Bernstein argued with Michael Burry: how long do Nvidia chips really live?

Investors are debating whether big tech companies should write off Nvidia's AI chips at six years. Michael Burry, the prototypical character from the movie "The Down Game," believes the lifespan of these chips is shorter, while Bernstein analyst Stacey Razgon argues that a six-year amortization is justified. Burry's view fits in with his warning about the AI market overheating: according to his data, stock indexes usually peak in the middle of investment booms and then - after another year or two of rising capital spending - there is a sharp collapse.

A dispute has erupted between prominent investor Michael Burry and Bernstein analyst Stacey Razgon over whether tech companies are correctly accounting for the cost of Nvidia's artificial intelligence chips that power their data centers, Barron's writes.

Burry, known for accurately betting against the mortgage market before the 2008-2009 crisis and becoming the prototypical character in the movie "The Downgrade Game," said on Nov. 10 that companies overestimate the economic life of hardware. He points out that graphics processing units (GPUs) are being written off at six-year amortization periods, while the actual life of the chips is shorter, in his opinion.

Depreciation allows the cost of an asset to be spread over the period of its use, rather than written off all at once. The longer the period, the lower the annual cost and the higher the profit - which is what Burry believes distorts financial statements.

Overclock disagrees with that assessment. "Amortization at most major cloud providers looks reasonable," he said in a Nov. 17 note, emphasizing that GPUs can be profitable over six years.

According to the analyst, even the Nvidia A100 - graphics gas pedals released about five years ago and in service for about the same amount of time - still provide good margins. Industry sources confirm that such chips often function for six to seven years or even longer.

"With limited processing power, demand for the A100 remains strong," added Overclocked. GPU cloud vendors have almost no spare resources left on the A100.

Executives at CoreWeave, which operates one of the world's largest clouds built specifically for AI and high-performance computing tasks, also noted this month that interest in older GPUs remains. The company was able to renew its contract to use the H100 chips at virtually the same price - a variance of just 5% - even though the model is three years old.

Microsoft CEO Satya Nadella explained why GPUs last longer than many assume. "First you use them for training, then you use them for data generation, then you use them for inferencing," he said on a recent podcast, The Dwarkesh Podcast. Inference is getting answers based on models that have already been trained. "It's not hardware that works for one task and then gets sent to the shelf," Nadella noted.

Burry published a new warning about the overheating artificial intelligence market on Nov. 16, Business Insider writes . He presented a graph of capital spending by S&P 500 companies over 35 years, adjusted for depreciation and U.S. GDP to show the share of new investment in the economy.

According to Burry, stock indexes historically peak in the middle of investment booms, while capital spending continues to rise for another year or two and then collapses sharply. His chart highlights the peaks of the Nasdaq before the dot-com bubble burst, the S&P 500 before the mortgage crisis and the S&P 500 Energy before the collapse in oil prices.

Burry also points to the Nasdaq 100's record high of about 26,000 points this quarter, suggesting that the market may be near the top of the current investment cycle. His conclusion: stocks may have already peaked, even though OpenAI, Meta and others are just about to invest trillions in AI infrastructure.

If the pattern repeats, the AI investment boom could last a maximum of two years, followed by a correction and possibly a recession, Burry said.

Burry attached a Lord of the Rings meme to the post to emphasize how investors have become blindly optimistic about the AI boom and are ignoring the warning signs.

This article was AI-translated and verified by a human editor