Liang Wenfeng

梁文锋（中文：梁文锋；拼音：Liáng Wénfēng；生于1985年）是一位中国企业家和商人，共同创立了量化对冲基金幻方量化，随后创立并领导人工智能公司DeepSeek。他以将机器学习应用于金融领域而闻名，后来开发了极具成本效益的开源AI模型，与价格高昂的西方系统相抗衡。DeepSeek在2024-2025年间因开发高性能、低成本的大型语言模型而获得全球关注。截至2026年，他被认为通过高效的工程而非庞大的计算预算挑战了美国在AI领域的主导地位。梁文锋的高效方法——强调架构创新而非蛮力计算——重塑了全球对AI开发成本的假设，并将DeepSeek定位为重要的颠覆者。他被描述为“极客”而非典型老板，拥有卓越的技术深度、快速学习能力和强大的资源调动能力。他是金融和AI领域雄辩的主题演讲者。梁文锋继续领导DeepSeek走向通用人工智能（AGI），专注于架构效率、开源发布和成本降低。该公司已扩展了负担得起的API服务以及编码/专用模型。2025年，他参加了高层政府座谈会，显示出政策影响力。最近的报道强调DeepSeek在引发AI价格战和挑战硅谷假设方面的作用。他保持极其低调，似乎没有官方公开的社交媒体账号。DeepSeek在GitHub、Hugging Face和中国社交媒体等平台上保持企业存在。没有可用的个人粉丝数。[编辑]

梁文锋于1985年1月1日出生在中国广东省湛江市吴川市覃巴镇的小村庄米历岭村。父母都是小学教师，家庭注重教育。当地人和老师从小就称赞他；他以阅读漫画书同时数学成绩优异而闻名。他就读于吴川一中，成绩名列前茅，自学了高中和大学水平的数学，并在17岁时提前离校进入浙江大学。他于2007年获得浙江大学电子信息技术学士学位，2010年获得信息与通信工程硕士学位。他的硕士论文聚焦于使用低成本PTZ摄像头的AI监控系统的目标跟踪算法。2008年读硕期间，梁文锋和一些同学开始抓取金融市场数据并尝试用机器学习进行交易预测。这些早期实验虽然最初不成功，但为量化金融奠定了基础。2013年，他与浙大校友徐进共同创立了第一家投资公司——杭州雅可比投资管理有限公司。这标志着他从工程正式转向AI驱动的交易。梁文锋在20世纪80-90年代成长于广东一个五线城市，当时许多当地人优先考虑快速赚钱而非教育。他父亲作为教师的角色在周围怀疑的氛围中灌输了一种强烈的学术精神。到了中学，梁文锋展现出非凡的数学天赋。在浙江大学，他开始将机器学习应用于现实问题。[编辑]

梁文锋的突破来自将AI与金融融合。2015年，他共同创立了杭州幻方科技，并与大学同学开始了正式的量化交易业务。2016年，他与两位工程同学共同创立了宁波幻方量化投资管理合伙企业（幻方量化）。该公司通过将AI和机器学习整合到交易策略中迅速扩张。到2019年，幻方量化建立了自己的AI超级计算集群；管理资产达到约100亿元人民币（约15亿美元）。该基金在2019-2023年面临中国监管压力，促使其将计算基础设施转向AI研究。到2021年，幻方量化管理资产超过1000亿元人民币（约140亿美元），梁文锋通过管理费和业绩分成成为亿万富翁。2023年，梁文锋创立了DeepSeek作为幻方量化的AI研究部门，利用该基金现有的基于英伟达的计算基础设施进行开源大型语言模型开发。2024年5月，DeepSeek发布了DeepSeek-V2，训练成本仅为600万美元——而OpenAI和谷歌的类似模型成本超过1亿美元——实现了约95%的成本降低，同时提供了有竞争力的性能。这通过多头潜在注意力（MLA）和DeepSeekMoE等创新引发了中国AI开发者的价格战。2025年1月，DeepSeek发布了DeepSeek-R1（及R1-Lite-Preview），这是一个推理模型，在多个基准测试中达到或超过OpenAI的o1性能，同时使用的先进芯片少得多。该发布立即引起市场动荡：英伟达单日市值蒸发6000亿美元，中国科技股飙升，DeepSeek的应用程序短暂超越ChatGPT成为苹果美国商店排名第一的免费应用。梁文锋在同一天受邀参加了与中国总理李强的座谈会。2025年，DeepSeek继续发布模型（包括两个月内以600万美元成本训练的V3）和负担得起的API服务，为小型开发者和企业提供了民主化的访问。[编辑]

梁文锋被认为是一个隐居、极度注重隐私的人物，回避公众视线，这与他的工作所带来的颠覆性影响形成了鲜明对比。媒体描绘强调他的‘极客’形象——一个拥有卓越技术深度的人，尽管领导着一家大公司，却仍然亲力亲为从事研究。内部人士称他拥有‘可怕的学习能力’，被视为一位将架构创新与成本效率相结合的战略家。他很少公开露面；在金融和人工智能领域，他是一个雄辩的演讲者，但没有社交媒体存在。粉丝和行业观察者欣赏他低调的风格和对开源贡献的专注，尽管没有形成正式的粉丝名称或文化。他的工作基调是务实和反体制的，挑战硅谷的常规。[编辑]

