Cathie Wood Steers Clear of Memory: Chinese Capacity Could Deepen the “Vicious” Cycle

Cathie Wood, founder and head of ARK Investment Management, does not plan to invest in the memory chip sector. Her reason is the “vicious” boom-and-bust cycles that characterize the industry. She says the problem may intensify as Chinese companies expand their production capacity. Bloomberg reported this on October 13, 2026. The comment comes amid elevated demand for memory from AI infrastructure, which is why it stands out.

Why memory is considered the most volatile segment

DRAM and NAND remain close to commodities: buyers compare modules mainly on price and capacity, and there is far less architectural differentiation than in processors. Supply grows in steps. A new fab takes years to build, the decision to build it is made when prices are high, and it comes online when the market is already saturated. As a result, prices fall faster than manufacturers can cut output.

The latest deep downturn illustrated this well. For fiscal 2023, Micron recorded a net loss of about $5.8 billion. Samsung's chip division posted an operating loss of roughly 14.9 trillion won in 2023. The DRAM market is highly concentrated: Samsung, SK hynix and Micron together control about 90% of output. Even that concentration did not protect them from the price collapse.

What China adds

The new element in Wood's thesis is Chinese supply. CXMT is growing in DRAM, and YMTC in NAND. Both companies are developing with significant state support. If such players decide how heavily to load their fabs on more than profitability alone, the balance of supply and demand becomes harder to predict. Equipment export restrictions hold back their technological level, but they do not stop volume growth itself, especially in mass-market segments.

The original Bloomberg note gives no details: Wood does not name specific companies, prices or timeframes. It is a general view of the industry's structure, not a quarter-by-quarter forecast.

The counterargument: AI and HBM

Demand for high-bandwidth memory (HBM), which accelerators need for training and inference of models, works against her position. It is an expensive, technologically complex product, and SK hynix, Samsung and Micron are competing for a place in accelerator makers' supply chains. But this argument has a weak spot. HBM depends on cloud companies' capital spending on AI, which is also cyclical. In addition, HBM production consumes silicon wafers that would otherwise go to conventional DRAM, so the segments are linked.

Key facts and metrics:

• Who is speaking: Cathie Wood, ARK Investment Management
• Position: avoiding the memory sector
• Argument: “vicious” boom-and-bust cycles, growth in Chinese capacity
• Date of report: October 13, 2026
• 2023 context: Micron's loss of about $5.8 billion for the fiscal year, Samsung chip division's loss of about 14.9 trillion won
• DRAM market concentration: three companies, about 90% of output

What to watch

Investors should look past the headlines about AI demand and watch cyclical indicators: contract prices for DRAM and NAND, inventories at manufacturers and buyers, and the capex plans of Samsung, SK hynix and Micron. Another signal is the share of CXMT and YMTC in supply and their announced investments in new lines.

Companies that buy servers and storage for AI clusters should avoid baking current memory prices into their budgets for years ahead. Long-term contracts with fixed volumes pay off in a growing market, but in a falling one they turn into overpayment. It makes more sense to combine them with some purchases at spot prices. Wood's position does not prove that a downturn is inevitable. It is a reminder that memory has a history in which shortages regularly gave way to overproduction.