Author | Zhiyu For more financial insights | BT Financial Data Pass Full text: 2,287 characters | Approx. 9-minute read

In May 2026, a figure who had long shunned external funding announced he would personally inject 20 billion yuan into his company’s Series A round.

Meet Liang Wenfeng, founder and CEO of DeepSeek.

That 20 billion yuan? It accounts for 40% of the round’s total 50 billion yuan raise. (Source: Sina Finance, May 10, 2026)

This isn’t a token 1-2% “confidence vote” from a founder. This is a founder staking a substantial portion of his personal fortune on his own venture.

Such a move is unprecedented in China’s AI sector.

What’s driving this high-stakes gamble?

This isn’t just another funding round—it’s Liang Wenfeng’s 20 billion yuan declaration to the market.

1

Who is Liang Wenfeng—and why does his “self-funding” matter?

To grasp the weight of 20 billion yuan, you must first understand Liang Wenfeng’s background.

He’s no typical AI entrepreneur—no BAT pedigree, no U.S. tech giant resume, minimal media presence, and rare public speaking engagements.

Liang’s expertise lies in quantitative investment.

DeepSeek grew out of High-Flyer Quant, the quantitative hedge fund he founded. A leader in China’s quant circles, High-Flyer boasts stable performance and healthy profits. It’s this financial backbone that let DeepSeek operate with extreme minimalism for years—rejecting financing, commercialization, and roadshows. Its R&D team of just 270 has built globally influential open-source large models. (Source: Synced Review, May 9, 2026)

This model is an anomaly in AI.

Most large model firms follow a playbook: raise funds, build teams, research, then raise more. DeepSeek flipped the script: research first, using its own cash, no external capital.

So why is Liang now investing 20 billion yuan of his own? What’s changed behind the scenes?

2

What’s shifted—and why now?

Quant hedge fund veterans don’t act on impulse. They understand risk better than most—never putting all eggs in one basket.

Liang’s 20 billion yuan decision must stem from cold, hard calculations.

The backdrop? A seismic shift in DeepSeek’s competitive landscape.

Domestically, ByteDance and Alibaba are pouring billions into AI, while upstarts like MiniMax and Yuezhi Anmian aggressively raise funds—Yuezhi just closed a $2 billion round, valuing it at $20 billion post-money, with ARR topping $200 million in April. (Source: Huafeng Capital, May 7, 2026)

Globally, Anthropic commands a valuation north of $900 billion, and OpenAI remains the pace-setter.

DeepSeek’s earlier “no-financing” stance was a deliberate strategy—when large model R&D costs were low, self-funding ensured absolute independence. But as competition shifts from a “research race” to an all-out war of “compute + talent + commercialization,” quant fund cash flow alone can’t keep pace.

Hence, the need to raise external capital.

But why 40% from his own pocket?

Because the signal matters as much as the money.

3

20 Billion Yuan Buys More Than Compute—It Buys Pricing Power

Had Liang made only a token co-investment or sat out entirely, external investors would’ve dictated pricing and strategic constraints.

His 20 billion yuan self-investment sends a clear message:

“I trust this company’s vision more than any external investor. I’m willing to bet my fortune here—so the valuation you propose must align with my logic.”

It’s a founder’s bargaining strategy at its finest. National funds, Tencent, Alibaba can all join the round, but their sway is naturally checked by Liang’s stake.

At a post-money valuation of 350 billion yuan (~$51.5 billion), DeepSeek’s valuation has quintupled from the $10 billion figure reported three weeks earlier when financing news first broke. (Source: Wall Street See, May 8, 2026)

With his 20 billion yuan, Liang has secured the right to set the tone within this valuation bracket.

4

The Risks: What’s He Betting On—and What Could Go Wrong?

Of course, a 20 billion yuan personal investment isn’t risk-free.

DeepSeek’s current monetization is relatively simple—primarily API call fees. While its open-source strategy has built a vast developer ecosystem, it’s also led many enterprise clients to self-deploy, diverting revenue from paid APIs.

Using the industry’s typical 100x price-to-sales ratio, a $51.5 billion valuation would require ~$500 million in annual revenue to justify. DeepSeek hasn’t disclosed its actual revenue, and whether it can scale rapidly in the next 12 months is the biggest question mark. (Source: Huxiu, May 10, 2026)

Liang is betting on the next 12 months—V4.1 launches in June, followed by enterprise tools and MCP protocol support. Can DeepSeek evolve from the “strongest open-source model” into a “commercially viable company with real revenue” this year?

If yes, the 20 billion yuan could be the smartest bet of his career.

If no, it’ll be the priciest personal “research project” tuition he’s ever paid.

5

What This Means for China’s AI Industry

China’s AI sector has long harbored a quiet concern: soaring large model valuations rest more on external investor confidence than founders’ genuine skin in the game.

When founders pitch, “We’re the AI era’s infrastructure,” investors buy in and assign high valuations. But have the founders themselves bet their own money? Often, no.

Liang’s 20 billion yuan shatters that unspoken rule.

It’s the first time a Chinese AI founder has backed his judgment with such massive personal capital.

Regardless of DeepSeek’s ultimate fate, this move alone has reshaped the credibility framework for China’s AI entrepreneurs.

The strongest vote of confidence isn’t saying, “I believe this will succeed”—it’s putting your money where your mouth is, then staying silent.

Disclaimer: This article is for information sharing and industry analysis only. It does not constitute investment advice, analysis, or solicitation. The market carries risks; invest cautiously. Any decisions based on this article are at your own risk, and the author/platform assumes no legal liability.

This is an original piece by BT Finance. Unauthorized use, reproduction, dissemination, or adaptation is prohibited. Infringement will be pursued legally.