A 500 billion yuan pre-IPO target would price DeepSeek at 148 times reported revenue

Cryptopolitan, citing Reuters, reports $7.4 billion invested in June against term sheets written at close to 450 billion yuan, with the final valuation settling at 350 billion yuan, or $52 billion ^[11]. Crypto Briefing reports the same June round raising roughly 50 billion RMB at a valuation near 52 billion RMB, with Tencent and JD.com among the backers ^[13]. Fifty-two billion RMB is roughly a seventh of 350 billion yuan ^[7]. The South China Morning Post's account of the listing preparations puts pre-IPO financing at around 500 billion yuan, or $74 billion ^[9]. On its own June figure Crypto Briefing calls that a nearly tenfold jump ^[14]; measured against 350 billion yuan it is 43% ^[1].

The 50 billion figure moves between the accounts too. Cryptopolitan has DeepSeek seeking 50 billion yuan in a round now open ^[10]. Crypto Briefing has 50 billion RMB as money June already brought in ^[13]. If the open round does raise 50 billion at a 500 billion valuation, the buyers get 10% of the company ^[2].

Then the price against the revenue. Crypto Briefing reports DeepSeek nearing $500 million in annual revenue ^[15]. At the $74 billion attached to the 500 billion yuan figure, that is 148 times revenue ^[3]. At June's $52 billion, 104 times ^[4].

There are two published measures of how much cheaper the Chinese models are, and they are not the same size. Juniper Research puts the cost of running Chinese models at up to 90% below some of the most popular US alternatives, which is one tenth ^[18][5]. The cached-input rates VentureBeat lists sit much further apart: $0.003 for DeepSeek-V4.1-Flash against $0.40 for GPT-5.6 Sol and $0.50 for Claude Opus 5, or 133 and 167 times ^[19][6]. Cached input is one rate on a bill that also charges for fresh input and output, so Juniper's 90% is the figure that describes a full workload.

Public money is not the near-term source of chips. Crypto Briefing has the process starting in 2026 and the listing itself targeted for 2027, about 18 months out ^[7][8]. Until then the compute and the hiring come out of the private round, and DeepSeek ran on cash from Liang Wenfeng's quantitative hedge fund High-Flyer from its founding in 2023 until it began taking outside money this year ^[22][23]. The Shanghai Stock Exchange published guidance in June on the fifth set of listing criteria for large-model companies, which asks an applicant to show at least a model already operating and widely used ^[16], and the exchange acknowledged that the field requires significant investment in computing power and specialist talent ^[17].

Yan was born in 1991 ^[2] and comes from GL Ventures, the venture arm of Hillhouse Investment, which is not itself a DeepSeek investor ^[4]; he has backed MiniMax, a DeepSeek rival ^[3]. His mandate is reported to cover investor relations and financial controls ^[5]. Crypto Briefing also reports DeepSeek pursuing founder control through non-voting shares acquired in earlier rounds ^[24].

In my view the pressure on American pricing already sits in the published rates, and the listing mostly changes who funds DeepSeek's losses. Two things would show that reading wrong. One is the follow-on round closing below 500 billion yuan, the way June closed below its own term sheets ^[11]. The other is performance: Stanford's 2026 AI Index put the best US models 2.7% ahead of the best Chinese model as of March 2026 ^[20], and RAND found the two countries' models converging in architecture and commercial positioning ^[21]. Widen that 2.7% and the $0.003 rate is buying a worse model.