Will Google's TurboQuant compress chip demand or accelerate AI prevalence?

U.S. and Korean stocks are being rattled by fears that Google’s newly unveiled compression algorithm TurboQuant could hurt the semiconductor sector by sharply reducing memory needs.

A screen in Hana Bank's trading room in central Seoul shows the Kospi closing at 5,460.46 points on March 26, down 181.75 points, or 3.22 percent, from the previous trading session. [YONHAP]

U.S. and Korean stocks are being rattled by fears that Google’s newly unveiled compression algorithm TurboQuant could hurt the semiconductor sector by sharply reducing memory needs.

The Kospi fell more than 4 percent intraday Friday from the previous session, briefly dipping into the 5,200 range before trimming losses to close at 5,438.87, down 0.4 percent.

“Concerns over TurboQuant eased somewhat, drawing bargain buying from retail investors,” said Lee Kyoung-min, an analyst at Daishin Securities. On Thursday, the S&P 500 and Nasdaq Composite Index fell 1.74 percent and 2.38 percent, respectively.

The immediate trigger for the market sell-off was TurboQuant, which Google revealed on Tuesday, a compression algorithm that the Big Tech giant says can cut memory use to one-sixth of previous levels while boosting processing speed eightfold. It remains at an experimental and theoretical stage, but the market quickly moved to price in the risk of weaker memory demand.

Over Wednesday and Thursday, shares of Micron and Sandisk fell about 10 percent and 14 percent, respectively. On Thursday and Friday, shares of Samsung Electronics and SK hynix also fell 5 percent and 7 percent, respectively.

The market appears to consider this latest development similar to the short-term jolt caused by the release of DeepSeek in early 2025. At the time, a large language model, DeepSeek-R1, was released by the Chinese AI startup DeepSeek, which claimed that the cost to build it was a fraction of that of ChatGPT, fueling fears that cheaper, more efficient AI could disrupt the existing semiconductor trade.

About $593 billion in market value was wiped out, marking the largest single-day loss in Wall Street history at the time. But the market later judged the panic to have been excessive, and the losses were recovered within a month.

A screen in Hana Bank's trading room in central Seoul shows the Kospi trading on March 27. [YONHAP]

“TurboQuant and DeepSeek are both meaningful in that they are software optimization technologies aimed at achieving low-cost, high-efficiency AI,” said Jeff Kim, the head of research at KB Securities.

“But technologies like these alone cannot absorb surging AI demand. On a map, a route may look like a straight line, but in reality, it can be an uneven dirt road. There is a gap between theoretical efficiency and real-world application.”

Some analysts say the long-term effect could be the opposite, with lower barriers to AI use ultimately expanding total demand. Even if an AI model’s memory usage falls to one-sixth, memory demand could still rise if service usage increases by more than six times.

“In the early 1990s, the spread of the internet was expected to reduce paper use,” Kim said. “In reality, however, paper consumption surged for 12 years as printer use, email use and web document printing all rose together. TurboQuant, too, is likely to lead ultimately to more computing demand and greater memory installation, making chipmakers the biggest beneficiaries.”

Still, the Middle East crisis remains a source of unease. “After the U.S. market closes on Friday, risk-off sentiment could intensify over the weekend if the frequency of hard-line remarks or unexpected actions by U.S. President Donald Trump rises,” the Daishin Securities analyst said.

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom. BY JANG SEO-YUN [kim.minyoung5@joongang.co.kr]