DeepSeek preps a STAR Market IPO — and cuts prices 60% the same day
China's most-watched AI lab spent Wednesday doing two things that pull in opposite directions: quietly hiring the bankers for a Shanghai listing, and publicly retreating from the price increases it imposed three weeks ago.
DeepSeek has hired CITIC Securities to prepare an IPO on Shanghai's STAR Market, with the listing process expected to begin within the year, according to people familiar with the matter. The Hangzhou lab is also working with accounting firms to close its books by the end of December — the financial report that has to exist before a filing can happen. It closed a record $7 billion funding round only weeks ago and is already talking to new backers about another raise at a pre-money valuation north of 480 billion yuan, up from roughly $50 billion in June.
The timing is the story. A company heading into a public listing needs a revenue curve, and DeepSeek's most obvious lever — raising API prices — just failed. On August 17 it put prices up across the V4 line, adding peak and off-peak billing and, in V4 Pro's case, raising cache-hit input by roughly 1,100%. On September 9 it walked most of that back: V4 Flash cache-hit input returns to its April level, a cut of up to 60% effective September 10. Output prices barely moved, and V4 Pro was untouched.
DeepSeek's problem is that it is competing with itself. Because V4 Flash ships open weights, 28 separate providers serve it on OpenRouter, and third parties undercut the official API by a factor of four or more — same model, different endpoint, one line of config changed. DeepSeek still holds about 23.8% of OpenRouter's text requests, more than OpenAI and Google combined, but most of that volume can switch away in an afternoon. An open-weights lab cannot price like a closed one, and a pre-IPO revenue story built on margin expansion runs straight into that wall. We covered the hiring push behind the same crunch in DeepSeek opens 150 engineering roles — and makes agent fluency mandatory.
Alibaba Cloud has entered Gartner's leaders quadrant for Strategic Cloud Platform Services for the first time, the only Chinese or Asia-Pacific vendor in a group that otherwise holds AWS, Microsoft, Google, and Oracle. Gartner said the global cloud infrastructure and platform services market reached $297 billion in 2025, with the evaluated vendors covering 97.8% of it — and credited Alibaba with 33.9% CIPS revenue growth, the highest among the six vendors that disclosed a rate.
The report's reasoning is the interesting part: Gartner frames Alibaba not as an infrastructure vendor that added AI, but as one converting itself into an "AI Native and Agent Native Cloud," with a vertically integrated stack running from its own Pingtouge Zhenwu chips through the Qwen model family to agent tooling and applications. Alibaba says the family now exceeds 460 open models, 3 billion downloads, and 300,000 derivative models. Read it as Gartner validating the full-stack thesis — own the silicon, the weights, and the agent runtime — at a moment when Western clouds are still assembling theirs from parts.
What to watch: whether DeepSeek's next model, V4.1 Flash — now in internal testing with claims of up to 420 tokens per second — lands before its books close, and whether an open-weights lab can convince public-market investors that a model nobody has to pay for is still a business.
Should a company whose weights are free be allowed to list on the strength of API revenue alone? Tell us in the comments.
