Liang Wenfeng

Liang Wenfeng is a Chinese entrepreneur, born in 1985 in Mililing village, Wuchuan, Guangdong Province, China. He is the founder and chief executive officer of DeepSeek, the open-weights AI research company he established in April 2023, and a co-founder of the High-Flyer quantitative hedge fund whose 2020 and 2021 GPU build-out provided the compute base for DeepSeek's subsequent research. As of May 2026, he leads DeepSeek following the January 2025 release of DeepSeek-R1 and the April 2026 release of DeepSeek-V4, and was named to the TIME 100 AI list and the TIME 100 Most Influential People list in 2025.

At a glance

• Education: Bachelor of Engineering in electronic information engineering, Zhejiang University (2007); Master of Engineering in information and communication engineering, Zhejiang University (2010), advised by Xiang Zhiyu, with a thesis on object-tracking algorithms based on low-cost pan-tilt-zoom cameras.
• Current role: Founder and Chief Executive Officer of DeepSeek, since April 2023; co-founder of High-Flyer since February 2016.
• Key contributions: founder of the open-weights frontier model line including DeepSeek-V3, DeepSeek-R1, and DeepSeek-V4; co-author on the DeepSeek technical reports; co-founder of the High-Flyer Fire-Flyer GPU clusters that funded DeepSeek's compute base.
• Recognition: TIME 100 AI 2025; TIME 100 Most Influential People 2025; Fortune 100 Most Powerful People in Business 2025.
• Public profile: Chinese-language interviews only; no public English-language podcast, conference keynote, or social-media presence as of May 2026. Wikipedia entry: Liang Wenfeng.

Origins

Liang was born in 1985 in Mililing, a village in Wuchuan city, Guangdong Province, in southern China. Both of his parents were primary-school teachers. Wikipedia and Chinese press coverage describe him as a top scorer in the Zhanjiang regional college-entrance examination, which placed him at Zhejiang University in Hangzhou at the age of 17.

He completed a Bachelor of Engineering in electronic information engineering at Zhejiang University in 2007 and continued at the same institution for a Master of Engineering in information and communication engineering, which he completed in 2010. His master's thesis, advised by Xiang Zhiyu, examined object-tracking algorithms based on low-cost pan-tilt-zoom cameras, an early indication of his combined interest in machine learning and engineering-driven cost reduction. Press coverage of his student years emphasizes the same disposition that has characterized his subsequent career: a preference for academic and research work over conventional entrepreneurial paths, including a documented decision to decline an early invitation to join DJI as a co-founder.

Career

After completing his master's degree in 2010, Liang moved to Chengdu and entered the quantitative-trading industry, applying machine-learning methods to algorithmic finance. In 2013 he co-founded Hangzhou Yakebi Investment Management with Xu Jin, and in 2015 he co-founded Hangzhou Huanfang Technology, now known as Zhejiang Jiuzhang Asset Management.

In February 2016 Liang co-founded High-Flyer, the Ningbo-registered, Hangzhou-headquartered quantitative hedge fund, with two Zhejiang University engineering classmates, Xu Jin and Zheng Dawei. High-Flyer's assets under management exceeded RMB 10 billion by 2019 and surpassed RMB 100 billion in 2021. In 2019 the firm commissioned the Fire-Flyer I supercomputer at a reported cost of around RMB 200 million; in 2021 it followed with Fire-Flyer II, a roughly RMB 1 billion deployment built around 10,000 Nvidia A100 GPUs. The clusters were originally configured for trading research and subsequently repurposed for AI work.

DeepSeek was incorporated in April 2023 as a research body inside High-Flyer, with Liang announcing the launch publicly the following month. Press coverage at the time noted that DeepSeek had acquired its initial GPU stockpile, including approximately 10,000 Nvidia A100s, before the October 2022 tightening of US export controls on advanced chips to China. DeepSeek's first models in 2023 and early 2024 were dense and mixture-of-experts language models in the 7-billion to 67-billion-parameter range, released open-weight, alongside specialized DeepSeek-Coder, DeepSeek-Math, and DeepSeek-VL lines.

