expected to exceed $700 billion in 2026. Microsoft, Google, Meta, Amazon, Oracle, and others are all expanding, but at the time, most AI companies did not have enough revenue to justify the enormous spending. CoreWeave sits at the center of this ecosystem. After the market opens, analysts everywhere will see the numbers that CoreWeave employees have spent the entire weekend poring over.
Day One: Debt Default
The market was calm in the early hours after opening on Monday, then a scheduled CoreWeave debt payment failed to appear in the bank clearing system. In the third quarter after its IPO, the company's interest expense reached $311 million, doubling in a year; its IPO fundraising target had been cut from $2.7 billion to $1.5 billion. Trading desks and analyst reports reacted quickly: CoreWeave fell 11% before lunch, while Nvidia fell 4%, wiping out an estimated $204 billion to $216 billion in market value. The market initially saw this as an isolated incident, but the value of GPU collateral became the key issue: H100s were worth about $50,000 when the loans were made, while their estimated value on the secondary market might be only $5,000 to $10,000—a decline of about 90%. In the afternoon, a private credit firm holding a large amount of CoreWeave debt instructed its lawyers to initiate margin-call proceedings.
Day Two: Servers Repossessed
Early Tuesday morning, a third-party asset management team hired by the lenders entered a New Jersey data center to repossess the Nvidia H100 servers pledged as collateral, as authorized by legal documents. The servers had to be inventoried and tagged, powered down, and have their fiber-optic and copper cables disconnected in sequence before being moved with specialized equipment. Fully loaded racks weigh hundreds of pounds, and a single system can be worth millions of dollars, so the removal process took several days. By 9 a.m., thousands of servers had been taken offline in New Jersey alone. AI companies that depended on CoreWeave's computing power lost resources already factored into their revenue forecasts and product roadmaps, disrupting or shutting down services such as legal document review tools and AI marketing platforms.
Pressure on Nvidia's Accounts Receivable
Nvidia has $27.8 billion to $33 billion in accounts receivable on its books. This revenue has already been included in its financial statements and used to support its valuation, but the cash has yet to arrive; essentially, it represents customers' promises to pay. Two anonymous customers account for 39% of Nvidia's quarterly revenue, and one of them has just defaulted on a payment. Nvidia's CFO spoke with investors, choosing their words carefully, but the core message, stripped of hedging, was not encouraging.
Day Four: The S&P 500 Plunges
On Wednesday, ordinary index fund investors found the S&P 500 down 6.4% early in the morning session, with the decline reaching 8.1% by midday. The S&P 500's top