The latest news from Big Epoch on July 26, 2026 reports that the second round of funding for the Chinese AI company DeepSeek has been temporarily suspended due to leaked recordings of internal meetings between DeepSeek founder Liang Wenfeng and investors.
According to a report by Bloomberg on July 25, sources revealed that DeepSeek has orally informed some investors that they will not be signing investment agreements in the next few days as originally expected. One of the reasons for the suspension is Liang Wenfeng’s dissatisfaction with the leakage of discussions he had with investors during the first round of funding being circulated online. Liang Wenfeng has decided to halt the new round of transactions and has requested the team to reassess the information disclosure process and investor communication mechanisms.
Sources indicate that the negotiation situation is still fluid, and DeepSeek may still decide to proceed with the funding. It is currently unclear whether DeepSeek has informed all potential investors of the aforementioned decision in this round of funding.
The leaked content of the internal meetings between Liang Wenfeng and investors has garnered widespread attention from both domestic and international media. According to meeting minutes compiled by several media outlets, including mainland China’s First Financial, some of Liang Wenfeng’s remarks may have displeased the Chinese government and raised concerns from the US government.
The meeting minutes reveal that Liang Wenfeng candidly stated, “China does not have a cost advantage in data labeling, especially not in labeling high-end data, making it difficult for us to invest in data labeling like the US does. This path is challenging in China because data labeling is very expensive.”
According to Amazon Web Services (AWS), data labeling in machine learning is the process of identifying and adding one or more meaningful and content-rich labels to raw data (images, text files, videos, etc.) to provide context for machine learning models to learn from. Various use cases require data labeling, including computer vision, natural language processing, and speech recognition.
Liang Wenfeng also mentioned, “There are too many companies in China making models now… maybe only three in the US are doing basic models… so resources are also more diversified, and to some extent, wasted.”
The meeting minutes show that Liang Wenfeng elaborated on the company’s vision, determination to achieve general artificial intelligence, and the gap between China and the US in the AI field. Liang Wenfeng admitted that China’s AI industry still heavily relies on NVIDIA chips and lags behind the US in maturity. He stated that DeepSeek currently has the equivalent computing power of about 20,000 NVIDIA H-series GPUs, while the US is already training models on a scale of 800 billion parameters, which would require 50,000 of the latest AI GPUs.
He openly acknowledged the possibility of “purchasing some non-compliant cards” to obtain the necessary resources, and mentioned that in the coming months, there will be a large number of machines arriving, mostly NVIDIA-based.
Political observer Xia Yan believes that the Chinese government has long emphasized the advantages of massive data and low labor costs in China. However, Liang Wenfeng’s remarks directly point out China’s deficiencies in “high-quality AI training data,” even admitting that DeepSeek can only afford to “label low-cost data first.” Moreover, his criticism of the rush by various local governments and tech companies in China to engage in a “battle of large models” despite insufficient computing power was described as “resource-wasting” by him. These statements contradict the propaganda narrative of the Chinese Communist Party and challenge the policies of local governments across the country that heavily subsidize and support the large model industry, thus touching a nerve within the CCP.
Xia Yan also mentioned that Liang Wenfeng’s comments directly relate to highly sensitive issues of smuggling and export controls for the US government, which could easily prompt actions to block such activities.
Technology industry discussions on the X platform regarding this incident suggest that abnormal procurement quantities, third-country transshipment, market premiums, and unusual payment entities are all areas of concern for the Bureau of Industry and Security (BIS) of the US Department of Commerce. Even if chips are deployed outside China, if the end-user or parent company is located in China, related companies typically cannot evade the licensing requirements for advanced computing chips through overseas subsidiaries. Therefore, the BIS may take action.