DeepSeek Finally Raised: $300M Funding at $10B+ Valuation

Honestly, when I saw this news, my first thought was—finally.

DeepSeek’s funding round has been circulating for over half a year, with no confirmation. The industry speculated whether their model wasn’t attractive to capital, or if there were issues with their technical approach. Then they dropped the bomb: $300 million, valuation over $10 billion.

What does this number mean in the AI landscape? OpenAI just secured $122 billion (though mostly for infrastructure), while DeepSeek landed $300 million. A significant gap, but for a Chinese team, this is a milestone.

The Cost of Independence

I’ve mentioned before that DeepSeek is one of the few “non-aligned” players in this space.

Major players building models either lean on cloud vendors (Alibaba’s Qwen, ByteDance’s Doubao) or hardware manufacturers (Huawei, Xiaomi). The advantage? No worries about compute, scenarios, or funding.

The downside? You’re always working for the cloud vendor, or serving as an accessory for hardware manufacturers. Truly independent products? Difficult.

DeepSeek took a different path. From day one, they pursued independence—open-source models, open-source tech stacks, even publishing their training methods in papers.

The benefits? You have a distinct product identity, a loyal developer community, a genuine technical brand.

The costs? No money.

Training a GPT-4-level model requires over $100 million in compute costs alone. What kept DeepSeek afloat? Early funding + open-source community reputation + a few commercial projects.

But here’s the thing: open source is about passion, not capital returns. Investors ask: where’s your moat? Where’s your business model? How do you avoid losing money?

So this funding round, in some sense, answers those questions.

The “Third Path” for Chinese LLMs

Chinese large models have basically followed two paths:

One is the “ecosystem approach,” like Alibaba’s Qwen. Models are free, but compute, cloud services, and ecosystem products are monetized. It looks busy, but requires a powerful cloud ecosystem foundation.

The other is the “application approach,” like ByteDance’s Doubao. Models are tools; the real products are consumer applications. Revenue comes from applications, subsidizing model development. The problem? Your model capabilities might be constrained by application scenarios, making true generalization difficult.

DeepSeek is attempting a third path: technical service provider.

Models are open-source, but enterprise editions cost money. API services cost money. Customized solutions cost money.

Sounds like a traditional software company model, but in the AI era, does it work?

My personal take—difficult, but possible.

The difficulty lies in this: the large model space evolves too fast. You’re ahead today, behind tomorrow. Open-source communities update at a pace commercial teams struggle to match.

The possibility lies in enterprise services being a slow business. Once customers adopt your solution, migration costs are high. As long as your model doesn’t underperform and service doesn’t falter, you can survive decently.

What Does a $10B+ Valuation Mean?

This valuation level positions DeepSeek among the top tier in the Chinese LLM space.

But I’m more curious—what exactly are investors betting on?

If it’s purely technology, that’s risky. Technology leadership today doesn’t guarantee tomorrow’s advantage.

If it’s the team, that makes more sense. DeepSeek’s team is relatively balanced across technical and commercial capabilities. Their open-source strategy has genuinely built a solid developer ecosystem.

If it’s the sector itself, that’s even more interesting. Among Chinese LLMs, truly independent operations without relying on big tech funding are rare. If DeepSeek proves this path works, it sets an example for the entire industry.

The Challenges Ahead

Securing funding doesn’t mean smooth sailing.

DeepSeek faces intensifying competition.

Domestically, Alibaba’s Qwen just topped global API call rankings, ByteDance’s Doubao is aggressively scaling consumer products, Huawei and Xiaomi are pushing edge-side deployment. Everyone’s scrambling for scenarios, users, and developers.

Internationally, GPT-6 just launched with 40% performance improvements and 2 million token context windows. Anthropic’s new model is on the way. In open source, Llama 4 rumors surface periodically.

Can DeepSeek maintain technical leadership? Can their business model work? Can their independent approach survive?

These questions aren’t solved by funding alone.

Honestly, I hope DeepSeek succeeds.

Not because I’m biased toward Chinese players, but because this space needs genuinely independent voices. Not big tech subsidiaries, not capital puppets—just people doing solid technical work, building products right.

Those are rare these days.

$300 million is a good start. But the road is still long.