DeepSeek reportedly planning mega IPO: $71 billion valuation under discussion

At the center is a new financing round that, according to Bloomberg, is targeting a volume of at least ten billion yuan. At a pre-money valuation of 480 billion yuan – equivalent to around 71 billion US dollars – the final amount could be significantly higher depending on investor demand.

Already at the end of May, DeepSeek had closed its first external financing round after a report first published by the Financial Times, raising around seven billion US dollars at a valuation of 52 billion US dollars. According to the reports, participants in this round included Tencent, battery manufacturer CATL, JD.com, NetEase, IDG Capital, as well as a state-run AI investment fund from Beijing, the National Artificial Intelligence Industry Investment Fund. Founder Liang Wenfeng reportedly contributed around three billion US dollars of private wealth; his fortune is currently estimated at around 36 billion US dollars.

Proprietary chips and gigawatt-scale data center

The rapid succession of financing rounds is, in the Financial Times' assessment, primarily related to the capital requirements for proprietary data centers and additional AI chips – both drivers for the development of AI agents. Reuters also reports that DeepSeek has been working on its own AI chips for about a year. In June, the company said it had advertised positions for its own data center team and plans to build a campus on a gigawatt scale.

Technological positioning of V4 Pro

Despite the ambitious valuation, the company's technological positioning is mixed. According to the benchmark service Artificial Analysis, the current top model V4 Pro ranks only 15th in the Intelligence Index with 44 points – far behind top US models such as Anthropic's Claude Fable 5 (59.9 points), OpenAI's GPT-5.6 Sol (58.9), or Claude Opus 4.8 (55.7). A year ago, DeepSeek's R1 models were still the leading open-weight reasoning models in the index. In the practical test GDPval-AA for real-world work tasks, however, V4 Pro leads with 1554 points ahead of other open models such as Kimi K2.6, GLM-5.1, and MiniMax-M2.7. At the same time, according to Artificial Analysis, V4 Pro has a very high hallucination rate – the model reportedly responds almost always, even when it does not know the answer. The price-performance ratio is also noteworthy: V4 Pro costs 0.02 US dollars per task, while GPT-5.5 at 1.04 US dollars and Claude Fable 5 at 2.75 US dollars cost 20 to 45 times as much.

Annualized revenue is approaching the 500 million dollar mark, according to The Information. Liang Wenfeng has reportedly signaled to potential investors that DeepSeek will continue to prioritize fundamental research over short-term commercialization and stick with open-source models.

Impact on Chinese AI stocks

The news also made waves in the stock markets. In Hong Kong, Zhipu's share price rose by 7.88 percent at times to 1,726.00 HKD, while MiniMax has noticeably retreated again over the course of the year after an initial price surge. However, the MiniMax Group share also rose significantly in the wake of the Bloomberg report: it climbed by 13.30 percent to 260.60 HKD.

Zhipu, MiniMax, and DeepSeek are collectively considered China's so-called AI Tigers, a group of start-ups that are supposed to lead China's AI development. Zhipu, developer of the GLM models, has been the more sought-after of the two stocks since its IPO in January and is currently valued at around 93 billion US dollars in Hong Kong.

The timing comes amid a phase of record-breaking tech IPOs: OpenAI and Anthropic have confidentially filed IPO documents in 2026 according to consistent reports, and SpaceX completed a record IPO in June. Should DeepSeek actually go through with the step of going public, it would be one of the highest-valued AI companies from China – and a further signal of the growing weight of Chinese AI start-ups in global competition.

This text is for informational purposes only and does not constitute an investment recommendation.