Over the past three weeks, the stock market’s hottest technology trade has gone into reverse, clipping more than 20 percent off the gains of some memory chip all stars.
Samsung Electronics, SK Hynix and Micron Technology — the three companies that make almost all the world’s memory chips — have each fallen from their highs into bear-market territory, dragging down (stock ticker) DRAM, the new exchange-traded ETF built to track them, more than 36% from a June 22 high of $80.72.
The timing is strange. The selloff arrived just as the companies were posting extraordinary results: a record operating profit of nearly $60 billion at Samsung and a $26.5 billion Nasdaq listing for SK Hynix. Record profits and falling prices show how the market is grappling with a single question: Has artificial intelligence permanently changed the economics of the memory industry, allowing one of the world’s most cyclical industries to escape its own history?
If so, these companies are no longer cyclical commodity producers but suppliers of a scarce, strategic input to AI processors, the fastest-growing platform in technology. They deserve higher valuations because their earnings should prove more durable. If not, today’s remarkable profits will eventually resemble every boom before them: spectacular but temporary.
Our answer is that the structural-change case is winning. We expect a durable re-rating of the industry’s earnings power, not simply another cycle that fades with the next glut. Memory will still boom and bust, but the downturns should be shallower and through-cycle profitability structurally higher than investors have assumed. Here’s why and what could prove us wrong.
When Memory Was a Commodity
For decades, memory was the purest commodity in technology. One maker’s chips were interchangeable with another’s. Buyers cared about capacity, speed and price—not brand.
That made the industry’s swings extreme. A leading-edge fabrication plant costs tens of billions of dollars to build, while producing one more chip costs very little. Small increases in demand sent profits soaring because fixed costs were already covered and barriers to entry limited new competitors. Personal computers, smartphones, cloud computing and the pandemic work-from-home boom all followed the same pattern: boom, overbuilding, and bust.
Those cycles also thinned the field. Japanese, European and nearly all American producers disappeared after failing to survive repeated downturns. Today, Samsung, SK Hynix and Micron produce almost all of the world’s DRAM, with Samsung at around 38 percent of market share, SK Hynix near 30 percent and Micron at about 20 percent. Yet despite this concentration, standardized products allowed buyers to switch suppliers easily, limiting pricing power and keeping competition focused on price like a commodity market.
The industry’s cyclical reputation left a lasting mark on equity valuations. For decades, Samsung, SK Hynix and Micron consistently traded at lower price-to-earnings (P/E) multiples. In effect, high earnings were capitalized at low multiples because investors discounted them as transitory rather than structural.
How AI Changed the Pattern
Artificial intelligence broke that pattern. AI accelerators run on high-bandwidth memory (HBM), an advanced, stacked form of DRAM built to feed the enormous data appetite of large language models. Demand has grown so fast that the constraint is no longer finding buyers but making enough chips. SK Hynix alone supplies close to 60 percent of the HBM market. For the first time, the fear runs the other way. Hyperscale cloud providers now worry about securing supply, and much of global HBM output for 2026 and 2027 is already committed under long-term contracts. Bargaining power has moved from buyers to sellers.
The deeper change is in what memory has become. HBM is not interchangeable. It requires advanced packaging, careful heat management and close engineering with the chip designers it serves, so qualifying a new supplier is slow and switching is costly.
Memory has shifted from a standardized commodity to a differentiated product with stronger pricing power and higher margins. HBM also consumes far more manufacturing capacity than conventional DRAM, leaving less capacity for everything else. AI is tightening supply from both sides, helping lift profits across the industry and not just in HBM.
One Risk and a Second Wave
The obvious risk is that this depends on continued AI spending. Today’s profits are supported by the enormous capital budgets behind AI data centers, and slower investment would weaken demand for advanced memory.
The market clearly recognizes that risk, which helps explain why Samsung’s record quarter was met with selling. But the next downturn would begin with advantages previous cycles lacked: real pricing power, high switching costs and differentiated products. A second wave of demand is also emerging as AI expands beyond data centers into phones, tablets, laptops and other consumer devices.
Is This Time Different?
The market is still pricing tomorrow’s memory industry through the lens of yesterday’s economics. Memory producers no longer simply manufacture an interchangeable commodity. They supply one of the most strategically important inputs into the global AI economy.
The companies emerging from this AI boom are fundamentally stronger than those that emerged from previous cycles. Their earnings should prove more durable, their competitive positions more defensible, and their valuations increasingly driven by structural rather than cyclical economics. That is a lasting fundamental change the market has yet to fully price.
