Multiple negative factors trigger a correction in Hong Kong storage concept stocks. GigaDevice once dropped nearly 7% during trading.

Financial Associated Press, March 27 (Editor: Hu Jiarong) - Affected by fluctuations in global capital markets, Hong Kong’s storage chip concept stocks collectively fell today, with GigaDevice Semiconductor at one point dropping nearly 7%.

As of the time of publication, GigaDevice Semiconductor (03986.HK) fell 3.50%, and Montage Technology (06809.HK) fell 1.66%.

During the same period, Southern Asset Management’s double long Samsung Electronics (07747.HK) and double long SK Hynix (07709.HK) both fell over 7%.

In news, according to the CFMS|MemoryS 2026 Flash Memory Industry Summit, the prices of storage products have experienced three consecutive quarters of sharp increases. It is expected that the growth rate will slow down starting in the third quarter of 2026, gradually converging, with some specific product prices showing differentiation.

For customers, securing storage capacity is more important than locking in prices. Compared to the booming AI market, the consumer market represented by smartphones is entering a period of pain, with costs rising rapidly and sales expected to decline by about 10%, with some smartphones potentially dropping by as much as 30%.

At the same time, overnight, the U.S. stock market’s storage chip sector suffered a heavy blow. Micron Technology fell for the sixth consecutive trading day, with a single-day drop of 6.97%, accumulating a retreat of over 23% from the historical high recorded on March 18; SanDisk, Seagate, and Western Digital fell 11.02%, 8.33%, and 7.70%, respectively.

Market sentiment is further pressured, partly due to the industry attention sparked by Google’s release of the TurboQuant AI memory compression algorithm on March 26.

It is reported that this algorithm is designed specifically for the KV cache scenario during the inference process of large language models. Through innovative compression technology, it compresses 16bit or 32bit cached data to 3bit, reducing memory usage to 1/6 of the original level, and achieving zero precision loss for long context inference without the need for model retraining or fine-tuning. Testing shows that its 4bit version on the Nvidia platform has an inference speed approximately 8 times faster than the 32bit baseline, achieving a breakthrough balance between compression efficiency, precision retention, and inference performance.

This technology features plug-and-play capabilities and has been adapted for mainstream open-source models such as Gemma and Mistral, allowing for widespread application in AI servers, edge computing, and mobile devices, significantly reducing the computational and memory costs of deploying large models.

Google officially released the revolutionary TurboQuant AI memory compression algorithm on March 26, 2026. This technology is optimized for the KV cache scenario of large models, using an innovative algorithm to compress 16bit or 32bit model cached data to 3bit, reducing memory usage to 1/6 of the original level, while ensuring zero precision loss for long context inference without requiring model retraining or fine-tuning.

Test data show that its 4bit version has an inference speed on Nvidia chip platforms that is 8 times faster than the traditional 32bit baseline, successfully achieving a breakthrough balance between compression ratio, precision loss, and performance.

Institutions claim that the decline in AI application deployment costs is expected to stimulate broader demand growth.

In response to market concerns, several industry research institutions have pointed out that TurboQuant AI’s core optimization focuses on cache efficiency in the inference stage, without touching core application scenarios of storage chips such as HBM high bandwidth memory and model weight storage. From a medium to long-term perspective, the decline in AI application deployment costs is expected to stimulate broader demand growth, thereby expanding incremental space for the storage chip market.

Current stock price fluctuations are mainly influenced by short-term emotional disturbances, and the industry’s fundamentals have not undergone substantial changes. Leading storage manufacturers, with their continuous technological iteration capabilities, advantages in capacity layout, and resilience in the supply chain, still have a solid foundation for robust development.

Massive news and precise interpretations can be found in the Sina Finance APP.