Today's Rumors About DeepSeek's Funding Are Quite Interesting

@BardKidd: Today's rumors about DeepSeek's funding are quite interesting. Alibaba's investment condition was to bind the ecosystem, which was rejected; Tencent's investment condition was to take a 20% stake, which was also rejected. Then Liang Wenfeng took out 20 billion out of his own pocket.

DeepSeek must have been under a lot of pressure over the past year. First, there's the rapid advancement of foreign closed-source models, with their leading edge growing larger and larger, all featuring multimodal + Agent capabilities and increasingly comprehensive abilities.
DeepSeek, with just a simple model, can no longer create the massive stir it did upon its initial release.

Then there's the issue of talent. DeepSeek R1 core researcher Guo Daya left for ByteDance.
Luo Fuli, the "post-95 genius girl" and V2 core developer, left for Xiaomi.
Wang Bingxuan, core author of the first-generation large model, left for Tencent.
Ruan Chong, a core contributor to multimodal capabilities, left for Yuanrong.
And who knows how many others whose names we don't even know...

Domestic peers are showing no mercy. In 2026, the average monthly salary for AI positions will exceed 60,000 RMB. No matter how rich Liang Wenfeng's quant fund is, it can't outspend Tencent.

Moreover, DeepSeek has never raised funding before, so what could they use to issue employee stock options? What true AI master with real technical skills wouldn't want to get rich? Just collecting a salary here—are they crazy? Therefore, this 50 billion is essentially half for computing power and half for an "option pool" to lock in talent.

Another shift with DeepSeek V4 is the complete transition from the Nvidia CUDA ecosystem (as mentioned in previous videos) to Huawei Ascend. This was also forced by circumstances. With US data centers and compute cards flourishing everywhere, continuing to rely on Nvidia cards means the new models are bound to fall behind, facing the embarrassment of being outdated the moment they are released. It's better to choose a shortcut and create the world's first model that doesn't rely on Nvidia; at least this way, there's a narrative.

From another perspective, if they successfully figure out Huawei's compute cards, it also carves out a survival path for other domestic models; otherwise, they will never surpass US models.

In the history of AI financing, it's unheard of for a founder to inject 40% of their personal funds.
Liang Wenfeng is now using 20 billion to retain nearly 100% of the voting rights. The company is entirely his. Purely from the perspective of a future IPO, this is not an ideal equity structure in the eyes of investors. Presumably, other major Chinese tech companies won't be very willing to collaborate deeply with DeepSeek later on, because as mentioned earlier, big companies want either an ecosystem or a stake in a partnership; otherwise, there's no need.

Indirectly, this also shows that Liang Wenfeng's personality is likely low-key yet ruthless. Of course, without being ruthless, he couldn't have created DeepSeek.
At the same time, this is also why rumors say DeepSeek ultimately accepted money from state-backed funds.

DeepSeek is reportedly valued at 350 billion currently, second in China only to Zhipu, without having disclosed a single cent of commercial revenue. Why is a company with no clear business model worth so much?

Because it has proven that without relying on any domestic internet giant, without relying on international capital, and even without relying on Nvidia, it can still build world-class models.