DeepSeek is raising money.
The impact of this news in the AI world may be as significant as any technical release. "Not accepting external investment" has long been one of DeepSeek's most important labels — Liang Wenfeng turned down Sequoia, turned down GGV, turned down nearly every well-known VC in China, and even some state-backed funds were politely declined.
The Information reported this week that DeepSeek is seeking its first round of external funding, targeting at least $300 million at a valuation of over $10 billion. The sources are four anonymous informed persons.
Why Now
This time point is not randomly chosen.
V4 is coming. DeepSeek V4 is expected to be released by the end of this month. The parameter count is expected to jump to the trillion level, with a maximum context window of 1 million tokens. Training and deploying a model of this scale requires massive computing power investment — even for DeepSeek, known for its cost efficiency, real money must be spent.
Huawei migration costs money. DeepSeek is migrating its infrastructure from Nvidia chips to Huawei's Ascend — this is not simply a hardware swap; it requires re-optimizing an entire software stack from the ground up, with considerable engineering costs.
Data centers need expansion. A new data center in Ulanqab, Inner Mongolia, is being planned. Servers, networking, and power — each item costs money.
Retaining talent requires money. DeepSeek has attracted top researchers with compensation levels close to academia, but the market has changed. ByteDance, Alibaba, and Tencent's AI salary packages have reached a very high position, and DeepSeek needs more resources to retain its people.
All of these factors combined mean that High-Flyer's own funds are no longer enough.
What This Means for DeepSeek's Operating Model
DeepSeek's previous logic was clear: the parent company provides funding, the research team focuses on technology, and there is no accountability to outside investors. This allowed them to make many decisions "without considering short-term commercialization" — such as open-sourcing models directly and charging no or only nominal API fees.
Introducing external funds, this logic begins to shake. Even with just $300 million, investors will bring expectations, KPIs, and growth pressure. How compatible these are with DeepSeek's current operating model is a real question.
Liang Wenfeng is certainly aware of this. His choice to seek funding at a time when V4 is about to be released and the company is at its highest negotiating leverage — rather than waiting until "they can't hold on anymore" — is itself a gesture of actively managing the pace. "Active" means: terms can be negotiated, and the people coming in are those I choose.
The $10 Billion Valuation: Is It Reasonable?
DeepSeek has not disclosed financial data, so comparison can only be horizontal reference:
| Company | Latest Valuation | Business Nature |
|---|---|---|
| DeepSeek | >$10B | Foundation model research + API |
| Mistral | ~$6B | European open-source models |
| Cohere | ~$5.5B | Enterprise AI API |
| Zhipu AI | ~$3B | Domestic enterprise services |
$10 billion is not low for a research institution without large-scale commercial revenue, but DeepSeek's technology brand — the global influence of V3 and R1 — and the upcoming V4 provide sufficient support for this valuation.
The real question is who will come in. The Chinese VC market's attitude toward AI foundation models has become highly polarized: top firms have either already bet on competitors or are waiting for a clearer commercialization path before committing. Foreign investment in China's AI sector is also subject to multiple restrictions.
More interesting than "how much money was raised" is "who ultimately came in."
After V4 is released, this answer will likely emerge.
Sources: CocoLoop, DeepSeek reportedly launches first funding round at over $10 billion valuation (CnTechPost / The Information); DeepSeek seeks outside funding for the first time at a $10B valuation (Tech Funding News)