TradingKey - Earlier this week, Google (GOOG) (GOOGL) launched its new extreme compression algorithm, TurboQuant. The algorithm is expected to reduce memory usage by approximately sixfold and increase computing speeds by up to eight times on the same GPU configuration, triggering a broad pullback in the semiconductor sector.
During Wednesday's U.S. trading session, Micron (MU) , SanDisk (SNDK) , Western Digital (WDC) and Seagate Technology (STX) all declined, with losses extending into Thursday's pre-market session.
Downside pressure from U.S. markets also spilled over into Asian equities. On Thursday, Samsung Electronics closed down 4.71%, while SK Hynix fell 6.23%, dragging the South Korean benchmark KOSPI index down 3.22%. Both companies have seen their shares decline for four consecutive sessions this week.
Google's algorithm optimizes storage bottlenecks during the inference process of Large Language Models (LLMs), reducing memory usage and boosting computing speeds without any loss in precision. However, will this technology truly disrupt semiconductor demand? How significant will the near-term impact be?
News of Google's TurboQuant algorithm launch quickly garnered market attention, and Cloudflare (NET) CEO Matthew Prince called it Google's "DeepSeek moment," representing a major historical breakthrough in AI efficiency.
Wall Street noted that if successfully implemented, storage demand could drop significantly, potentially undermining the DRAM and NAND Flash demand previously bolstered by the AI boom. In January, Samsung Electronics raised NAND flash contract prices by more than 100%, following a nearly 70% hike in DRAM prices, underscoring the fervor in the semiconductor market.
Despite the collective cooling of the global semiconductor sector, Goldman Sachs (GS) technology specialist Peter Callahan believes the market is not in extreme panic, but that investors are conducting a reality check on the recent extraordinary rally in storage stocks.
In fact, the market had already moved before the Google TurboQuant "black swan" emerged—storage giant Micron, after posting strong earnings, saw its stock price trail the Philadelphia Semiconductor Index by nearly 20% within five days, the largest short-term relative underperformance since 2011.
Although the TurboQuant algorithm can theoretically reduce memory requirements, the technology is still in the research phase and has not been commercially validated.
From a technical perspective, the technology is actually only applicable to dynamic VRAM consumption during the inference process (primarily KV Cache) and does not involve the model weights themselves. In other words, the greatest breakthrough of this technology is the improvement in operational efficiency, but the storage space required by large models themselves cannot be reduced.
In addition, considering that the number of parameters in AI models is also growing exponentially, the algorithm can currently only save up to six times the storage space through compression, which may be a drop in the bucket.
The introduction of this technology is most likely to cool down the currently overheated semiconductor market and puncture the valuation bubble of memory stocks, proving that memory demand may not be entirely limitless. With the launch of similar technologies and algorithmic progress, the growth of storage demand may encounter a bottleneck.
It is worth noting that compared to standard DRAM chips, this technology will have a smaller impact on HBM (High Bandwidth Memory). TurboQuant is mainly used to optimize AI model inference, a stage that mostly requires only ordinary DRAM chips. However, HBM remains a necessity in the AI training phase. For the three HBM giants—Micron, Samsung Electronics, and SK Hynix—the TurboQuant algorithm will have almost no material impact in the near term.