Liang Wenfeng's AI high-stakes gamble: raised 50 billion yuan, why can't he keep people?

Author: Zeno ; Editor: Zeno

In October 2023, DeepSeek's financing plan first surfaced, causing a direct stir in the capital markets. 50 billion yuan—this number not only broke the record for a single-round funding of domestic AI large models but also pushed the usually low-profile Liang Wenfeng and his team into the spotlight.

Just a year ago, Liang Wenfeng was confidently shouting the "Three No Principles": no fundraising, no going public, no commercialization. Now, with DeepSeek embracing capital, is it out of necessity or a proactive breakthrough?

Luxurious financing lineup, valuation soaring to 350 billion

According to disclosures, in this round of financing, DeepSeek contributed 20 billion yuan, becoming the largest "financier." Other major players also stepped in:

· Tencent invested 10 billion yuan

· CATL invested 5 billion yuan

· JD.com, NetEase, and IDG Capital each contributed 3 billion yuan

· The National AI Industry Investment Fund also entered, investing 1 billion yuan and holding voting rights

After the financing was completed, DeepSeek's total valuation surpassed 350 billion yuan, directly ranking it among the top domestic AI unicorns.

From top scorer in college entrance exams to quantitative genius, he has long "stockpiled supplies"

Liang Wenfeng's story itself is a classic underdog tale.

In 2002, he scored 806 points to become the top scorer in Zhanjiang City’s college entrance exam, declined Tsinghua University, and chose Zhejiang University’s Electronic Information Engineering as his first choice. During college, he became obsessed with computers, and after graduation, he started with 80k yuan, diving into quantitative trading.

In 2015, he founded Fantasia Quant, and by 2019, it became China's first quantitative hedge fund to raise over 80k yuan, with annual net profits reaching 5-6 billion yuan.

But what truly allowed DeepSeek to catch up later was Liang Wenfeng’s "foresight"—starting in 2021, Fantasia heavily stockpiled NVIDIA A100 GPUs. By 2023, only five companies in China owned over ten thousand GPUs, and Fantasia was the only non-internet giant with such a large "cardholder" base.

This computing power became the confidence behind the emergence of DeepSeek V3.

After the explosion, crises followed one after another

December 2024 saw the release of DeepSeek V3, which gained over 100 million users in 7 days, with growth outpacing ChatGPT (which took two months), becoming a phenomenon-level dark horse among domestic large models.

However, after the hype, troubles also arose.

❶ Talent loss, no equity to retain people

Luo Fuli was poached by Lei Jun to Xiaomi, Guo Daya jumped to ByteDance, and at least five core R&D personnel left one after another. The AI talent war intensified, and without equity or options incentives, DeepSeek had no advantage in retaining talent.

❷ Computing power race, endless spending

Large models are no longer just about algorithms but a hardware arms race. OpenAI, ByteDance, Alibaba spend billions annually on chips, data centers, and electricity. The new version of DeepSeek requires 2-3 times the computing power of V3, with a single H100 GPU costing over 200,000 yuan, and expansion costs starting at over a billion yuan. The early stockpiled A100s are now "old assets," and the profits from quant trading are simply not enough to fill this bottomless pit.

❸ The IPO window waits for no one

After the listings of Zhipu AI and MiniMax, their stock prices soared, and Jié Yuè Star also is preparing for IPO. If DeepSeek misses this round of capital dividends, its valuation in the primary market could directly shrink or even collapse.

Kaola Finance’s observation: Compromise is for survival

The original "Three No Principles" were idealistic, but reality is harsh.

Liang Wenfeng once wanted to grow slowly with his own funds, but the AI large model track has long turned into a brutal game of burning money to buy time and scale. No fundraising means falling behind in computing power, talent loss, and valuation shrinkage—each step is a dead end.

Now, DeepSeek’s choice to embrace capital is not a "betrayal of the original intention," but a sober recognition of industry rules. Once 50 billion yuan is in place, whether it can truly transform into technological barriers and a commercial closed loop is the key to the second half.

After all, raising money is just an entry ticket; creating the future through burning is the real skill.