DeepSeek, China's low-cost AI player, reportedly raising $15 billion

100 billion yuan, or about $15 billion. That is the amount DeepSeek is reportedly raising, according to two separate articles from Reuters and Bloomberg published on October 22. It would be one of the largest funding rounds in China.

Both news agencies rely on people familiar with the matter, and DeepSeek has not officially confirmed the deal. According to Reuters, the deal was due to close in the following days and its terms could still change. Bloomberg describes it as completed. Among the largest investors, both agencies name Tencent, the company behind the WeChat messaging app, and CATL, the Chinese battery giant.

A target that doubled along the way

According to Bloomberg, the target was raised at least twice during the talks: 50 billion yuan at the start, then 80 billion, then about 100 billion. According to Reuters, as cited by Next, several investors are reportedly tied to the Chinese state or to major Chinese groups. The Reuters and Bloomberg articles mention no valuation.

Where does investor interest come from? On September 24, The Information reported that DeepSeek's annualized revenue run rate had reached $1 billion, more than double its level a few months earlier. Its chief executive, Liang Wenfeng, reportedly presented this figure to investors. Most of it comes from paid access to its models by companies and developers through its application programming interface (API).

Closing the gap with OpenAI and Anthropic

Despite its size, the sum remains smaller than what the US leaders have raised. According to Bloomberg, as cited by Investing.com, OpenAI and Anthropic have each raised more than $100 billion. This money could help DeepSeek narrow the resource gap. The startup has already increased its spending on data centers and open models.

That is what made it famous: in 2025, DeepSeek showed that competitive models could be built at a relatively low cost, which sent tech stocks tumbling. The company continues to publish the weights of its models, which anyone can download and run on their own servers. It also sells access to its API far cheaper than OpenAI or Anthropic. On October 18, it claimed a cost of less than $0.20 per programming task, based on its own tests.

What to keep in mind in Europe

The funding round comes a few days after Washington made public its ban on DeepSeek in its civilian agencies. For a European company, two uses need to be distinguished. Calling the company's API means entrusting it with your data. Downloading the weights to run the model in-house avoids sending that data, but does not remove the need to evaluate the model before using it.

A possible next step: an IPO in Shanghai, mentioned for 2027 but never confirmed by the company. Until then, we will have to see whether this money translates into new models, and at what price.