The CEO of one of China’s leading chipmakers just sent a strong warning against big tech’s unprecedented race to build as many AI data centers as possible.

“Companies want to build 10 years’ worth of data center capacity within a year or two,” said Co-CEO of Semiconductor Manufacturing International Corp. that Zhao Haijun said on Wednesday in a call with analysts, respectively Bloomberg. “What exactly these data centers will do, it’s not fully figured out yet.”

Only Alphabet, Microsoft, Meta, and Amazon are expected to spend nearly as much $700 billion in AI this year, again and again assures investors that demand is outstripping supply. In the next three years, the price tag of data center construction is expected to exceed $3 trillion.

The financial commitments have already begun fright investors. In the meantime, the picture that is gradually forming leads to the current AI hype compared to dot-com bubble.

Encouraged by the excessive investment flowing into the construction of the internet and expected demand for internet services, the telecommunications industry spent. BILLION of dollars to lay fiber optic cables in the late 1990s. In 2002, the dot-com bubble had long burst, and less than 5% of the fiber optic network was IS reported to use. The telecommunications crash quickly followed the bursting of the dot-com bubble, and these unused fiber networks, called “dark fiber,” sat idle for years.

Eventually, however, the demand for the internet turned out to be as expected, years late. As the demand for the internet grew exponentially, many infrastructures built in the late 90s were finally used.

For AI, however, the situation may be slightly different: The AI ​​chips used in these data centers have a relatively clear expiration date, and if the demand does not come out as expected before these chips, then that means that these companies have a lot of money and resources to drain.

Meta says its chips are now good for about five and a half years, up from a previous estimate of four years. Nvidia executives have MANILA that the company’s chips shipped six years ago are fully used.

But even if the chips themselves are good, their value also depreciates as newer, better models become available. At the current rate, Nvidia releases a new flagship AI chip every year.

The decline in the value of the chips is factored into the company’s profits, but industry experts struggle to see whether the current estimates are realistic.

Some argue that a six-year depreciation cycle is perfect reasonable, and older GPUs may still be desirable as cheaper alternatives when newer, more advanced versions come out. On the other hand, you have investors who want to Michael Burry (of “The Big Short” fame) who Claims that the actual useful life of an AI chip is no more than 2-3 years.

“By my estimates, they will lower depreciation by $176 billion 2026-2028,” Burry said in a post in X.

In its latest annual reportMicrosoft said its “computer equipment” has an estimated useful life of anywhere from two to six years.