DeepSeek raises about $15B from Tencent, CATL and others at a roughly $90B valuation

DeepSeek has long been an outlier among frontier labs, funded largely by its parent quant fund High-Flyer rather than outside capital. A roughly 100-billion-yuan round, about $15B, from strategic investors including Tencent and battery maker CATL marks a sharp change. At a reported valuation of about $90B, DeepSeek still prices far below US peers such as OpenAI ($1.25T target) and Anthropic ($1.5-2T IPO range), despite competitive open models.

The investor mix is telling. Tencent brings distribution across WeChat and cloud. CATL signals industrial and energy interest in AI, which matters as power becomes the constraint on compute everywhere. The money should fund training compute that has been constrained by US export controls, likely through domestic accelerators.

On the product side, DeepSeek upgraded V4.1 Flash to revision 1017. Tom's Hardware reports the revision uses about a quarter of the HBM and an eighth of the SSD storage of comparable setups, but remains more prone to hallucination than proprietary models in agentic workloads. Its larger V4-Pro uses 1.6T total parameters with 49B active.

The headwind is Washington. The same week, DHS banned DeepSeek products across federal civilian agencies, including derivative tools. Watch how the raise translates into training scale, and whether hallucination rates improve enough to win agentic production work outside China.