Liang Wenfeng is the founder and CEO of DeepSeek (2023–present) and the founder of the quantitative hedge fund High-Flyer Quant (2016–present). He studied information and electronic engineering at Zhejiang University and entered quantitative finance through High-Flyer. He was a central figure in the 2024–2025 surge of Chinese frontier-model development, and DeepSeek's V3 and R1 releases under his leadership reshaped Western assessments of Chinese AI capability.
Background
Liang founded High-Flyer in 2016, and by 2021 it had become one of China's top-tier quantitative trading firms. High-Flyer's GPU procurement, reportedly around 10,000 A100s acquired before US export controls tightened, became the hardware foundation of DeepSeek. Liang spun DeepSeek out in mid-2023 as an independent AI research lab. DeepSeek is structurally unusual among Chinese AI developers: privately funded, research-first, publishing detailed technical reports, and operating with a relatively small team compared with Baidu, Alibaba, or Zhipu.
Liang founded DeepSeek in May 2023 and oversaw the releases of DeepSeek-V2 (2024), V3 (late 2024), and R1 (early 2025). The R1 release triggered what commentators termed the "DeepSeek moment," a reassessment of Chinese frontier capability. In January 2025 Liang met with Premier Li Qiang, an elevation for a private-sector AI founder that observers read as a signal of CCP endorsement.
Liang retains roughly 78 percent of DeepSeek; amid the company's July 2026 talks for a second funding round at an approximately $71 billion pre-money valuation, his net worth reached about $36 billion on July 14, 2026 (Source: techtimes.com).
Positions on AI policy and research
Efficiency over scale
In public interviews in 2024, Liang stressed that DeepSeek's advantage is architectural — mixture-of-experts (MoE), Multi-head Latent Attention (MLA), and training-efficiency innovations — rather than compute scale. CSIS analysis notes that DeepSeek's own framing has been that it could produce equivalent results with 2–4x more compute than US peers, which it presents as vindication of its efficiency thesis while also acknowledging the ongoing compute gap. In a leaked transcript of a May 20, 2026 investor call that circulated on July 23, 2026 — a session running 3 hours and 44 minutes — Liang said China's gap with the U.S. is primarily compute resources rather than talent (Source: luizasnewsletter.com; aiproem.substack.com). He described chip access as the binding constraint in concrete terms: he sought 200,000 Huawei 950 chips for a frontier training run but could obtain only 16,000, against Huawei's expected 750,000-chip output for 2026 (Source: fredgao.com; transformernews.ai).
Open weights
DeepSeek releases V3 and R1 under permissive open-weight terms, described as the most aggressive open-release strategy of any frontier-tier lab as of 2026. Liang has framed open release as core to DeepSeek's identity and to Chinese AI ecosystem strategy. In the leaked May 20, 2026 investor call, he defended open-sourcing as DeepSeek's core strategy and predicted consolidation among AI labs (Source: luizasnewsletter.com; aiproem.substack.com).
Research posture
DeepSeek's technical reports (V3, R1) are detailed by Chinese-lab standards, disclosing training data volumes, architectural choices, and ablations. Liang has publicly discouraged hype and emphasized fundamental research.
Talent and organization
Liang has stated that DeepSeek hires primarily recent graduates from Chinese universities rather than senior researchers returning from abroad, an explicit contrast with the Baidu and Alibaba model.
Relationships
- related: DeepSeek — founder and CEO
- related: DeepSeek-V3, DeepSeek-R1 — flagship releases under his leadership
- related: CSIS on DeepSeek and Huawei — source for the "2–4x more compute" framing
- related: Export Controls on AI — DeepSeek's operating environment
- related: Distillation, Fast-Follow Problem — both frames DeepSeek challenges
- related: Open-Source AI — DeepSeek a principal exemplar
- related: China/US Different AI Races
- supports: efficiency-over-scale research thesis; open-weight strategy