Memory chip stocks slump as Google cuts AI cache needs sixfold
Alphabet's TurboQuant algorithm compresses AI key-value cache storage by at least sixfold without precision loss, with tasks like code generation and question answering unaffected. Memory chip stocks including Samsung, SK Hynix, Kioxia, Sandisk and Micron slumped as traders weighed the breakthrough ahead of its April ICLR presentation.
1. Google unveils TurboQuant breakthrough
On March 24, Alphabet introduced TurboQuant, an algorithm that compresses the AI model’s key-value cache by at least sixfold without degrading precision. The technology maintains performance on tasks such as code generation, question answering and text summarization and is currently confined to laboratory testing.
2. Market reaction
Shares of major memory hardware producers fell sharply following the announcement, with Samsung down nearly 5%, SK Hynix losing 6% and Kioxia off almost 6%. In the U.S., Sandisk and Micron also recorded declines during Wednesday trading as investors reassessed demand projections.
3. Analyst perspectives
Industry experts cautioned that reducing inference memory needs often paves the way for more advanced AI models that may require greater hardware support. Some analysts attributed part of the sell-off to profit-taking after a period of sustained growth in a cyclical sector.
4. Outlook and potential impact
TurboQuant offers no relief for the high RAM demands of AI training and lacks widespread deployment, with a formal presentation scheduled for ICLR 2026 in April. Some forecasts suggest lower entry barriers could spur localized AI projects, potentially driving overall long-term chip consumption higher.