Google AI Compression Tech Sparks Jitters in Memory Chip Market

Google's artificial intelligence (AI) data compression technology, 'TurboQuant,' has drawn significant attention to the memory semiconductor market since its unveiling. Concerns that the technology could enable more efficient data processing, thereby reducing the need for memory chips, have impacted the stock prices of related companies, including Samsung Electronics and SK Hynix. Micron's stock experienced a correction following the TurboQuant announcement, with Samsung Electronics and SK Hynix also seeing their share prices decline.

TurboQuant is a technology that compresses the capacity of KV cache—the temporary storage space AI uses to remember conversation context—to one-sixth of its original size. This breakthrough opens up the possibility of overcoming the limitations of existing methods, which required significant investment to increase the capacity of memory semiconductors themselves.

The AI market has recently shifted from the learning phase to the inference phase, increasing demand for high-bandwidth memory (HBM), DRAM, and NAND flash. While this surge in memory semiconductor demand had boosted stock prices for related companies, the emergence of TurboQuant technology has led to assessments that memory semiconductor demand could falter.

The industry, however, has not shown significant agitation over the potential decrease in memory demand due to TurboQuant. Memory supply shortages persist, and big tech companies are striving to secure long-term supply agreements (LTAs) for three to five years or more. Consequently, memory semiconductor manufacturers are cautiously selecting contracts. Cloudflare CEO Matthew Prince described TurboQuant as a 'Google deep-seek moment.' Although there were concerns about a contraction in the HBM market when China's AI model DeepSeek emerged, HBM demand subsequently increased.