DeepSeek's First Funding Round: Tencent Wants 20%, Valuation Hits $20 Billion

It took less than 48 hours for DeepSeek’s opening ask to become obsolete.

When the Chinese AI lab first began approaching prospective investors in late April, it sought to raise around $300 million at a valuation of at least $10 billion — a respectable number for a startup that had, until then, operated without any outside capital. But investor demand was so overwhelming that the valuation had already more than doubled to above $20 billion before formal term sheets were even circulated, according to multiple people familiar with the matter reported by The Information and Bloomberg.

Now Tencent Holdings is proposing to acquire a stake of as much as 20% in the company. Alibaba Group is also engaged in discussions. And DeepSeek — the lab that triggered a market selloff in January 2025 when its open-source R1 model outperformed American rivals at a fraction of the cost — finds itself navigating its most consequential business decision yet: who to let in the door, and on whose terms.

From Quant Fund Side Project to National Asset

DeepSeek’s origin story is unlike any other AI lab’s. It was founded not by academics or ex-Google engineers, but by High-Flyer Capital Management, a Hangzhou-based quantitative hedge fund that has quietly accumulated one of China’s largest private GPU clusters over several years. High-Flyer, run by founder Liang Wenfeng, initially built the AI research operation to improve its trading algorithms. It was never meant to become a globally consequential AI company.

Then came the R1 moment. In January 2025, DeepSeek released its R1 reasoning model and claimed it had been trained for roughly $6 million using a cluster of NVIDIA H800 chips — far below the $100 million-plus typically associated with frontier model training at that time. Nvidia stock fell 17% in a single day. Silicon Valley went into a minor panic. The narrative that American AI dominance was permanent and well-funded suddenly seemed shakier.

Over the following year, DeepSeek continued releasing competitive models, culminating in the V4 series launched in late April 2026 with a 1.6-trillion-parameter Pro variant and a lighter Flash version. Deployed in partnership with Huawei’s Ascend 950 chips, DeepSeek V4 topped multiple open-source coding benchmarks and demonstrated significant advances in reasoning and agentic task completion. The lab had done what most thought impossible: matched frontier American models while operating under US semiconductor export controls.

Why Tencent Wants 20%

Tencent’s aggressive opening position — seeking as much as 20% of the company — reflects a strategic calculus that goes beyond financial return. For China’s internet and gaming giant, owning a meaningful stake in the country’s leading AI research lab would provide several advantages.

First, it would give Tencent a credible AI research partner at a moment when every major Chinese tech company is racing to build or buy AI capabilities. Baidu has its Ernie series, Alibaba has Qwen, and Tencent itself has been developing its Hunyuan models — but none has achieved DeepSeek’s international recognition or the kind of efficiency breakthroughs that have made it a genuine alternative to Western frontier models.

Second, a 20% stake would give Tencent real influence over DeepSeek’s product roadmap and distribution strategy. WeChat, Tencent’s social and payments super-app, has over 1.3 billion monthly active users in China. Embedding DeepSeek’s capabilities into WeChat’s ecosystem would represent an extraordinary distribution advantage.

DeepSeek, however, is reportedly reluctant to cede that level of control. The company appears wary of becoming too dependent on any single strategic investor — a concern that is well-founded given how Tencent’s previous strategic investments have tended to integrate acquired companies into its broader ecosystem over time.

Alibaba’s Position and the Competitive Subtext

Alibaba’s interest adds a competitive dimension to the deal that neither party would prefer to acknowledge publicly. The e-commerce and cloud giant has invested heavily in its own AI capabilities through the Qwen model family, which competes directly with DeepSeek in some enterprise applications. Investing in DeepSeek would represent either a hedge against its own internal AI efforts or an acknowledgment that DeepSeek’s research capabilities are difficult to replicate internally.

Alibaba’s Tongyi Cloud unit already provides inference infrastructure for some of DeepSeek’s API traffic in China — a relationship that creates natural alignment but also potential conflict of interest as both companies develop competing products.

The terms of Alibaba’s proposed involvement have not been disclosed. Unlike Tencent’s reported 20% proposal, no specific stake size has been attributed to Alibaba’s discussions.

The Valuation Calculus

A $20 billion-plus valuation places DeepSeek in rarefied company among global AI startups. For context, Mistral AI — the French lab that has become Europe’s leading open-source model developer — was valued at around $6 billion in its most recent funding round. Cohere, the Canadian enterprise AI company, has been valued in a similar range.

The comparison point that matters most, however, is with American frontier labs. OpenAI raised at a $300 billion valuation in its most recent round. Anthropic’s Google investment implies a valuation of tens of billions. xAI has been valued at over $50 billion. By those standards, DeepSeek at $20 billion might look like a relative bargain — which partly explains why domestic tech giants are moving so aggressively.

There is also a geopolitical dimension. As US-China tech tensions have intensified, Chinese regulators and policymakers have made clear that they view domestically developed frontier AI as a strategic priority. Having DeepSeek remain independent — or fall into the hands of a Chinese corporate backer rather than foreign investors — aligns with broader government priorities around AI sovereignty.

What Remains Unresolved

As of the time of publication, no deal has been finalized. The terms, valuation, and structure of the round are still fluid. Representatives for DeepSeek, Tencent, and Alibaba have not commented publicly.

Several questions remain open. Will DeepSeek accept a 20% stake offer from Tencent, or negotiate a smaller position? Will it take investments from multiple parties to preserve balance? Will it seek to maintain the operational independence that has enabled its research breakthroughs?

The answers will shape not only DeepSeek’s trajectory but also the structure of China’s AI ecosystem for years to come. A startup that built its reputation on doing more with less — on scrappy efficiency over brute-force spending — is now navigating a funding negotiation that could define its soul.

High-Flyer’s Liang Wenfeng has long projected an image of a researcher-operator who prioritizes scientific progress over commercial ambition. The $20 billion question is whether that image survives contact with Tencent’s strategic interest and Alibaba’s competitive calculations.