Wood, Thompson warn 10x HBM price surge is a negative signal

Memory and storage chip stocks face a structural risk their current rally is hiding, as two prominent investors argue that surging high-bandwidth memory prices are creating the forces that will eventually eliminate demand.

"Most people think this is a huge positive, but it's actually a negative signal," Cathie Wood, founder of ARK Invest, said in a recent podcast, explaining why her firm holds no storage-chip positions.

Wood pointed to HBM prices that have tripled, quadrupled, and in some cases risen tenfold, calling the surge abnormal for a semiconductor segment she describes as the most commoditized and cyclical in the industry. She cited Cerebras and Groq — whose inference chips are architected without HBM — as evidence that engineering is replacing the technology, drawing a parallel to Tesla removing cobalt from its batteries.

The bearish view challenges the AI-infrastructure trade that has lifted HBM suppliers including Micron, SK Hynix, and Samsung. If inference-chip architectures and algorithmic optimization reduce HBM reliance, current prices could accelerate the shift rather than sustain the rally.

Wood's caution stems from years of watching cyclical semiconductor markets swing between boom and bust. Storage, she argues, sits at the most commoditized end of the chip chain, where price spikes historically invite capacity expansion and substitution. The current HBM surge, driven by AI training demand, is no exception, she said.

The more consequential development is on the demand side. Cerebras and Groq, both building inference-focused chips, have designed their architectures to avoid HBM entirely; ARK's venture fund holds a stake in Groq. Wood compared the shift to Tesla's decision to strip cobalt from its batteries: when a supply-chain input becomes expensive or risky, engineers find a way around it. "In inference, we're seeing demand for HBM replaced by engineering," she said.

The Iran Analogy for Memory Pricing

Ben Thompson, a tech strategy analyst, frames the same risk in competitive terms. He likens memory makers to Iran, whose leverage over the Strait of Hormuz works only while the threat remains unused. "Now they've played that card, and it worked," Thompson said. "But the UAE and Saudi Arabia will build pipelines and new ports so it doesn't happen again." He noted that Apple is lobbying to bring in Chinese memory suppliers, and that the top algorithmic optimization priority has become reducing memory usage.

Both arguments converge on the same conclusion: the storage rally's strength is accelerating its own replacement. For investors, the question is timing. HBM remains a critical input for Nvidia's GPUs and hyperscaler data centers, and near-term demand shows no sign of easing. Nvidia this week announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise more than $500 billion in third-party capital for AI infrastructure, a scale of buildout that currently supports HBM demand.

Yet the structural shift Wood and Thompson describe — inference architectures that skip HBM, algorithmic efficiency, and supply diversification — could erode the pricing power that underpins the trade. The same week ARK bought the dip in Nvidia shares, it continued to avoid storage names, a choice that reflects a bet on the technology's eventual replacement rather than its current profitability.

This article is for informational purposes only and does not constitute investment advice.