The Emergence of DeepSeek... Rather an Opportunity for Domestic Companies

by Ryu Hyunseok

Published 31 Jan.2025 07:55(KST)

Sangsangin Securities analyzed on the 31st that the launch of a high-performance AI model by the Chinese AI startup DeepSeek could present an opportunity for domestic software (SW) companies.

The Chinese AI startup DeepSeek recently unveiled an inference AI called 'DeepSeek-R1.' DeepSeek-R1 is an AI that verifies its own answers and performs reasoning to correct hallucinations through multiple stages, ultimately finding accurate answers. In some performance tests, it was even evaluated to outperform OpenAI's 'o1' model.

DeepSeek claimed to have developed this AI model in just over two months with a cost of less than 6 million USD. They also explained in a technical report that they used about 2,000 units of NVIDIA's affordable chip 'H800.' The H800 is a product made by NVIDIA for export to China to circumvent U.S. government export regulations. It has reduced performance and is much cheaper.

Sangsangin Securities explained that DeepSeek's high performance is due to optimization. Researcher Choi Seung-ho of Sangsangin Securities said, "DeepSeek's learning method is not new. Reinforcement learning, a core methodology of model training, is already widely used, but DeepSeek succeeded in optimizing it to an astonishing level, creating a stark difference in results even with the same methodology."

In particular, since DeepSeek introduced the model based on open source, it is expected to be an opportunity for domestic companies. Korea belongs to the open-source camp, so the success of open-source models creates a trickle-down effect.

Researcher Choi Seung-ho analyzed, "In the future, AI may no longer be the exclusive domain of big tech companies. The reason DeepSeek shocked the market is that it implies that overwhelmingly high computing resources may not be necessary to implement high-performance models."

He emphasized, "If DeepSeek's methodology can be properly followed, AI development at home and abroad, which has been difficult due to infrastructure cost investments, is expected to become more active. As AI moves from a few to everyone, the overall total capital expenditure (CAPEX) and the AI cycle's upward trend are expected to accelerate."

Sangsangin Securities identified NAVER as the biggest beneficiary in domestic AI on the internet side and Krafton in gaming. He said, "Many domestic companies have already completed LLM development, but due to the nature of AI requiring overwhelmingly high computing resources, their performance compared to global players was poor. If high-performance models can be created using low computing resources, the commercialization of software can be faster."