DeepSeek has first brought the attention of the world to China’s AI capabilities, and it’s now officially China’s most valuable AI startup.
The Hangzhou-based lab has closed its first-ever external funding round, raising more than 50 billion yuan — approximately $7.4 billion — at a post-money valuation exceeding $50 billion. This makes DeepSeek China’s most valuable AI startup.
Until this week, DeepSeek had operated entirely on capital from High-Flyer, the quantitative hedge fund that founder and CEO Liang Wenfeng co-founded in 2015 out of Zhejiang University. The decision to take outside money at all marks a significant shift for a company that had made a point of not needing it. But the structure Wenfeng chose makes clear he isn’t interested in sharing control. Investors were required to funnel their capital not into DeepSeek itself, but into a limited partnership managed directly by Wenfeng. They get a five-year lock-up and no voting rights. Wenfeng personally contributed around 20 billion yuan — roughly $3 billion — making him the round’s single largest contributor by a wide margin.
The only exception to the no-voting-rights structure is China’s National Artificial Intelligence Industry Investment Fund, which received direct equity, voting rights, and freedom from the lock-up. The explicit carve-out for the state-backed fund formalizes what was already widely understood: DeepSeek operates with Beijing’s backing, and the governance structure now makes that official.
Among the commercial investors, Tencent committed around 10 billion yuan and battery maker CATL contributed approximately 5 billion yuan. Tencent’s interest is defensively rational — it has its own AI platform in Hunyuan, but DeepSeek’s models have been pulling developer traffic at a scale that Hunyuan cannot match. Sitting out would have been the costlier choice. CATL’s participation reads differently — it’s a strategic energy-infrastructure play, not a pure AI bet. The company has been expanding into AI data centers and power storage, and an anchor position in China’s most prominent AI lab serves that ambition well.
The $50 billion valuation is a sixfold increase from where DeepSeek was valued in early 2025, and it arrives as the company’s real-world usage numbers have grown to match the hype. DeepSeek commands 16.3% of all token volume on OpenRouter, more than any other single provider. Its V4-Pro model is priced at a fraction of what frontier closed-source labs charge and scores second among open-weights models globally on the Artificial Analysis Intelligence Index. The cost advantage has been translating into real adoption decisions — the CEO of AI agent platform Lindy recently switched entirely to DeepSeek V4, citing millions of dollars in savings and better performance on core use cases.
That trajectory back to the top of the rankings after a competitive stretch makes the funding round feel like it arrives at the right moment. DeepSeek’s R1 release in early 2025 was what put the company on the map globally, triggering a $600 billion single-day drop in Nvidia’s stock price and briefly making it the most downloaded free app in the US App Store. The model matched OpenAI’s o1 on reasoning benchmarks at a reported training cost of around $6 million, at a time when American labs were spending hundreds of millions for comparable results.
The gap between DeepSeek and its Western counterparts in raw funding terms is still substantial. Anthropic closed a $65 billion Series H last month at a $965 billion valuation. OpenAI secured $122 billion at an $852 billion valuation earlier this year. At $50 billion, DeepSeek is operating in a very different financial weight class — but the comparison matters less given the lab’s open-weights strategy and dramatically lower compute costs. Publishing model weights freely has allowed DeepSeek to build a developer ecosystem without needing to run the same compute-intensive infrastructure as its American rivals, and Western chip export restrictions have reinforced the company’s incentive to stay efficient.
The fresh capital is expected to go toward new model development, compute infrastructure, and talent. DeepSeek’s V3.2-Speciale, released late last year, delivered gold-medal-level performance at competitive mathematics olympiads. V4-Pro followed in April. The lab’s own assessment — that it trails frontier closed-source models by roughly three to six months — is the kind of honest framing that has kept the industry from underestimating it twice.
What’s also become apparent from this round is that China’s AI ecosystem has consolidated behind DeepSeek in a way it hasn’t behind any other domestic lab. Chinese models broadly now account for roughly 44% of token volume across the top 10 providers on OpenRouter, but DeepSeek is the one with the valuation, the state backing, and the investor syndicate that spans platform giants, industrial conglomerates, and government capital. That combination doesn’t produce a startup in the conventional sense anymore. It produces a national champion.