The late-2024 and early-2025 releases reframed DeepSeek's profile globally. DeepSeek-V3 launched in December 2024 as a 671-billion-parameter mixture-of-experts foundation model with 37 billion active parameters. The accompanying technical report stated that the full training run consumed approximately 2.788 million H800 GPU hours, estimated at $5.576 million at a $2-per-GPU-hour rental rate, a figure that drew substantial industry skepticism over whether the published number reflected total project cost or only the final pre-training run. DeepSeek-R1 followed on January 20, 2025, applying large-scale reinforcement learning to V3 to produce a reasoning model that matched OpenAI's o1 on math, code, and reasoning benchmarks. R1 became the most-downloaded free app on the US Apple App Store within weeks of release, in an event press coverage labeled the "DeepSeek shock". On the same day as the R1 release, Liang attended a closed-door symposium hosted by Premier Li Qiang in Beijing, where he was the only large-language-model-company representative invited to give a speech.

DeepSeek-V4, previewed on April 24, 2026, was the company's first frontier-tier release built and shipped on Huawei's Ascend AI chips rather than the Nvidia-CUDA ecosystem that has dominated frontier training to date. The V4 release coincided with reports that DeepSeek had begun discussions with external investors for a $300 million round at a $10-billion-plus valuation, the company's first external raise after three years of self-funding through High-Flyer profits.

Affiliations

• Hangzhou Yakebi Investment Management: Co-founder, 2013.
• Hangzhou Huanfang Technology (now Zhejiang Jiuzhang Asset Management): Co-founder, 2015.
• High-Flyer: Co-founder, 2016-02 to present.
• DeepSeek: Founder and Chief Executive Officer, 2023-04 to present.

Notable contributions

Liang's body of public work is concentrated on the founding, capital structure, and research direction of DeepSeek and the parent High-Flyer, rather than on individual research authorship at the level of a typical frontier-lab chief scientist. The notable contributions track that emphasis.

• DeepSeek founding (April 2023). Established the company as a research arm of High-Flyer with a mandate of pursuing artificial general intelligence outside both the trading business and the venture-capital fundraising cycle, a structural choice Liang has cited as enabling longer research horizons.
• DeepSeek-V3 (December 2024). The 671-billion-parameter mixture-of-experts foundation model whose technical report documented FP8 mixed-precision training, multi-head latent attention, multi-token prediction, and an auxiliary-loss-free load-balancing scheme for the MoE router. Liang is listed among the contributing authors.
• DeepSeek-R1 (January 2025). The reasoning-optimized model trained from V3 with large-scale reinforcement learning. The accompanying paper introduced the Group Relative Policy Optimization training method and documented R1-Zero, a variant trained with reinforcement learning alone from the base model. R1 forms the lineage for community fine-tunes and distillations on Hugging Face.
• DeepSeek-V4 (April 2026). The 1.6-trillion-parameter mixture-of-experts release built and shipped on Huawei Ascend silicon, accompanied by the smaller DeepSeek-V4 Flash sibling.
• High-Flyer Fire-Flyer GPU clusters (2019 to 2021). The Fire-Flyer I and Fire-Flyer II supercomputers built for High-Flyer's trading research and subsequently repurposed for AI training, including the 10,000 Nvidia A100 deployment that anchored DeepSeek's first generations of model training.
• Open-weights distribution thesis. Every DeepSeek model has been published under an open license with weights and inference code distributed through Hugging Face and GitHub, a structural choice Liang has framed as the company's primary distribution channel rather than a parallel-track to a closed-weights API.
• 36Kr / Waves interview (July 2024, English translation by ChinaTalk November 2024). The long-form Chinese-language interview that has served as the most-cited primary articulation of DeepSeek's research direction, hiring posture, and open-source strategy.
• Recognition. TIME 100 AI 2025; TIME 100 Most Influential People 2025; Fortune 100 Most Powerful People in Business 2025. Added to China's 100 richest list in November 2025 with an estimated net worth of $11 billion.

