[Weekend Money] What Do Semiconductors and Paper Have in Common?... The Paradox of Google's 'TurboQuant'

Google has introduced its 'TurboQuant' technology, which can significantly reduce the use of memory semiconductors, causing the stock prices of memory chip makers such as Samsung Electronics, SK hynix, and Micron to plummet. However, analysts in the securities industry believe that, in the end, the biggest beneficiaries of TurboQuant will actually be the memory semiconductor companies themselves.

TurboQuant, released by Google, is a data compression algorithm that can reduce the memory usage of artificial intelligence (AI) models by up to six times. After Google published a related paper on March 24 (local time), concerns over slowing memory demand due to increased AI technology efficiency came to the fore, sending shares of Micron and SanDisk into a steep decline for two consecutive days. Samsung Electronics and SK hynix also experienced significant drops.

However, some believe these concerns are unfounded. Lee Sunghun, a researcher at Kiwoom Securities, stated, "It is unlikely that the emergence of TurboQuant will lead to a structural slowdown in memory demand." He added, "Currently, TurboQuant exists only at the research paper level, so there will be a time lag before it is actually implemented. Moreover, we should consider the scenario in which TurboQuant reduces AI operating costs and increases efficiency, leading to more latecomer companies joining the AI ecosystem, resulting in a larger AI market and a corresponding increase in total memory demand (Jevons paradox)."

Notably, when DeepSeek was introduced previously, related stocks also plunged in the short term but quickly rebounded. Kim Dongwon, a researcher at KB Securities, commented, "After DeepSeek was unveiled in January last year, Nvidia's stock price dropped sharply by 17% in the short term, but it quickly recovered within a month and even surpassed previous levels, which is highly instructive. Both TurboQuant and DeepSeek are meaningful as software optimization technologies for achieving low-cost, high-efficiency AI. However, considering the pace of anticipated investment in AI data centers over the next five years, there are practical limits to how much these technologies alone can accommodate the explosive growth in AI demand."

Lee also noted, "Similarly, during the DeepSeek incident last January, the initial market response was mostly shock, but in the medium to long term, stocks rallied again. Ultimately, the recent plunge in memory stocks triggered by TurboQuant is largely attributable to short-term profit-taking by investors, following a year-long rally in semiconductor shares, particularly as investment sentiment has weakened due to the ongoing war. Therefore, it is inadvisable to reduce exposure to the semiconductor sector in response."

Efficiency innovations are expected to lead to a surge in demand. Kim explained, "While the apparent purpose of TurboQuant is to reduce costs and improve power efficiency by making AI models more lightweight, the deeper strategic intent is to structurally expand overall AI inference demand. The AI industry is now moving beyond simple model performance competition into an era of ecosystem competition centered on applications and operating systems. Technologies like TurboQuant draw more users into the AI environment and encourage developers and services to become entrenched within the Google stack. This will ultimately result in an explosive increase in AI usage, stronger ecosystem lock-in, and an expansion of dominance by AI platforms."

Similarly, in the early days of the Internet in the 1990s, it was expected that the spread of email and digital documents would reduce paper consumption. In reality, however, the increase in PC, printer, and email use, along with the growth of web document printing, led to a surge in paper consumption over 12 years (1995–2007). Kim noted, "This is a classic example of the rebound effect, where efficiency improvements do not lead to decreased demand but rather expand total demand. The same trajectory is expected for AI."

In the end, the ultimate winners from Google's TurboQuant are expected to be memory semiconductor makers. Kim emphasized, "Low-cost AI technologies such as TurboQuant, Grok, and SRAM lower the barriers to AI adoption and exponentially expand total demand. This directly translates into increased computational loads and greater memory capacity requirements. Thus, the biggest beneficiaries in the competition to expand the AI ecosystem will be memory semiconductor companies." He highlighted Samsung Electronics, SK hynix, and Micron as his top picks in the sector.

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