HANGZHOU: DeepSeek, the Chinese AI startup that upended the global AI industry in 2025 by building a frontier-class model at a fraction of what U.S. competitors spent, is doing something it has never done before: asking for outside money. The company’s first-ever fundraising round began quietly in mid-April with a $300 million target at a $10 billion valuation. Within days, interest from investors including Tencent and Alibaba pushed that valuation north of $20 billion, according to reporting by The Information. Talks are ongoing, no deal has been finalized, and the terms remain in flux. But one thing is now clear: even the most capital-efficient AI lab in the world has hit a wall.
The Quant Trader Who Built an AI Lab Without Asking for a Dollar
To understand why this fundraise matters, you have to understand why it did not happen sooner. DeepSeek was founded in July 2023 by Liang Wenfeng, a 40-year-old computer scientist who built his career running quantitative trading strategies, not pitching venture capitalists. Liang co-founded High-Flyer Capital Management in 2016 with two Zhejiang University classmates. The Hangzhou-based hedge fund grew to manage roughly $8 billion in assets, and by 2023 Liang was ready for a new challenge: artificial general intelligence.
Unlike many AI founders, Liang never took the VC meeting circuit. DeepSeek was funded entirely by High-Flyer’s balance sheet. That internal backing, combined with an obsessive focus on algorithmic efficiency, produced a string of models that outperformed expectations at a fraction of the cost. When DeepSeek-R1 landed in early 2025, it briefly wiped billions off Nvidia’s market cap and triggered a Washington conversation about whether America’s chip export controls were actually slowing China down. It also sparked a broader reckoning about what the U.S.-China chip war actually means for founders building AI products.
Three years in, the math has changed. Building the next generation of frontier AI models requires datacenter-scale compute investment that even a well-capitalized hedge fund struggles to absorb alone. Competitors including ByteDance, Baidu, and U.S. labs are spending at a pace that makes internal funding increasingly untenable. DeepSeek V4, released today with 1.6 trillion parameters in its Pro configuration and a one-million-token context window, is the clearest signal yet of how much heavier this bet is getting. Bloomberg reports that the round is also intended to set a fair market valuation for employee stock options, a critical retention tool as rival firms aggressively recruit DeepSeek engineers.
What Does DeepSeek’s First Fundraise Mean for Founders Building with AI?
The capital efficiency story has a limit. DeepSeek was the most celebrated example of doing more with less in the AI era, and even it could not hold that line forever. For founders who have built their strategies around the assumption that lean, efficient AI development can outrun the megacap labs, this is a data point worth sitting with.
It does not mean the efficient approach was wrong. DeepSeek’s R1 and V3 models genuinely changed what small teams could do. But those models were built at one stage of the AI cost curve. The V4 generation, with trillion-parameter mixed-expert architectures running at million-token context windows, represents a fundamentally different compute requirement. The same dynamic is playing out across the industry: Amazon’s $33 billion bet on Anthropic and Perplexity’s billion-dollar build both reflect the same reality. Frontier AI now costs frontier money.
For investors, the demand surge around this round tells a parallel story. Tencent reportedly proposed acquiring up to 20% of DeepSeek, a stake the company is said to be reluctant to grant given the control implications. Alibaba is also at the table. The fact that two of China’s largest technology conglomerates are competing to enter DeepSeek’s first round reflects how rare this opportunity is: a proven frontier AI lab, with no prior dilution, at a moment when AI infrastructure is treated as strategic national infrastructure across every major economy. PYMNTS notes that deals like this, with a first-time raise at this scale from a company this established, are structurally uncommon.
What to Watch as DeepSeek’s First Deal Takes Shape
Talks are still active and no terms have been confirmed. The gap between Tencent’s proposed 20% stake and what DeepSeek is willing to offer is likely the central negotiating friction. If DeepSeek holds firm on control, the round could end up structured as a smaller, less-concentrated raise that spreads the capital across a broader investor base to limit any single party’s influence.
The V4 launch today adds another variable. A successful model release strengthens DeepSeek’s hand in negotiations and could push the valuation ceiling even higher. Conversely, it may accelerate timeline pressure. Liang presumably wants this round closed before the market reacts to V4’s benchmarks and revises its expectations further. Watch for official deal announcements in the coming weeks. If Tencent or Alibaba confirm participation, it will mark the first time either tech giant holds a formal stake in China’s most closely watched AI lab.
