October 07, 2026 – One of China’s most closely watched AI labs is weighing a raise of up to 100 billion yuan, double its original goal, as it prepares for a possible listing.
In Summary
DeepSeek’s round could reach 100 billion yuan, about $14.9 billion, double its original 50 billion yuan goal, reports say.
Commitments reportedly exceed 80 billion yuan, with Tencent and CATL among the largest backers.
The reported 500 billion yuan valuation target equals about $74.6 billion, under 9% of OpenAI’s $852 billion.
A roughly 4% rise in the yuan this year adds about $600 million to the dollar value of a 100 billion yuan round.
DeepSeek has reportedly hired CITIC Securities for a possible STAR Market listing; it has not confirmed any figures.
The DeepSeek funding round could reach 100 billion yuan, or about $14.9 billion, according to reports on October 6. That would double the 50 billion yuan the Hangzhou lab first set out to raise. Commitments already exceed 80 billion yuan, roughly $11.9 billion, people familiar with the talks said.
DeepSeek has not published any figures for the round. Therefore, Catenaa treats every figure from the talks as reported, not confirmed.
Inside the DeepSeek Funding Round
The round reportedly began in July with a target valuation of 500 billion yuan. At current rates, that equals about $74.6 billion. Tencent and battery giant CATL have reportedly made some of the largest commitments. Strong demand after its latest model release pushed the total well past the first goal, the reports added.
It also follows a first outside raise of about 50 billion yuan earlier this year. Taken together, DeepSeek could pull in up to 150 billion yuan within months. That is roughly $22 billion of fresh capital in a single year.
All dollar figures here use the Federal Reserve’s rate of 6.70 yuan per dollar on October 2, published on FRED.
A Stronger Yuan Adds a Bonus
Currency moves have quietly lifted the dollar value. The yuan traded at 6.99 per dollar on January 2. By October 2, it had firmed to 6.70, a gain of about 4%. As a result, 100 billion yuan now buys around $14.9 billion, against $14.3 billion at January rates.
That swing adds roughly $600 million without a single extra yuan from investors. Conversely, backers paying in dollars now need more of them for the same stake.
Still Small Next to US Rivals
Even at the top end, the round looks modest beside American peers. OpenAI closed a round with $122 billion in committed capital at an $852 billion valuation in March. Two months later, Anthropic raised $65 billion at a $965 billion valuation. Put differently, Anthropic’s valuation is nearly 13 times DeepSeek’s reported target.
By comparison, DeepSeek’s reported 500 billion yuan target is less than 9% of OpenAI’s price. Similarly, its maximum raise equals about an eighth of OpenAI’s haul. That gap frames DeepSeek’s pitch, which has long centred on cost.
Revenue disclosure is another gap. Anthropic said its run-rate revenue crossed $47 billion in May. OpenAI, for its part, cited $2 billion of revenue a month in March. DeepSeek has disclosed no revenue figures, so investors are pricing it largely on its models.
Efficiency Built the Brand
DeepSeek’s reputation rests on doing more with less. Its V3 technical report put the full training run at 2.788 million H800 GPU hours. At an assumed rental price of $2 per hour, that came to about $5.6 million. However, the paper notes the figure excludes earlier research and experiments. It also says each trillion training tokens took about 3.7 days on a cluster of 2,048 H800 chips.
Simple arithmetic shows how far the stakes have moved. The top end of the new round could cover roughly 2,700 runs at that cost. Reports say the cash will instead fund more computing power and new models. One plan reportedly involves a data centre in Inner Mongolia using at least 160,000 Huawei AI chips.
The models keep coming. R1 arrived on January 20, 2025, followed by a V4 preview in April 2026. Most recently, DeepSeek-V4.1-Flash launched on September 10. DeepSeek calls it the smallest model in a new architecture family, with built-in image understanding. It holds 552 billion parameters but activates just 8 billion for input and 16 billion for output.
In other words, fewer than 3% of the model’s weights work on any single step. Consequently, each answer needs far less computing power than the headline size suggests.
What Comes Next
DeepSeek has reportedly hired CITIC Securities to prepare a listing on Shanghai’s STAR Market. Timing and valuation remain undecided, according to the same reports. Meanwhile, investors will watch whether the round closes nearer 80 billion or 100 billion yuan. A listing would also force fuller disclosure, including revenue and costs DeepSeek has so far kept private.
Either way, the raise marks a turn for a lab that reportedly took no outside money until this year. It also shows that Chinese giants such as Tencent and CATL will back homegrown AI at scale. Until the company confirms terms, the DeepSeek funding round remains a work in progress.
