Liang Wenfeng (梁文峰), founder of Chinese AI startup DeepSeek, which triggered the "DeepSeek shock" in the AI industry, cited computing resources such as graphics processing units (GPUs) as the key variable in U.S.-China AI competition. He said the monopoly power of Nvidia's AI chip ecosystem is gradually weakening and assessed that China's AI chip ecosystem is facing a new opportunity, while noting that he will prioritize developing artificial general intelligence (AGI) over maximizing profits.

Liang Wenfeng, DeepSeek founder. /Courtesy of Baidu

According to China Business News on the 24th, DeepSeek recently completed its first external funding round. Participants included Liang Wenfeng himself, Chinese big tech company Tencent (腾讯), battery maker CATL, private investment firms, and government funds. In an investor meeting that lasted about four hours, Liang detailed the company's management philosophy, AGI strategy, and plans to secure computing resources.

◇ "Nvidia ecosystem's monopoly power is shaking"... confidence in China's AI chip ecosystem

According to the leaked transcript, Liang Wenfeng pointed to computing resources as a "gap that looks very hard to close" in U.S.-China AI competition and said that although they already know the technical direction, there are not a few cases where research must be postponed due to a lack of computing resources. He added, "We expanded compute capacity very aggressively this year and currently have roughly 20,000 Nvidia H100 chips," and said they plan to secure as many units as possible going forward.

Liang Wenfeng also said he plans to build a large in-house AI cluster. As for whether to develop their own AI chips, he said, "If we can buy chips at a reasonable price, there is no need to make chips ourselves," adding that the decision will be based on economic feasibility.

He offered an optimistic outlook on Chinese AI chips that have drawn attention due to U.S. export controls on semiconductors to China. Nvidia not only sells hardware such as GPUs, but also built the dedicated software development platform CUDA early on, creating the Nvidia ecosystem. Because of this, even if competitors catch up with Nvidia's hardware performance, it is difficult to replace the software and developer ecosystem, so Nvidia's market dominance has been maintained.

However, Liang Wenfeng said, "As AI makes it possible to build a new ecosystem, the monopoly power of Nvidia CUDA is rapidly crumbling," adding, "Within a year, it will be proven that there is no problem at all with the Chinese chip ecosystem. Chinese AI chips have met a historic opportunity." He also revealed plans to participate deeply in Huawei's AI chip ecosystem. He noted, however, that insufficient production capacity is a limitation.

DeepSeek logo. /Courtesy of Reuters Yonhap News

◇ Top priority is AGI... "We won't compete to acquire users or maximize profits"

Liang Wenfeng repeatedly emphasized that the company's top priority is not competing to acquire users but developing AGI. AGI refers to AI that is not confined to a specific field and, even without users' "perfect" prompts, can learn on its own like a human and perform all intellectual tasks that humans can do.

He said that even when DeepSeek received unexpected attention during last year's Lunar New Year holiday, they did not focus on retaining users or expanding consumer services, saying, "The opportunity for AGI is very big, and many opportunities before us are just small things," and, "We have no intention at all of becoming a super app to compete with someone and becoming the second ByteDance Ltd. or Tencent."

He also made clear the policy of maintaining an open-source and low-price strategy. From the outset, the AI market cannot be monopolized, so rather than sticking to a closed strategy to maximize profits, growing an open ecosystem is more advantageous for survival. Liang Wenfeng said, "We are very willing to help competitors such as Alibaba, Zhipu, and Moonshot AI," adding that even the highest-performing model to be released in the future is likely to adopt an open-source approach.

He went on to say that, in the long term, the AI market could become so large that it accounts for 10% of global gross domestic product (GDP), adding, "If we wanted to maximize profits, we could double usage prices, but DeepSeek's purpose is not to make the most money; it is to enable everyone to use affordable, high-performance services."

He also said that it is sufficient for API prices to be at a level where server investment expense can be recovered in about 10 months, indicating a policy of pursuing only a reasonable level of profit. He added, "There is still room to lower prices."

※ This article has been translated by AI. Share your feedback here.