Investments and boards

• High-Flyer (Software / AI): Co-founder, 2016 to present. Privately held quantitative hedge fund headquartered in Hangzhou with assets under management reported at approximately US $10 billion as of December 2025.
• DeepSeek (AI): Founder and Chief Executive Officer, 2023 to present. Privately held; reportedly 84.3 percent indirect ownership through High-Flyer-affiliated investment vehicles plus a 1 percent direct holding, with a $300 million external round in process at a $10-billion-plus valuation as of April 2026.

No public personal angel-investor activity on record outside the High-Flyer and DeepSeek roles in AI, semiconductors, datacenters, software, or energy as of May 2026. Liang's footprint in this section is concentrated in the founder-and-operator positions at High-Flyer and DeepSeek rather than a parallel investing program.

Network

Liang's principal long-running professional relationships run through the Zhejiang University engineering cohort and the Chinese quantitative-finance industry. His High-Flyer co-founders are Xu Jin and Zheng Dawei, both classmates from his master's program at Zhejiang University; Lu Zhengzhe is the current chief executive of High-Flyer. DeepSeek's research and engineering team has not been publicly profiled in the depth that is typical of US frontier labs. Researchers credited on DeepSeek papers include senior contributors recruited from Tsinghua, Peking, and Zhejiang and from Chinese technology companies. The team's compact size, reported under 200 people in 2025, limits the publicly available picture of senior research leadership beyond Liang.

Liang's documented engagement with peer Chinese technology executives is concentrated in two state-organized events. At the January 20, 2025 symposium hosted by Premier Li Qiang, he was the only large-language-model-company representative present. At the February 17, 2025 private-enterprise symposium hosted by Xi Jinping at the Great Hall of the People, the front-row attendees included Liang alongside Jack Ma of Alibaba, Pony Ma of Tencent, Wang Xing of Meituan, and Zeng Yuqun of CATL. Industry coverage cites comments from Qihoo 360 founder Zhou Hongyi expressing public support for DeepSeek's strategy, though there is no public record of a formal advisory or investor role. Among Chinese-American AI researchers, Kai-Fu Lee is one of the most-cited public commentators on DeepSeek's strategic positioning, though without a documented direct working relationship.

Press coverage has not recorded sustained working relationships with the chief executives of OpenAI, Anthropic, Google DeepMind, Meta, or other US-based frontier labs, and Liang has not appeared at the international AI-policy convenings (Bletchley, Seoul, Paris) where peer chief executives have been visible.

Position in the field

As of May 2026, Liang occupies a structurally distinctive position among the chief executives of frontier-tier AI labs. DeepSeek is the only frontier-capable lab that originated as a research subsidiary of a profitable financial-services firm and remained self-funded for the first three years of its existence. Liang's reported 84.3 percent indirect economic ownership combined with the High-Flyer financing structure produces a capital configuration that has no close parallel among the labs covered on this site, including OpenAI, Anthropic, Google DeepMind, Mistral AI, or peer Chinese labs such as Alibaba Qwen, Moonshot AI, Z.AI / Zhipu, MiniMax, ByteDance Seed, and Tencent Hunyuan.

The V3 and R1 release sequence in late 2024 and early 2025 produced two reframings of US industry consensus. The V3 technical report's $5.576 million training-cost figure for the final training run sat orders of magnitude below contemporary US frontier-lab training-cost estimates and forced public reappraisal of the assumed compute floor for frontier capability. The R1 release demonstrated reasoning capability comparable to OpenAI's o1 from a Chinese open-weights lab, and the corresponding US AI-hardware-equities sell-off, including a Nvidia single-day market-capitalization decline of approximately $600 billion, is widely cited as the moment at which Chinese open-weights frontier capability entered US public-policy conversation.

Liang's public profile is concentrated in Chinese-language press, including the July 2024 36Kr / Waves interview and several shorter Chinese-language interviews and event appearances. He has given no English-language podcast, conference keynote, or major-outlet interview as of May 2026, a posture that distinguishes him from peers including Sam Altman at OpenAI, Dario Amodei at Anthropic, Demis Hassabis at Google DeepMind, Mira Murati at Thinking Machines Lab, and Ilya Sutskever at Safe Superintelligence.

