Michael Burry, the investor famously portrayed in "The Big Short," has intensified his criticism of NVIDIA and other AI giants, revealing that he is shorting NVIDIA and Palantir.
In a post on his newly launched paid Substack newsletter on Tuesday, Burry addressed NVIDIA's recent memo to Wall Street analysts, calling it a sophistical response to claims he never made.
In an article titled "Unicorns and Cockroaches: Lucky Frauds," Burry expressed disbelief that NVIDIA's response came from the world's most valuable publicly traded company. He described the document as filled with "one straw man after another," adding that the entire memo "reads almost like a scam."
Burry, who recently closed his hedge fund to focus on writing, clarified in his post that he never suggested NVIDIA was delaying depreciation of its property, plant, and equipment (PP&E), as the company is primarily a capital-light chip designer rather than a manufacturer.
"No one cares about NVIDIA's own depreciation," he wrote. "One straw man burned down."
Burry also dismissed NVIDIA's argument about its older-generation chips still being in use, stating his concern lies in newer chips potentially becoming functionally obsolete between 2026 and 2028.
"I'm looking forward because I see issues relevant to today's investors," he wrote. "Second straw man burned down."
The investor added that NVIDIA's rebuttal was "hypocritical to the core and disappointing."
In his latest post, Burry disclosed his short positions: "I continue to hold puts on Palantir and NVIDIA, with both companies' puts to be discussed next."
One of Burry's primary concerns revolves around AI companies' depreciation accounting—how quickly they forecast asset value declines and their end-of-life valuations.
By spreading these costs over five to six years instead of three, companies can boost short-term profits and book asset values. However, Burry warned this could pave the way for massive future write-downs.
He highlighted Microsoft CEO Satya Nadella's recent interview, where Nadella admitted slowing data center construction earlier this year over concerns about overbuilding infrastructure for current-gen AI chips, as next-gen chips will require different power and cooling solutions.
"These hyperscalers have been systematically extending chip and server lifespans for depreciation purposes after investing hundreds of billions in accelerated planned obsolescence for GPUs," Burry wrote.
He suggested his depreciation comments triggered a stronger reaction than anticipated: "I got swept up in something much bigger than me."