Bloomberg reported on Tuesday that DeepSeek, an artificial intelligence laboratory based in Hangzhou, is nearing a funding round that could raise at least 80 billion yuan, or 10.6 billion euros, according to the European Central Bank's exchange rate as of October 6. This amount surpasses the laboratory’s initial target of 50 billion yuan. Interest has grown significantly since the launch of DeepSeek's latest AI model, with the total potentially reaching 100 billion yuan, or 13.2 billion euros. Among the largest commitments are CATL, a leading battery manufacturer, and Tencent, which operates the messaging platform WeChat. DeepSeek initially aimed for a valuation of 500 billion yuan, or 66 billion euros, and plans to go public in early 2027, though final details are still being negotiated. CATL, Tencent, and DeepSeek did not respond to the report, which was published on a holiday in China. CNBC confirmed the same day, citing two sources, that DeepSeek is considering raising the round to 100 billion yuan and is still seeking a valuation of 500 billion yuan. The funding round is being pursued by public funds, investment arms of listed companies, and venture capital funds, but the laboratory is carefully selecting participants, excluding private funds raised from individuals. According to one source, CATL, Geely, a major automotive manufacturer, and investment firms Monolith Management and Loyal Valley Capital have made commitments. The round, expected to be completed by the end of August, has been delayed due to some investors hesitating on the price. The funding round is described as having a state character, with public funds and industrial groups playing a significant role. DeepSeek operated without external capital until 2026, when it began seeking investment. Founded in 2023 by Liang Wenfeng, the founder of the quantitative fund manager High-Flyer, the laboratory was initially funded by profits and the GPU resources of the fund. Reuters described the first round in June as a major shift in strategy. This initial round, launched in April and completed in June, raised about 50 billion yuan, giving DeepSeek a post-money valuation of around 350 billion yuan. Liang personally invested 20 billion yuan, while Tencent contributed 10 billion yuan and CATL 5 billion yuan, the two largest external commitments. The structure of this first round was designed to give Liang control: investors entered through a limited liability company managed by him, without voting rights in DeepSeek and with a five-year lock-up. According to TrendForce, the national investment fund was an exception, entering directly for 1 billion yuan, or 2 percent of the round. Nothing indicates that the ongoing round adopts this structure. In July, DeepSeek briefly paused further discussions after a viral summary of a meeting attributed to Liang mentioned the reliance on Nvidia chips and the Chinese lag in AI. By the end of August, the round was active again, expected to be completed by the end of the month. On September 9, Reuters reported that DeepSeek had commissioned CITIC Securities to prepare a listing on the STAR Market, the technology sector of the Shanghai Stock Exchange, with the intention of starting the process before the end of the year. Bloomberg noted that once the round is completed, the next step will be restructuring the company in preparation for the listing. The list of participants in the funding round highlights an industrial and political financing strategy. As early as June, Reuters noted that the list of investors "highlights the efforts of China to build an increasingly self-sufficient AI industry, from models to energy infrastructure." CATL, a dominant player in batteries, has committed to electrical equipment and data center storage. Tencent, whose Hunyuan AI model is outpaced by Doubao from ByteDance and DeepSeek, seeks to integrate AI into its platforms. The arrival of Geely and the filtering of private funds reported by CNBC align with this strategy. Z.AI, formerly Zhipu, and MiniMax listed in Hong Kong on January 8 and 9, with MiniMax doubling in value on its listing day. Moonshot AI, valued at 50 billion dollars, has confidentially filed with Hong Kong and also aims for the beginning of 2027. This timeline explains the urgency. Reuters recalls that these valuations remain far below those of Anthropic and OpenAI, "a gap that reflects a strong disparity in revenues." According to the prospectus of MiniMax, the company generated 53.4 million dollars in revenue for the first nine months of 2025, with a net loss of 512 million. At 500 billion yuan, or 75 billion dollars in Reuters' conversion, DeepSeek would be valued significantly above the stock market capitalization of Z.AI, approximately 54 billion dollars on September 9, without published accounts. The premium rewards a national champion status, the qualification from Reuters, and a technical reputation. Bloomberg reported that DeepSeek plans to deploy at least 160,000 Huawei accelerators in the data center it is building in Inner Mongolia, making it one of the largest known clusters of Huawei chips. This project was reported by ActuIA at the beginning of September, noting that the chips are not intended for training, which remains on Nvidia. DeepSeek has also published with Huawei software to program these chips, including DeepGEMM-Ascend, a port of its matrix calculation libraries compatible with Nvidia GPUs. Reuters adds that the capital must finance computation, model development, and talent retention, with DeepSeek having recently lost researchers to better-funded companies like ByteDance and Xiaomi. Liang committed during at least one meeting with investors to continue developing open models while pursuing the goal of general artificial intelligence, indicating that his main goal was to push the limits of technology rather than monetize. Recent actions align with this. The weights of DeepSeek-V4.1-Flash are available on Hugging Face under the MIT license since September 10, the day the API update log presented this model as "the smallest model of our new architecture family," with a price reduction. ActuIA detailed the attention cache reduced to 890 bytes per token for this model. On the Artificial Analysis index consulted on October 6, V4.1 Flash scores 39 points, one more than the new French model, but behind five open Chinese models. For European teams using these weights, this promise has so far only the author's guarantee. Liang retained control of the laboratory in June, where he was the first subscriber with 20 billion yuan. The national fund, if it indeed entered directly with voting rights, is the other shareholder whose voice will weigh on the maintenance of openness. A listing in Shanghai will require DeepSeek to publish accounts. After Z.AI and MiniMax, this would be another measure of what is reported by open-weight models offered at low prices, this time for the laboratory that imposed this business model. It will also fix DeepSeek in a framework that European buyers already know through its frictions. In June, ActuIA described how Beijing seals the training data of Qwen and DeepSeek when AI Act requests a summary. On September 8, the NSA, CISA, and FBI named six Chinese companies, including DeepSeek, accused of having distilled at an industrial scale the American models. A listed DeepSeek, funded by the country's largest industrial groups and backed by a state fund, will be even more exposed to these tensions. Its European users will have to take this into account in their sourcing choices. As of October 6, the DeepSeek update log does not mention any fundraising. Its last entry, dated September 10, announces V4.1-Flash and reduced prices. The next verifiable deadline is the initiation of the listing process in Shanghai, which Reuters placed before the end of the year.