"‘The boom in AI memory chips is sowing the seeds of collapse’! This time, the risk lies on the demand side."

```

How long can the boom in AI memory chips last?

Wall Street Journal columnist James Mackintosh wrote on May 16 that the current prosperity of AI memory chips is sowing the seeds of its own destruction. The article uses Micron Technology as a core case to systematically examine the cyclical risks behind this round of AI chip mania.

Micron Technology recorded its largest loss ever three years ago, but it is now predicted to become the sixth most profitable company in the US, with net profit close to $100 billion in the next 12 months—surpassing Meta and Berkshire Hathaway. Meanwhile, Samsung Electronics and SK Hynix in South Korea are also benefiting from the explosion in demand for high-bandwidth memory (HBM), making Korea's stock market one of the best-performing markets globally this year.

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

Low valuation does not equal cheap—history repeatedly proves otherwise

Micron Technology currently has a price-to-earnings (PE) ratio of less than 10 times, making it one of the cheapest stocks in the S&P 500. Many investors see this number and think “it’s so cheap, worth buying.”

To this, James Mackintosh pours cold water on the idea: “This doesn’t mean it’s cheap—it simply shows that investors know the good times for memory chips won’t last forever.”

Historical data confirms this. When Micron’s share price peaked in early 2022, its PE ratio was also just 9 times—after which its stock price was cut in half that year. At the cyclical top in 1984, the PE ratio was 15 times, and it took 9 full years for the price to return to that high. At the 2018 top, the PE was even lower at 5.5 times.

The pattern is clear: low PE ratios appear at the cycle peak, not as buy signals, but as warning signals. Investors fooled by “cheapness” suffer heavy losses each time.

Biggest risk: AI itself becomes more “memory efficient”

Mackintosh believes the current danger mainly comes from the demand side, not the supply side. The new capacity added this year and next won’t crush profits as long as demand remains strong.

Will demand collapse? He listed several risks. The hardest to quantify—and most fatal—is: AI technology itself becomes more memory efficient.

In March this year, researchers from Alphabet (Google) published a paper showing a huge jump in memory efficiency. As soon as the news broke, memory chip stocks took a hit, though they later rebounded to some degree.

James Mackintosh wrote in the article: “Large language models are an immature technology, and engineering improvements for dedicated data centers are to be expected—but how large those improvements are, and when they arrive, cannot be known in advance.”

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

Risk two: Data center expansion falls short of expectations

In addition to technical efficiency risks, James Mackintosh also listed risks faced throughout the AI supply chain: data center construction plans may be trimmed, the speed of AI adoption may be lower than expected, and political resistance could slow expansion.

His assessment: “These risks can all happen, but the AI bulls driving share prices higher don’t seem to care.”

This sentence itself is worth investors’ attention—the market consensus is often most fragile at its most optimistic.

Risk three: High profits attract new entrants

The third risk is changes in the competitive landscape.

James Mackintosh notes that there are no obvious new entrants in Micron’s high-speed memory sector yet, but in other high-profit AI chip sectors, competition has already begun.

Alphabet has developed Tensor Processing Units (TPUs) specifically for AI training, directly replacing Nvidia’s expensive GPUs. Amazon’s Graviton chip handles CPU needs for AI inference tasks, diverting Intel’s market share.

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

James Mackintosh concludes: “When AI demand is high, all this extra supply can be absorbed, and the impact on margins is small. But the longer this goes on, the more competitors come in, and the more capacity gets built.”

This is a typical “winner’s curse”: The more profitable, the more people come to compete, eventually diluting the profits.

Memory chips are a classic strongly cyclical industry. Building a wafer fab requires massive investment, and after demand rises, supply takes years to catch up, causing prices and profits to soar. High profits then spur CEOs to expand production, and high fixed costs force factories to run at full capacity—even if supply is already excessive. The crash in 2022–2023 was the result of this process.

Mackintosh notes that investors are not unaware of this pattern. But his view is: As long as AI demand continues booming, this new supply can temporarily be digested by the market, limiting the impact on profit margins. “But the longer it lasts, the more new competitors enter, the more capacity gets built.”

He wraps up with this line: “Like all commodities, success itself sows the seeds of self-destruction—even if the beautiful vision for AI ultimately comes true.”

Risk Disclosure and DisclaimerThe market has risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular situation. Investment based on this, responsibility is at your own risk. ```