Chipmakers, Big Tech brace for fallout from DeepSeek breakthrough

Home > Business > Industry

Chipmakers, Big Tech brace for fallout from DeepSeek breakthrough

Published: 30 Jan. 2025, 18:24

LEE JAE-LIM
[email protected]

Audio report: written by reporters, read by AI

The logo of DeepSeek is displayed alongside its AI assistant app on a mobile phone, in this illustration picture taken on Jan. 28. [REUTERS/YONHAP]

The successful debut of a new AI model developed by Chinese startup DeepSeek will likely deliver a jolt to Korean chipmakers like SK hynix, a major memory chip supplier to Nvidia, as well as U.S. Big Tech players.

Nvidia’s market value dissolved by $593 billion — a record one-day loss for the AI giant — on Monday following reports that DeepSeek’s low-cost AI model is not only exceptionally cheaper to operate but functionally competitive with OpenAI’s GPT. The domestic stock market remained shielded from falling share prices as trading was closed on Monday for a temporary holiday followed by the Lunar New year holiday through Thursday.

Financial Supervisory Service Gov. Lee Bok-hyun on Thursday expressed concerns that domestic stocks would take a hit from DeepSeek’s entry into the AI industry.

“The emergence of low-cost AI models like DeepSeek, along with concerns over the overvaluation of U.S. Big Tech stocks, could increase market volatility and bring significant changes to the structure of the AI industry,” Lee said at the policy meeting following the U.S. Federal Open Market Committee. “Therefore, it is essential to closely monitor related developments.”

The Chinese company said that training its large language model, V3, cost $5.57 million and took two months, which is approximately 10 percent of the estimated cost to train models like GPT-4. The H800 GPUs used in this process are modified version of Nvidia’s H100 chips, adjusted to comply with U.S. export restrictions on high-performance chips to China.

DeepSeek’s AI chatbot application that utilizes its AI model, R1, outpaced ChatGPT to become the most-downloaded free app on the U.S. App store on Monday — leading experts to question the need for astronomical investments in AI chip development or establishing large-scale data centers to power AI models.

R1 is the latest open-source model released on Jan. 20 by the startup that can mimic the reasoning process of humans, with optimization for coding and mathematics. V3 is the general-purpose language model that was released last month.

However, domestic experts have cautioned against using AI models from the Chinese startup, citing concerns of data breaches.

“Not only does it collect users’ device information, but also keyboard input patterns, typing rhythm, IP information, device IDs and even cookies,” wrote Ha Jung-woo, head of AI innovation at Korea’s largest portal site, Naver, in a Facebook post while linking to DeepSeek’s privacy policy.

“Not surprisingly, the collected user data is stored on security servers located in China. Users should be aware of this in advance and weigh their options carefully before using the service.”

Naver, the dominant search engine in Korea, has opted to advance and integrate its proprietary AI model into its core functions in its search and shopping platforms.

Meanwhile, OpenAI and Microsoft launched a probe against DeepSeek over allegations that the Chinese company utilized data from OpenAI without authorization, according to reports from Bloomberg and the Wall Street Journal on Wednesday.

BY LEE JAE-LIM [[email protected]]