Samsung and SK hynix Unlikely to Face Another 'Plunge'... Google TurboQuant Under Scrutiny Ahead of ICLR [Into the World of AI]

ChatGPT Image Generation.

원본보기 아이콘

"The global artificial intelligence (AI) industry is turning its attention to ICLR (International Conference on Learning Representations), which will be held in Rio de Janeiro, Brazil, from April 23 to 27. This is because Google’s AI memory compression technology, 'TurboQuant,' will undergo concrete verification during the event."

As the market impact of Google's TurboQuant is drawing significant attention, AI experts from both Korea and abroad are focusing on ICLR. ICLR is one of the three major conferences in the AI field. At this event, Google plans to officially present its paper on TurboQuant, undergo peer review, and release the actual program code worldwide around June.

On March 24 (local time), Google published the TurboQuant paper on the Google Research Blog, which rapidly sparked debate over its potential impact on the global memory semiconductor market. TurboQuant is a technology that compresses the 'KV cache' (temporary memory) used by large language models (LLMs) to retain context. Previously, this process used 16 bits without compression; TurboQuant drastically reduces that, improving efficiency.

Since TurboQuant reduces memory usage to one-sixth, concerns have grown that this could negatively affect U.S.-based Micron, as well as Samsung Electronics and SK hynix in Korea, all of which produce memory semiconductors. This even caused a sharp drop in their stock prices. However, a growing number of voices in academia and the AI semiconductor industry argue that the so-called 'TurboQuant shock' in the market is excessive.

Jinwon Lee, Chief Technology Officer (CTO) of HyperAccel, explained, "Traditional methods used 16 bits without compression, but techniques such as 3- to 4-bit quantization for structural optimization have already been consistently discussed in academia." He added, "While Google's quantization approach is indeed a good idea, 4-bit quantization is not entirely new, and the market’s reaction at that time was excessive."

원본보기 아이콘

CTO Lee stated, "By increasing the efficiency of AI models, TurboQuant will generate more demand from companies that previously could not afford to use expensive hardware like NVIDIA’s graphics processing units (GPUs). On the contrary, if TurboQuant is thoroughly validated at ICLR-both in terms of securing memory storage space and achieving the targeted speed-it could actually spur semiconductor consumption by overcoming current limitations in AI services." He explained that cutting memory usage to one-sixth means greater efficiency, enabling the implementation of complex, large-scale AI services. This, in turn, would not shrink the market share for memory semiconductor manufacturers but instead drive up demand through infrastructure expansion.

Dongjoo Shin, CEO of Mobillint, also commented, "TurboQuant became a hot topic simply because it went viral in the media; it is not a technology that appeared out of nowhere." He continued, "With efficient technologies like TurboQuant, market demand will expand into areas that previously could not utilize such solutions, so demand for memory manufacturers such as SK hynix and Samsung Electronics will continue to grow."

CEO Shin expressed the view that the future stock prices of Samsung Electronics and SK hynix, which have recently posted record earnings, will be determined more by supply-side factors than demand. He predicted, "Competition among Micron, Samsung Electronics, and SK hynix, the race to expand production capacity, and the pursuit by Chinese memory manufacturers will play a bigger role in influencing stock prices going forward."

Dongpyeong Shin, Director of the Center for Technology Foresight at the Korea Institute of S&T Evaluation and Planning (KISTEP), said, "TurboQuant is an attempt to solve hardware problems through software innovation." He added, "As TurboQuant’s influence expands AI services in the long term and enables AI computation on more devices, the on-device AI market will also grow."

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.