“The prosperity of AI memory chips is sowing the seeds of a collapse!” This time, the danger lies on the demand side.

How long can the prosperity of AI memory chips last?

The Wall Street Journal columnist James Mackintosh wrote on May 16th that the current boom in AI storage chips is sowing the seeds of self-destruction. Using Micron Technology as a core case, the article systematically outlines the risks behind this round of AI chip frenzy.

Micron Technology recorded its biggest loss in history just three years ago, but now is predicted to become the sixth most profitable company in the U.S., with net profits approaching $100 billion in the next 12 months, surpassing Meta and Berkshire Hathaway. Meanwhile, South Korea’s Samsung Electronics and SK Hynix are also benefiting from exploding demand for high-bandwidth memory (HBM), making the Korean stock market one of the best performers globally this year.

But the more prosperous it is, the more dangerous it becomes.

Low valuation does not equal cheap, history keeps proving it wrong

Micron’s current P/E ratio is less than 10, making it one of the cheapest stocks in the S&P 500. Many investors see this number and think, “It’s so cheap, it’s worth buying.”

In response, James Mackintosh’s analysis poured cold water: “This doesn’t mean it’s cheap, it just indicates that investors know memory chip good days won’t last forever.”

Historical data confirms this. In early 2022, when Micron’s stock peaked, its P/E ratio was also only 9—then the stock price halved that year. At the cycle top in 1984, the P/E was 15, and it took a full nine years for the stock to return to that high. At the 2018 cycle top, the P/E was even lower at 5.5.

The pattern is clear: low P/E appears at cycle peaks, not as a buy signal, but as a warning signal. Investors lured by “cheapness” have suffered heavy losses each time.

The biggest risk: AI becomes more “memory-efficient” itself

Mackintosh believes that the current danger mainly comes from the demand side, not the supply side. The additional capacity added in the next one or two years, as long as demand doesn’t collapse, is not enough to crush profits.

So, will demand collapse? He listed several risks. The most difficult to quantify and the most deadly is: AI technology itself becoming more memory-efficient.

In March this year, researchers at Alphabet, Google’s parent company, published a paper showing a significant improvement in memory usage efficiency. Once the news broke, memory chip stocks fell sharply, though they later recovered somewhat.

James Mackintosh wrote in the article: “Large language models are an immature technology; engineering improvements targeted at dedicated data centers are expected— but how much and when they will arrive is unpredictable.”

The logic is simple: if AI models run faster and use less memory, data centers won’t need to buy as many HBM chips, and demand will shrink.

Risk two: Data center expansion falls short of expectations

Besides technological efficiency risks, James Mackintosh also listed the risks faced by the entire AI supply chain: plans for data center construction may be scaled back, the pace of AI adoption may be slower than expected, and political resistance could also slow down expansion.

The article’s assessment is: “All these risks could happen, but the AI bulls pushing stock prices higher don’t seem to take them seriously.”

This statement alone should alert investors—market consensus is often most fragile at its most optimistic.

Risk three: High profits attract new competitors to accelerate entry

The third risk involves changes in the competitive landscape.

James Mackintosh pointed out that in the high-speed memory field where Micron operates, there are currently no obvious new entrants, but in other high-profit AI chip sectors, competition has already begun.

Alphabet has developed specialized tensor processing units (TPUs) for AI training, directly replacing expensive NVIDIA GPUs; Amazon’s Graviton chips handle AI inference workloads, diverting market share from Intel.

More noteworthy is Cerebras. The company launched its first large chip for AI training and inference in 2019, raised $5.55 billion in IPO funding this year, and its stock price doubled on the first day of trading.

This is a classic “winner’s curse”: the more profitable it gets, the more players rush in, ultimately diluting profits.

Memory chips are a typical cyclical industry. Building a wafer factory requires huge investments, and when demand rises, supply takes years to catch up, causing prices and profits to surge. High profits then stimulate CEOs to expand capacity, and the high fixed costs force factories to operate at full capacity—even when supply is already overshooting. The crash from 2022 to 2023 was exactly this cycle.

Mackintosh pointed out that investors are not unaware of this pattern. But his judgment is: “As long as AI demand remains high, these new capacities can be temporarily absorbed by the market, with limited impact on profit margins. However, the longer this lasts, the more competitors enter, and the more capacity is built.”

He concluded with a sentence: “Like all commodities, success itself sows the seeds of self-destruction—even if the optimistic vision of AI ultimately comes true.”

Risk warning and disclaimer

Market risks exist; invest cautiously. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. Invest accordingly at your own risk.