At the end of November, Michael Burry, the famous investor who predicted the collapse of the bubble real estate agency in 2008, dealt a blow to the real estate industry artificial intelligence (IA) by denouncing that the markets’ perception of the technology fashion is distorted.
At the center of his criticism is Nvidiathe giant of the chips that thanks to the fever of investments that surrounds the IA has managed to become the most valuable company in history with a market capitalization which has exceeded 5 billion dollars. Your desired graphics cards are the backbone that supports the training and execution of the models. generative AI. Their strategic status explains that the most powerful, such as H200can cost up to $30,000 per unit.
From the hand of giants like Google and emerging firms such as OpenAIin 2025 Silicon Valley invested around $400 billion in IAa volume without historical precedents. Approximately half of that spending goes to the purchase of advanced chips. However, its useful life is very short, a problem for the aspirations of companies that aspire to recover their investment and generate benefits that allow them to continue setting the course of AI.
Columns of servers inside a Google data center. / Google
Nvidia assures that its chips last between four and six years. Based on this, the companies that buy those semiconductors to feed their data centers —clients like OpenAI, MicrosoftGoogle, Amazon y Meta— they amortize them over that period of four to six years.
However, Burry and other analysts say that chip technology is advancing so quickly that it really only takes two to three and a half years for them to become obsolete. His warning struck a chord. So much so that Nvidia responded by denying the majority.
“One of the most common frauds”
Burry’s response, particularly in the media after being represented in the film The Big Shortit was devastating. “Underestimating depreciation by artificially lengthening the useful life of assets increases profits, one of the most common frauds of the modern era. (…) Yet this is exactly what all the hyperscalers have done,” he tweeted. If right, companies like Oracle or the matrix Facebook e Instagram They could be overvaluing their earnings by more than 20%.
This obsolescence makes the AI bubble “completely different” from previous ones, says the investor, consultant and MIT researcher Paul Kedrosky. And the fact is that, while infrastructures such as railway tracks or power cables fiber optic last for decades, the rapid aging of chips would force companies to change their equipment more frequently. A higher expense that makes them much less profitable and, therefore, it is much more difficult for them to access capital to finance their operations.
A future without generative AI
There is also another risk of a technical aspect that is less talked about. The sector’s inflated valuations are based on the assumption that the big investment model deployed by Silicon Valley and the great language models (LLMsfor its acronym in English) that shape the call generative AI and, therefore, chatbots like ChatGPT, Gemini, Claude o Grok They are the future.
Last January, the Chinese company DeepSeek He has already questioned the first premise by launching a model capable of competing with Silicon Valley with a much lower investment. The commitment of influential experts such as Demis Hassabisgeneral director of Google DeepMindo Yann LeCunformer head of AI at Meta, for other types of intelligent systems opens the door to a second, even more disturbing question: What if the industry has put all the money on the losing horse?
Subscribe to continue reading
