In the corridors of Wall Street, the word "memory" no longer refers to human recollection, but to the "gold" of artificial intelligence: High Bandwidth Memory (HBM). As we navigate May 2026, the euphoria surrounding semiconductor firms like SK Hynix, Micron, and Samsung has reached levels reminiscent of the mid-90s boom. Yet, amidst this climate of extreme optimism, a voice from Harvard Business School is making waves by challenging the consensus.
Willy Shih, a professor of management practice and one of the world’s leading experts on the chip supply chain, is sounding the alarm. His analysis is straightforward: the market is falling for a classic linear projection fallacy. “Curves that just go to the sky with no end… never continue forever,” he notes, reminding investors that the semiconductor industry is, by its very nature, deeply cyclical.
The Illusion of Eternal Scarcity
The current narrative suggests that AI is creating an insatiable thirst for data and, consequently, for memory. Large Language Models (LLMs) require massive amounts of HBM to operate efficiently, leading to shortages and skyrocketing prices. Wall Street analysts are pricing in the assumption that this demand will remain unabated for the next decade.
Shih, however, sees the telltale signs of an impending oversupply. When prices rise, every manufacturer rushes to increase capacity. Capital expenditures (CapEx) for industry giants are at all-time highs. “The problem with semiconductor fabs is that they take years to build, but when they come online, they flood the market all at once,” he explains. This phenomenon, known as the “bullwhip effect,” has historically led to spectacular price collapses, leaving companies with massive inventories of unwanted silicon.
The Real Cost of AI and Geopolitical Chess
Beyond pure economic theory, Shih focuses on the actual cost of infrastructure. Big Tech—Microsoft, Google, Meta, and Amazon—is spending billions every quarter to build out data centers. However, there is a limit to how long this spending can continue if the return on investment (ROI) from generative AI fails to materialize in the short to medium term.
- Dependence on TSMC and Taiwan remains the “elephant in the room” for global stability.
- Subsidies from the CHIPS Act in the US and EU are creating artificial incentives for excess capacity.
- China is accelerating its own self-sufficiency efforts, which could displace Western players from a massive market segment.
According to Shih, semiconductor history is littered with the “graveyards” of companies that believed the cycle had finally been broken. From the DRAM wars of the 1980s to the dot-com bubble, the pattern remains identical: excessive optimism, massive over-investment, and eventually, a hard landing back to reality.
Survival Strategy in a Volatile Market
For investors and decision-makers, the Harvard professor’s message isn’t necessarily doomsday-prophesying, but it is a stern warning. AI is a genuine revolution, but its infrastructure isn’t governed by the laws of software (where margins are nearly 100%). Instead, it follows the brutal laws of physical manufacturing and capital intensity.
“People forget that memory is a commodity. And with commodities, all it takes for the price to collapse is a 5% drop in demand or a 5% increase in supply.”
As we move into the second half of 2026, the focus is shifting from how many chips can be produced to how many are actually needed. If the AI bubble begins to leak air, memory companies will be the first to feel the squeeze, as they sit at the very base of the supply chain. Willy Shih reminds us that wisdom in technology lies not just in predicting the future, but in understanding the eternal cycles of the past.