Liang Wenfeng commands a unique position in global technology, bridging the gap between reclusive engineering and massive market disruption. While he shuns traditional celebrity culture, his impact has propelled him into the highest echelons of public recognition. According to the iFANN Global Fame Rankings, Liang is currently the 54th most famous person from China, a standing that reflects his nation's rising influence in artificial intelligence. His specific dominance in the tech sector is even more pronounced; he ranks as the 14th most famous person in AI, surpassing many Western counterparts who rely on larger marketing budgets. These placements are not merely statistical artifacts but reflect a genuine shift in how the world perceives Chinese innovation. The rankings track global attention across media coverage, search trends, and consumption signals worldwide, capturing the moment DeepSeek challenged U.S. hegemony with radical efficiency. Despite this fame, there is no formal fanbase or organized cult following dedicated to Liang personally. Unlike pop stars, his audience consists of developers, investors, and industry analysts rather than adoring crowds. There are no fan rituals, birthday billboards, or charity drives in his name. Instead, the public interaction is defined by intense professional scrutiny and the viral spread of his management philosophies. The lack of a named fandom underscores his identity as a serious industrialist whose legacy is built on code and capital rather than cultural performance.[编辑]

Liang Wenfeng began experimenting with quantitative trading at age 23 in 2008, using scraped market data and machine learning algorithms alongside classmates. High-Flyer faced regulatory pressures in China from 2019 to 2023, which prompted Liang to pivot the firm's computing infrastructure toward AI research. DeepSeek trained its V2 model for just $6 million, a fraction of the over $100 million spent by OpenAI and Google on similar systems. Insiders describe Liang as having a 'terrifying ability to learn,' remaining technically stronger than many first-line researchers while excelling at high-level strategy. He grew up in an environment where parents viewed studying as 'useless' compared to quick business opportunities, only for views to reverse by his generation. In 2025, he was named to TIME's 100 Most Influential People in AI, listed in Fortune's Most Powerful People in Business, and included in the Bloomberg Billionaires Index. DeepSeek-V2 was recognized for the most cost-efficient training of a frontier-level AI model in 2024. A widely retold story from July 2026 involves Liang enforcing a strict policy against overtime work and KPIs, defying the pervasive '996' culture common in Chinese tech firms. Another key moment in his lore is his aggressive stance that 'OpenAI is Not God,' prioritizing AGI goals over immediate profits. As of August 2026, media outlets described a 'surprise offensive' where DeepSeek released the V4 Pro model, allegedly outperforming Grok 4.6 in head-to-head tests against Elon Musk's systems.[编辑]

梁文锋的父母是广东省一个小村庄的小学教师。他没有公开透露任何关于兄弟姐妹、配偶或子女的信息。他保持着高度私密的个人生活，除了成长背景外，没有经过证实的家庭细节。没有公开披露过兄弟姐妹、配偶或子女。可靠消息来源称没有恋爱关系、订婚、结婚、离婚、风流韵事或公开传闻。他从未在媒体报道中与任何伴侣有过关联。[编辑]

梁文锋与同为浙江大学校友的徐进共同创办了杭州雅克比投资管理有限公司。他与两位未具名的工程系同学共同创办了幻方。他的公司DeepSeek和幻方雇佣了研究者和工程师团队，他们合作开发了DeepSeek-V2和V3等模型。[编辑]

梁文锋一直保持着高度私密的个人生活，在任何可信来源中都没有公开的恋爱关系、订婚、结婚、离婚或传闻伴侣的记录。他在媒体报道中从未与任何人有过关联。截至2026年，据报道梁文锋单身，没有确认的关系或子女。没有公开披露过兄弟姐妹、配偶或子女。可靠消息来源称没有恋爱关系、订婚、结婚、离婚、风流韵事或公开传闻。他从未在媒体报道中与任何伴侣有过关联。[编辑]

Liang Wenfeng's net worth remains a moving target for the global financial press, yet iFANN's editorial analysis places his current valuation at exactly $36 billion. This figure anchors itself in the June 2026 funding round where DeepSeek secured $7.4 billion at a $50 billion valuation, a moment that saw Liang personally invest $3 billion of his own capital. Bloomberg now identifies him as the world's richest creator of artificial intelligence models, a title earned through a trajectory that began with modest estimates hovering between $1 billion and $3.2 billion before 2025. By late 2025, the launch of the reasoning model R1 had already propelled cited figures to approximately $11 billion. The surge to $36 billion reflects a calculated stake of roughly 78% following dilution, effectively doubling the previous estimate of $16.7 billion. These numbers fluctuate wildly because DeepSeek operated without outside funding for most of its existence, forcing analysts to guess values ranging from $1 billion to $150 billion based on internal metrics. With Liang retaining ownership between 76% and 84% depending on the timing of the assessment, even minor shifts in assumed company value create massive swings in his paper wealth.[编辑]