Retail investors, known as 'seohakgaemi,' in the United States, began to buy on dips despite fluctuations in American semiconductor stocks, including NVIDIA, caused by the emergence of the Chinese artificial intelligence startup 'DeepSeek.'

According to the Korea Securities Depository on the 30th, domestic investors net purchased $283.2 million (approximately 408 billion won) worth of 'NVDL' on the 27th (settlement date 29th). This ranked first in net purchase volume of foreign stocks. NVDL is an exchange-traded fund (ETF) that tracks NVIDIA's daily stock price increase at twice the rate.

Illustration = ChatGPT DALL-E 3

Despite a loss of 33.78% ($24.7) in NVDL's stock price on the 27th due to the shock from DeepSeek, there was a significant buying trend among domestic investors. The NVDL stock price plunged from $73.11 on the 24th to $48.41 on the 27th, before recording $56.88 on the 28th and $52.29 on the 29th.

The second most purchased stock by domestic investors on the 27th was also NVIDIA. While NVIDIA's stock price fell by 16.97% ($24.2) on the same day, domestic investors showed a buying advantage with $222.56 million (approximately 321 billion won).

Domestic investors also net purchased $174.31 million (approximately 251.5 billion won) worth of 'SOXL,' an ETF that tracks the daily increase rate of the American semiconductor index at three times the rate, on the 27th. SOXL also recorded a decline of 23.25% ($7.59) on the same day.

China AI corporations Deepseek logo. /Reuters·Yonhap News

The reason semiconductor stocks, including NVIDIA, were shaken by DeepSeek is that there are growing doubts about whether large-scale investments in high-performance AI are justified. DeepSeek claims to have developed the open-source AI model 'DeepSeek-R1' for $5.58 million within two months. This is one-tenth of the training cost that Meta Platforms incurred for its latest AI model, Llama 3. Nevertheless, the service quality of DeepSeek-R1 did not lag behind existing American open-source AI models. This led to market concerns that AI development corporations might reduce capital expenditures (CAPEX) on related facilities, including data centers.

Nevertheless, many global investment banks noted that the DeepSeek case is unlikely to lead to a decrease in AI investments or a downturn for big tech in the United States. This aligns with the judgment of retail investors regarding buying on dips.

Citi Securities said, 'The dominance of American AI corporations may be challenged, but access to advanced semiconductors will remain a competitive advantage for American companies.' Bernstein also noted that even if DeepSeek achieved equivalent performance at a cost 10 times cheaper, the current model's cost increase rate is rising by tenfold each year, stressing that innovation is necessary. They added, 'The outlook for AI-related companies such as NVIDIA and Broadcom remains positive.'

Wedbush said, 'Ultimately, the American big tech companies that have built a massive AI infrastructure and ecosystem will maintain their advantage' and predicted that investments in facilities will continue not only for AI but also for robotics and autonomous driving.

However, there is a consensus that the impact of DeepSeek will affect the market for the time being. Given that stock prices fluctuate wildly in one direction without moving consistently, this suggests that leveraged investments could be risky.

Morgan Stanley pointed out, 'DeepSeek has the potential to redefine the landscape of global AI technology' and noted that, in the long run, the United States' export restrictions on advanced semiconductors could create an unfavorable environment for Chinese AI corporations, potentially increasing the volatility of stocks related to semiconductors and AI technology.

Despite avoiding wild fluctuations due to the Seollal holiday closure, stock prices of Samsung Electronics and SK hynix are also expected to undergo adjustments. Goldman Sachs noted, 'Considering that high-bandwidth memory (HBM) contracts are made on an annual basis, changes in demand could have a more pronounced impact after 2026,' and added that this could have a short-term negative effect on the valuation of Korean memory semiconductor suppliers.