Report: DeepSeek restarts financing, pre-investment valuation of 500 billion yuan

Original: Finance Magazine
Multiple trading individuals revealed to Finance that the large model company DeepSeek has restarted its second round of financing, planning to raise 50 billion yuan in this round, with a pre-investment valuation of approximately 500 billion yuan, and plans to complete the signing by late August.

DeepSeek initiated its first round of financing in April this year and completed the delivery in June, raising 50 billion yuan with a valuation exceeding 350 billion yuan, becoming the company with the largest financing scale in the history of China's AI large models for the first round.

Several investors mentioned that DeepSeek's second round of financing had at least started in mid-July, but was suddenly paused at the end of July. At that time, some investors on the negotiation standby list were informed that the plan to sign the financing agreement was postponed.

According to media reports on July 26, DeepSeek suspended its second round of financing, one reason being that DeepSeek founder Liang Wenfeng was dissatisfied with widely circulated comments online, which mainly revolved around the suspected leaked "meeting transcript for investors."

Finance sought confirmation from DeepSeek regarding the above market information, but had not received a response by the time of publication.

The aforementioned trading individuals told Finance that DeepSeek and the investors at the negotiation table hope that this financing can proceed discreetly after it restarts. Currently, some investment institutions that had previously been actively in contact indicated that they had not received news of DeepSeek restarting the financing, and the channel remains in a paused state.

According to several trading individuals, on the negotiation list for the second round of financing, there are both standby institutions that "failed" in the first round and VC (venture capital) and industrial capital that entered in the first round considering additional investments. The investors in DeepSeek's first round of financing include: National Artificial Intelligence Industry Investment Fund, Tencent, CATL, Puxuan Capital, NetEase, JD.com, Lisi Capital, IDG Capital, Zhengxin Valley Investment, and Shixiang Capital.

Since opening financing, DeepSeek's shares have been in high demand. An investor involved in the transaction told Finance that the scale of funds expressing investment intentions around DeepSeek in the first round exceeded 100 billion yuan, and ultimately raised 50 billion yuan, indicating that there is at least another 50 billion yuan of funds interested in entering. These funds continue to compete for new shares after the second round opens.

The pre-investment valuation for the second round of financing is 500 billion yuan, an increase of about 43% compared to the first round. If this round of financing goes smoothly, DeepSeek will raise over 100 billion yuan in total across the two rounds.

DeepSeek's financing scale far exceeds that of its competitors, including the recently hotly financed company Dark Moon, as these AI large model star companies have significantly accelerated their financing pace this year.

Dark Moon completed its Series F financing at the end of July with a valuation of approximately 31.5 billion USD (about 210 billion yuan), and immediately opened a new round of Pre-IPO financing with a valuation of 50 billion USD after the round ended. According to insiders, Dark Moon's current Pre-IPO financing is expected to be completed in August, with plans to file for listing within this year.

Also stirring market sentiment alongside financing dynamics are the model releases from these star companies. The capabilities and efficiency of the models directly relate to the pricing of model companies in the primary market. In an era where model iterations are measured in months and rankings change weekly, investors hope to anchor the first ship to reach AGI (Artificial General Intelligence) amidst uncertainty and seek tickets. One trading individual believes that "the pricing of large models is essentially an option, rather than a financial model based on cash flow."

According to previous reports from Finance, the initial heat of Dark Moon's Series F financing, which started in June, was tepid, with many channels that received quotas "selling for several days without anyone wanting it," and there were even many transactions of old shares at an 18 billion USD valuation circulating in the market. However, after the release of the new flagship model Kimi K3, which gained global attention, its shares suddenly became sought after, and some sellers of old shares even raised their prices.

An investor close to DeepSeek told Finance in late July that DeepSeek is under considerable pressure to train new models. The official version of DeepSeek-V4, originally scheduled for release in mid-July, only went live for public testing in the latest log of the official API documentation on July 31, with only V4-Flash showcased, while the official version of V4-Pro has not yet been released, and the specific public testing time has not been announced.

According to official documentation, DeepSeek-V4-Flash did not make adjustments to the model architecture and parameter size but strengthened post-training on the original basis, significantly enhancing its Agent capabilities, with multiple benchmark tests far exceeding the V4-Flash preview version released in April.

According to the latest Intelligence Index released by the independent third-party AI model evaluation platform Artificial Analysis, the intelligence level of the official version of DeepSeek-V4-Flash reached 50 points, second only to Zhiyu GLM-5.2's 51 points in China, and above the average level of similar models. DeepSeek continues its model strategy of "low cost and high cost-effectiveness," with the output price of V4-Flash being 0.28 USD per million tokens, lower than Zhiyu GLM-5.2's output price of 4.29 USD per million tokens.

In the context of rising demand for inference computing power and continuously increasing costs, many large model companies are pursuing extreme inference efficiency, with Tencent's Hunyuan also following a high cost-effectiveness route domestically. Recently, the American AI model company OpenAI also announced a reduction in the API call prices for the GPT-5.6 series models, with the entry-level model Luna seeing an 80% price drop. Before the official release of DeepSeek-V4-Flash, the cost-effectiveness of GPT-5.6 under the same model capability was even higher than that of DeepSeek. However, this cost-effectiveness advantage was quickly erased.

On the latest list of the AI model aggregation platform OpenRouter aimed at global developers, the official version of DeepSeek-V4-Flash ranked first in weekly token consumption with 7.1 trillion tokens.