NVIDIA (NVDA) sent a memo to Wall Street analysts last weekend, firmly denying any involvement in "supplier financing"—a controversial practice where suppliers invest in or lend to their own customers.
However, prominent short sellers Jim Chanos and Michael Burry remain unconvinced.
The 7-page document refutes claims that NVIDIA artificially inflated revenue by investing in customers. The memo was issued in response to allegations made by a little-known Substack blogger last week, who accused the $5 trillion AI chip giant of employing a "round-trip financing model" to boost sales. The blogger drew parallels between NVIDIA and infamous accounting fraud cases during the dot-com bubble, such as Enron and Lucent.
Enron became notorious for manipulating financial data and using off-balance-sheet debt to hide losses in its broadband business. Meanwhile, Lucent, an internet infrastructure supplier, aggressively invested in and extended loans to struggling telecom clients—who then used the funds to purchase Lucent equipment they couldn’t otherwise afford. When the dot-com bubble burst, these telecom startups defaulted, forcing Lucent to write off billions in revenue.
Chanos, who famously predicted Enron’s collapse, believes the comparison between NVIDIA and Lucet "has some merit."
"NVIDIA is funding money-losing companies specifically so they can order its chips," Chanos told Yahoo Finance.
NVIDIA has made significant investments in several of its own customers, including ChatGPT developer OpenAI (OPAI.PVT, private), Elon Musk’s xAI (XAAI.PVT, private), and AI cloud service providers like CoreWeave (CRWV) and Nebius (NBIS)—drawing close scrutiny from Wall Street.
The memo obtained by Yahoo Finance states: "NVIDIA is fundamentally different from historical accounting fraud cases because our core business is built on solid economic foundations, our financial reporting is complete and transparent, and we place the highest value on our reputation for integrity."
The company further clarified: "Unlike Lucent, NVIDIA does not rely on supplier financing agreements to drive revenue growth." NVIDIA emphasized that its customers typically pay for chips within an average of 53 days, whereas traditional supplier financing involves multi-year repayment terms.
Michael Burry, known as "The Big Short" for predicting the 2008 housing crash, took a harsher stance than Chanos last week on X (formerly Twitter). He labeled NVIDIA as one of several AI companies with "questionable revenue recognition," attributing the issue to its customer investments.
**"The Real Achilles’ Heel of the AI Tech Market"**
Beyond supplier financing, Chanos warns that debt in the AI sector poses another major risk. He noted that, similar to Enron, some NVIDIA clients—like Meta (META) and xAI—are financing chip purchases with off-balance-sheet debt, while others, such as Anthropic (ANTH.PVT, private), rely on traditional debt financing.
"Layering massive credit and highly complex financial structures on top of these money-losing entities—that, to me, is the real Achilles’ heel of the AI tech market," Chanos told Yahoo Finance on Tuesday.
While accounting concerns may artificially inflate AI demand, both short sellers argue that the core issue is simpler: major tech firms are racing to invest billions in AI data centers before demand materializes, leading to oversupply.
Burry, in his new Substack newsletter "Cassandra Unchained," likened the current AI market to the dot-com bubble, warning of a "supply glut crisis with demand nowhere near expectations." In other words, there are too many chips, servers, and data centers, while actual enterprise and consumer demand for AI applications falls far short.
NVIDIA, however, maintains that the market is accelerating. In its latest earnings report, the company described AI chip demand as "record-breaking" and dismissed bubble concerns. On Tuesday, NVIDIA also claimed its technology is "a generation ahead of competitors"—though its stock rebounded Wednesday after an earlier dip due to rising competition from Google’s AI chips.
Chanos remains wary of unchecked AI expansion amid uncertain demand: "If we eventually realize that by 2027 or 2028, we don’t need nearly as many data centers or chip capacity as currently planned, cancellations could follow. That’s a major risk, but few are paying attention to it yet."
