DeepSeek seeks funding at reported $45b valuation

Chinese AI lab DeepSeek is in talks to raise its first funding round at a reported US$45 billion valuation.

DeepSeek’s valuation was estimated at about US$20 billion only weeks ago after its AI models drew attention for lower training costs.

Founder Liang Wenfeng controls nearly 90% of DeepSeek.

The company decided to seek funding after rivals tried to poach researchers so staff could get company shares.

China Integrated Circuit Industry Investment Fund may lead the round while Tencent and Alibaba are also in talks.

The fundraising comes as China tries to build local AI with limited access to US chips.

🔗 Source: TechCrunch

🧠 Food for thought

Implications, context, and why it matters.

The fundraising valuation tracks V4. DeepSeek says it performs on par with top models from OpenAI and Anthropic, an AI startup backed by Amazon and Google, on benchmark tests 1.
DeepSeek prices V4 far below similar US models. V4-Pro costs US$1.74 per million input tokens, while V4-Flash costs about US$0.14 1.
That price edge comes from a compute-efficient architecture such as mixture-of-experts (MoE), a design that activates only part of the model for each task. This can cut computing needs during inference, the stage when a trained model generates answers 2.
In a 1-million-token context, DeepSeek says V4-Pro uses 27% of the computing power and 10% of the memory required by V3.2 1.

DeepSeek’s rise is seen as an early sign that China is building a parallel AI stack in response to US export controls on advanced chips 1.
V4 is DeepSeek’s first model tuned for Chinese chips such as Huawei’s Ascend series. DeepSeek still appears to rely on Nvidia in part, since its technical report says Chinese chips handle inference while training may still depend mainly on Nvidia chips 1.
DeepSeek also releases open-weight models, meaning trained parameters others can use or fine-tune and deploy. That broadens AI development beyond US dominance 2.
The strategy weakens the idea that AI leadership depends on exclusive access to the most advanced US hardware, and it puts more weight on algorithmic efficiency. China’s AI push still is not fully autonomous and has relied on Nvidia hardware 2.