Liang Wenfeng

From Zhejiang to a hedge fund

Liang was born in 1985 and grew up in Guangdong. He studied electronic information engineering at Zhejiang University and took a master's degree in information and communication engineering there in 2010. In 2015 he co-founded High-Flyer (幻方量化).345612

Its stock of graphics chips grew step by step, he told Waves: 100 cards in 2015, 1,000 in 2019, then 10,000. In 2021 High-Flyer built Fire-Flyer II, a 1-billion-yuan computing cluster that Waves reported held about 10,000 Nvidia A100s.1214

DeepSeek

High-Flyer announced a separate large-model company in May 2023, and Liang founded DeepSeek (深度求索) that July. He was plain about the aim. "我们要做的是通用人工智能，也就是AGI," he told Waves. In English: what we want to do is general artificial intelligence, that is, AGI.41415

DeepSeek-V2, in May 2024, cost about 1 yuan per million tokens and set off a price war among Chinese AI firms. Liang said the price left only a small margin above cost, with no subsidy. DeepSeek-V3 followed in December 2024. Its technical report put the official training run at 2.788 million H800 GPU hours, or $5.576 million at an assumed $2 an hour, not counting earlier research. Then came R1, a reasoning model posted on arXiv in January 2025 and published in Nature that September, with Liang as corresponding author.39101322

He has kept the models open, saying in 2024 that DeepSeek would not switch to closed source. He frames the contest with the US as one of originality. In ChinaTalk's translation of his 2024 interview: "We often say that there is a gap of one or two years between Chinese AI and the United States, but the real gap is the difference between originality and imitation. If this doesn't change, China will always be only a follower." And on chips: "Money has never been the problem for us; bans on shipments of advanced chips are the problem."3

Beijing takes notice

In January 2025 Liang spoke at Premier Li Qiang's symposium on the draft government work report. Weeks later he was at Xi Jinping's meeting with private entrepreneurs in Beijing. Nature named him among the ten people at the heart of 2025's big science stories, and TIME put him on its TIME100 list.11161718192728

In June 2026 DeepSeek closed its first outside funding round, of more than 50 billion yuan. Reuters reported that Liang put in about 20 billion yuan himself.2021

Disputes over DeepSeek

The disputes below concern DeepSeek, the company. Its cost figures are contested: research firm SemiAnalysis estimated DeepSeek's hardware spending at about $1.6 billion and its fleet at about 50,000 Nvidia Hopper GPUs, Tom's Hardware reported. DeepSeek's own Nature supplement put R1's reasoning training at about $294,000 and disclosed that A100 chips were used in preparatory work.10232930

In a February 2026 memo to the US House Select Committee on China, OpenAI accused DeepSeek of trying to "free-ride" on US labs' models through distillation, Reuters reported. DeepSeek and High-Flyer did not respond to Reuters. In its Nature supplement DeepSeek said it did not intentionally include OpenAI-generated data, while noting that crawled web pages contain answers generated by OpenAI models.72331

In April 2025 the House Select Committee on the CCP alleged that DeepSeek sends US user data to China, censors answers in line with Communist Party positions, was trained on material unlawfully obtained from US models and may have used export-controlled Nvidia chips. Its release spelled his name "Lian Wenfang". We found no response from DeepSeek in our sources. In June 2025 a senior US State Department official told Reuters that DeepSeek supports China's military and intelligence work. Reuters said it could not independently verify the procurement data.832

Regulators have moved against the app. Italy's data regulator limited DeepSeek's processing of Italian users' data on 30 January 2025. The companies said they had not entered the Italian market. South Korea's privacy regulator found in April 2025 that DeepSeek had sent users' data abroad without consent, and DeepSeek stopped the prompt transfers on 10 April. Berlin's data protection commissioner reported the app to Apple and Google in June 2025, and Australia and Taiwan barred it from government devices, Reuters reported.3334353637

Liang, year by year

1. 2015

   Co-founded the quantitative hedge fund High-Flyer (幻方量化).4512

2. 2021

   High-Flyer built its Fire-Flyer II computing cluster, a 1-billion-yuan project that 36Kr's Waves reported held about 10,000 Nvidia A100 GPUs.1214

3. 2023

   High-Flyer announced a separate large-model company, DeepSeek (深度求索), in May; Liang founded it that July.41415

4. May 2024

   DeepSeek released the open-source DeepSeek-V2 at about 1 yuan per million tokens, setting off a price war among Chinese AI firms.3

5. 20 Jan 2025

   Spoke at Premier Li Qiang's symposium on the draft government work report in Beijing.1116

6. 22 Jan 2025

   DeepSeek posted the DeepSeek-R1 reasoning model paper on arXiv.9

7. 17 Feb 2025

   Attended Xi Jinping's symposium with private entrepreneurs in Beijing.1718

8. 17 Sep 2025

   The DeepSeek-R1 paper was published in Nature, with Liang as corresponding author.13

9. Dec 2025

   Named in Nature's 10, the journal's list of people at the heart of the year's big science stories.19

10. Jun 2026

    DeepSeek closed its first external funding round, of more than 50 billion yuan; Reuters reported Liang put in about 20 billion yuan himself.2021

Questions people ask about Liang Wenfeng

Liang Wenfeng on Kenning

We haven't written about Liang Wenfeng yet. New stories will appear here.

DeepSeek on Kenning