Outlook

Open questions over the next 6 to 18 months:

• Closure of the $300 million external round. Whether DeepSeek's first external raise closes at the reported $10-billion-plus valuation, who the lead and participating investors are, and whether sovereign-fund or strategic-corporate participation appears.
• Successor model release cadence. The timing and capability profile of the next DeepSeek release after V4, the production rollout of V4 Pro and V4 Flash beyond preview, and any successor reasoning model after R1.
• Huawei Ascend integration trajectory. Whether V4 production deployment runs at scale on Ascend with comparable performance and cost economics to Nvidia-based deployment, and whether the integration extends to inference and developer tooling.
• US export-control environment. US AI-chip export controls and any further policy moves directed at Chinese frontier labs may shape DeepSeek's compute supply for training and inference at successor-model scale.
• Open-weights license posture. Whether DeepSeek maintains open-weights distribution for the V4 generation and successor models, and how the license terms evolve as model size increases.
• English-language public profile. Whether Liang gives an English-language interview, conference keynote, or podcast appearance, which would be a notable change from the documented Chinese-language-only posture through May 2026.
• Talent retention. Whether DeepSeek's reportedly compact under-200-person team holds at scale or expands materially as the company moves into its first external-fundraising posture.

Sources

• Liang Wenfeng. Wikipedia biographical entry covering education, family background, the High-Flyer co-founding, and the DeepSeek period.
• High-Flyer. Wikipedia entry on the parent quantitative hedge fund, including the Fire-Flyer GPU cluster builds.
• DeepSeek-V3 Technical Report. The December 2024 arXiv paper documenting the 671-billion-parameter mixture-of-experts model and the reported $5.576 million training-cost figure.
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. The January 2025 paper introducing the Group Relative Policy Optimization training method and the R1 and R1-Zero releases.
• Deepseek: The Quiet Giant Leading China's AI Race. The ChinaTalk November 2024 annotated English translation of the July 2024 36Kr / Waves interview with Liang Wenfeng, conducted by Lily Yu.
• Interview with Deepseek Founder: We're Done Following. It's Time to Lead. The China Academy English translation of the An Yong / Waves interview with additional commentary on price-war strategy and architectural innovation.
• Liang Wenfeng: The 100 Most Influential People in AI 2025. TIME 100 AI 2025 entry.
• Liang Wenfeng: The 100 Most Influential People of 2025. TIME 100 Most Influential People 2025 entry.
• Liang Wenfeng on Fortune's 100 Most Powerful People in Business 2025. Fortune 100 list entry.
• Beijing meeting puts spotlight on China's new face of AI, DeepSeek founder Liang Wenfeng. South China Morning Post coverage of the January 20, 2025 Premier Li Qiang symposium.
• DeepSeek Founder Becomes a Guest of China's Premier, on the Same Day of the R1 Model Launch. Geopolitechs coverage of the January 20, 2025 symposium and its overlap with the R1 release.
• China's Xi urges entrepreneurs to 'show their talents' in sign of support for private business. CNBC coverage of the February 17, 2025 Xi Jinping private-enterprise symposium.
• TechCrunch: DeepSeek previews new AI model that 'closes the gap' with frontier models. April 2026 V4 preview coverage and Huawei Ascend integration context.
• Tech Startups: DeepSeek seeks $300M in first fundraise at $10B+ valuation as AI costs surge. April 2026 funding round context.
• DeepSeek Hugging Face organization page. Public weights and inference code for the DeepSeek model line.
• Photo: text-mode card generated via scripts/make_lab_card.py --text "Liang Wenfeng". No CC-licensed portrait was found on Wikipedia or Wikimedia Commons as of May 2026; commercially licensed images from Getty and Chinese state media (Xinhua) were not used.
• YouTube embed: omitted. Liang has not given an English-language podcast, conference keynote, or interview as of May 2026, and per People.md §8 a news-anchor segment about him is not a substitute for a primary on-camera appearance. Chinese-language interview material exists but is not on a verifiably licensed channel suitable for embed.