'NVIDIA's CUDA dominance is crumbling,' said the founder of DeepSeek during a four-hour presentation, revealing a policy of prioritizing the realization of AGI over profit maximization.
Liang Wenfeng, founder of the Chinese AI company DeepSeek, held a four-hour briefing for investors on May 20, 2026, and a transcript of the discussion has now been released. In the briefing, Liang explained the company's future plans, stating that 'we will not aim to maximize profits, but will prioritize
DeepSeek's Liang Wenfeng Breaks His Silence - by Fred Gao
https://www.fredgao.com/p/deepseeks-liang-wenfeng-breaks-his
Wenfeng Liang: Four-Hour Investor Meeting Transcript | elsewhere
https://elsewhere.news/en/elsewhere/wenfeng-liangs-four-hour-investor-meeting-full-transcript
DeepSeek puts AGI research ahead of products and commercial growth · TechNode
https://technode.com/2026/07/23/deepseek-puts-agi-research-ahead-of-products-and-commercial-growth/
DeepSeek is developing open-source AI models that rival the cutting edge models of OpenAI and Anthropic, and while continuously improving performance, it maintains a stance of prioritizing research and development over monetization. In May 2026, it was reported that the company was in negotiations to raise a large amount of funding, 70 billion yuan (approximately 1.6 trillion yen), and on May 20, 2026, Mr. Liang gave a presentation in a dialogue format as an investor briefing. Compared to American founders who often appear in the media, opportunities for Mr. Liang to clearly explain his vision for the AI industry are rare.
Mr. Liang does not view AI as merely a tool for monopolization, but rather as 'a massive wave that will change the course of human history, a trend that no company can monopolize.' Therefore, he argues that excessive profits should not be pursued, and that information should be openly shared to reduce resistance, emphasizing the realization of AGI as the true goal, rather than short-term profits or user growth. Furthermore, Mr. Liang believes that the most important key to realizing AGI is not finding the best talent, but rather having a team that can collaborate over the long term.
Liang stated that he believes the dominance of top American models like OpenAI, Anthropic, and Google is temporary and will not last long. Anthropic has gained a significant advantage in agent coding with the emergence of '
According to Liang, American companies are ahead because they have abundant resources and their business idea is based on a closed model and high profitability to dominate the market. However, Liang points out that this pursuit of high profitability is weak in the long run and will inevitably lose to companies that are content with reasonable profits. In particular, Liang believes that China has an inherent advantage in terms of cost and user experience because American companies do not have a strong reason to cut costs to the extreme.
NVIDIA's CUDA ecosystem excels in its ability to deliver high performance even when compared to hardware with high theoretical performance through software optimization, and has therefore been prioritized in the AI field, which demands high computing power. However, according to Mr. Liang, the dominance of the CUDA ecosystem is being challenged by new technologies, and building a new ecosystem for Chinese-made chips is seen as a historic opportunity. The lack of a CUDA-compatible environment is a barrier to the widespread adoption of Chinese AI chips, so if this is overcome, the fundamental software gap between China and the United States will narrow, leaving only pure computing power as the remaining difference.
Liang predicts that in the future, 'just like in other manufacturing sectors, low-cost Chinese products will become widespread in the field of AI.'
Liang explained that coding agents are a priority for DeepSeek, both now and going forward, because they have the potential to improve both DeepSeek's products and its own research processes. On the other hand, features such as 3D generation, video generation, and world models, while useful applications, are not considered to be at the core of intelligence development and are therefore not prioritized for development. Similarly, specialized agents in fields such as finance and medicine have a lower priority, and the wisest approach is to focus all efforts on general-purpose agents.
Mr. Liang repeatedly explained that DeepSeek is a research company, not a product company, and that DeepSeek's API pricing is designed not to maximize profits, but to recoup hardware costs in about 10 months. He also revealed that DeepSeek plans to release its highest-performing model as open source, rather than releasing lower-performing versions to the public, so that the same model can be used both internally and externally.
Furthermore, Liang denied any plans to develop DeepSeek Chat into a 'super app' that integrates a wide variety of services. Liang explained, 'The reason we don't compete in that is because there's a watermelon at the end of the road, and what we have in front of us now is just a sesame seed. Even if that sesame seed were a little bigger, it wouldn't be that big in the end.'
Regarding the open-source orientation and the goal of AGI, Mr. Liang said, 'Open source and low cost give employees a sense of accomplishment, strengthen organizational cohesion, contribute to society, and make colleagues and the general public happy. Therefore, from a long-term perspective, these constraints increase the probability of achieving AGI. If you are aiming to maximize profits, you have already lost. You may face even greater difficulties. That's how the world works.'
Related Posts:
in AI, Posted by log1e_dh