Amid the AI infrastructure investment boom, memory chip stocks appear to benefit from surging computing power demand, yet two prominent investors and tech analysts remain cautious. Their reasoning converges on the same core judgment: the high prices and margins in memory are breeding forces that will ultimately undermine them.
Cathie Wood, founder of ARK Invest, recently explained on a podcast why she does not hold memory stocks. She argued that memory is the most commoditized and cyclically volatile segment of the semiconductor supply chain, and the current sharp price surge is not normal for the tech industry—it is essentially a negative signal.
Meanwhile, tech strategy analyst Ben Thompson issued a sharper warning, likening memory makers to Iran blockading the Strait of Hormuz—effective in the short term, but ultimately pushing the entire industry to find alternative routes.
Their assessments directly challenge the current market sentiment favoring memory stocks. As AI inference demand expands rapidly, high-bandwidth memory (HBM) suppliers have seen sustained stock price rallies, but the logic of these two investors suggests that investors should be wary of structural risks stemming from technology route shifts and demand-side avoidance.
Cathie Wood: High Prices Are a Warning, Not a Positive
Cathie Wood directly addressed questions about why she does not hold memory stocks in a recent video. She admitted that years of investment experience may have made her more vigilant about cyclical industries.
In her view, memory is the most commoditized segment of the entire semiconductor supply chain, historically subject to violent cyclical swings. The current trend of HBM prices tripling, quadrupling, or even rising tenfold is abnormal for the tech industry. "Most people see this as a huge positive, but it's actually a negative signal," she said.
She further backed this judgment from a cash flow perspective. She noted that a recent chart showed free cash flow of chip stocks moving in exactly the opposite direction to that of hyperscalers—the former benefiting while the latter is under pressure. But she stressed that this state is temporary.
More critically, Cathie Wood pointed out that technological innovation is actively reducing reliance on HBM from the demand side. She cited Cerebras and Groq—with ARK's venture fund holding a stake in Groq—noting that the architectures of both inference chip companies do not require HBM.
She likened this trend to Tesla's removal of cobalt from its batteries: once a supply chain component becomes too expensive or risky, engineers find ways to work around it. "In the inference space, we're seeing HBM demand being replaced by engineering innovation," she said.
Ben Thompson: Memory Makers Are Setting Themselves Up as Targets
Tech strategy analyst Ben Thompson's concerns about the memory industry focus more on competitive dynamics. He used a geopolitical analogy to describe the strategic dilemma memory makers currently face.
"I compare memory makers to Iran," Thompson said. His logic: the deterrence of the Strait of Hormuz lies in it always being a card that can be played. Once it is actually used, it triggers the opponent's determination to bypass it entirely. "Now they've played that card, and it did work. But the UAE and Saudi Arabia will build pipelines and new ports so this doesn't happen again."
He believes the current high-price strategy of memory makers is creating the same effect. On one hand, Apple is lobbying to bring in Chinese memory suppliers to break the existing supply structure; on the other, the primary optimization goal in algorithms has shifted to "how to reduce memory usage."
"I worry memory makers may have done the same thing," Thompson said. "No one will let themselves be caught in such a passive position on memory again." His conclusion: in the long run, by creating such a massive "target on their backs," memory makers may end up shooting themselves in the foot.
Two Lenses, One Conclusion
Although Cathie Wood and Ben Thompson use different analytical frameworks, their conclusions are strikingly aligned: the current strength of memory stocks is accelerating their own replacement.
Cathie Wood approaches from the technological evolution path, emphasizing that innovation in inference chip architectures will systematically reduce demand for HBM; Ben Thompson approaches from competitive dynamics and supply chain politics, noting that high prices and high concentration will drive buyers to seek alternatives—whether through algorithm optimization, introducing new suppliers, or redesigning system architectures.
Together, their judgments point to a risk worth noting for investors: within the AI infrastructure investment narrative, the case for memory stocks may be more fragile than the market expects, and the current boom may in fact be the strongest force driving the industry to accelerate its search for alternatives.
Comments