Samsung and SK hynix Shares Rattle on Google's "TurboQuant" as Semiconductor Industry Reacts

Paper Introduces Technology to Reduce AI Memory Usage

Google's New Innovation After Ushering in the AI Era with 'Transformer'

Memory Chip Stocks Fall, CPU Makers Surge

Getty Images Yonhap News

Google, which has been at the forefront of artificial intelligence (AI) innovation, has presented a solution to the ongoing memory semiconductor shortage. As the shortage and bottleneck issues regarding memory semiconductors have intensified following those of graphics processing units (GPUs), this new solution is already having an immediate impact on the stock prices of semiconductor companies such as Samsung Electronics, SK hynix, Intel, and AMD.

On March 25 (local time), Google Research published a paper on its blog introducing a new AI compression algorithm called "TurboQuant." Google claimed, "TurboQuant will redefine AI efficiency with its unparalleled compression ratio." The company explained that the algorithm can significantly reduce memory usage without any loss of AI accuracy.

Google announced that TurboQuant can compress the KV cache down to a 3-bit level without accuracy loss. Foreign media outlets that reviewed the paper reported that this technology could reduce the ‘KV cache’ memory required by large language models (LLMs) by at least sixfold and provide up to an eightfold increase in speed. The KV cache is a type of ‘temporary memory storage’ that allows LLMs to retain past information in order to continue conversations.

The expansion of the KV cache has led to the "memory wall" bottleneck phenomenon, which significantly increases the burden on AI hardware. This is because the volume of interaction with AI has surged as AI models have become larger and demand for inference has grown. For large LLMs, the KV cache alone now requires a tremendous amount of memory. Agentic AI has also come to demand even more memory. This is one of the main reasons why the stock prices of Samsung Electronics and SK hynix have soared since late last year.

TechCrunch, an IT-focused media outlet, described TurboQuant as "a new AI memory compression algorithm." Another IT publication, Ars Technica, evaluated the technology as one that could greatly reduce memory usage and potentially disrupt the cost structure of existing AI infrastructure.

Experts in the AI industry are also responding enthusiastically. Matthew Prince, CEO of Cloudflare, referred to this achievement as "Google's DeepSeek moment" on his X account.

The announcement of TurboQuant had an immediate impact on the stock market. On this day, the share prices of memory-related semiconductor companies such as Micron and SanDisk fell sharply on the US stock market. On March 26, the stock prices of Samsung Electronics and SK hynix also declined in tandem. In contrast, CPU manufacturers such as Intel and AMD saw their share prices surge, highlighting the contrast.

Google has a history of disrupting the AI landscape. In particular, the Transformer played a catalytic role in ushering in the era of generative AI. Google developed the TPU as an alternative to NVIDIA’s GPUs, which are essential for AI training, and utilized it to introduce Gemini AI.

Lee Jaewook, Director of the AI Research Institute at Seoul National University and former visiting scholar at Google, said, "Google is a company that always comes up with alternatives whenever technical bottlenecks arise. What stands out is that Google has consistently sought new solutions instead of being tied to specific architectures, even in AI infrastructure."

This content was produced with the assistance of AI translation services.