档案库 · AI 与模型 · 战略决策 · 2023–2026
01.AI押注中国，转向B2B，计划港交所IPO
李开复2023年创办01.AI；与阿里合并后转向To B；2026年订单超15亿元，计划港股IPO。
01.AI (零一万物)
做的是什么生意
AI large-model startup, early on building the Yi series foundation models; from 2025, pivoting to enterprise AI transformation services and a sovereign AI platform.
起因
Founded in March 2023, it entered at the same time as Baichuan Inc. (百川智能), Moonshot AI (月之暗面), and StepFun (阶跃星辰), collectively called the 'AI Six Little Tigers' (AI六小虎). Kai-Fu Lee bet that in the first year of generative AI, China could produce a world-class foundation model, claiming to build a full stack of 'model + AI Infra + application', with the goal of becoming the next OpenAI.
经过
In December 2024, DeepSeek released the V3 technical report; in early 2025, R1 became widely known, and a model price war fully erupted. Kai-Fu Lee judged that foundation models were a resource war and that startups could not outspend large companies, so in early 2025 he merged most of the pretraining and AI Infra teams with Alibaba (阿里), abandoning the ultra-large-model competition; according to Jiemian News (界面新闻) citing Lee's own account, during that period one-third of the company's workstations were empty and morale hit rock bottom. Afterwards, 01.AI pivoted to To B and sovereign AI (主权AI), and Lee met more than 100 CEOs one-on-one over a year, pitching the product as a 'top-executive project' (一号位工程) that could improve enterprise financial statements.
结果
In 2025, orders were about RMB500 million and audited revenue was RMB250 million; as of May 2026, contracted orders exceeded RMB1.5 billion, and the company aims to reach RMB1.5–2.0 billion in contracts within the year and pursue break-even. In July 2026, it launched Wan Ce AI (万策AI) and the Yihaowei AI (一号位AI) product matrix, is dismantling its offshore shareholding structure, advancing Pre-IPO financing, and plans to list in Hong Kong in 2027.
背景
李开复2023年3月创办01.AI，赌中国能媲美OpenAI。它以全栈叙事加入“AI六小虎”，但融资最少，靠李开复个人品牌撑场。
转折点：DeepSeek低成本开源模型和价格战。李开复把预训练团队并入阿里，转向To B和主权AI，对标Palantir。早期转型艰难：三分之一工位空置，士气低迷。
李开复当起头号销售，见100多位CEO。2025年订单约5亿元；2026年合同订单超15亿元；Pre-IPO融资，计划2027年港股上市。
截至2026年9月，01.AI推新品、拆海外架构。批评者指出订单5亿元只带来2.5亿元收入，过度依赖李开复，订单转ARR的转化率未经验证。
这件事要成立，得有什么
- 李开复判断基础模型是资源战，行业最终会集中到少数巨头，初创烧钱必输。
- To C流量被大厂垄断，传统To B是几十万元代理的竞标生意，两者都不值得all-in。
- 他赌生成式AI会让企业高管为“改变财报”付费，而不是为单个工具付费。
- 主权AI对应30-40个国家的AI自主需求，他相信中国几乎没人愿意或能做。
- 他本人的国际顾问角色和人脉，是能打开国家和企业高层大门的资产。
可借鉴之处
当赛道变成打不赢的烧钱战，早让步并把现有资产重新包装成赚钱模式，可能保证生存。
后续进展
截至2026年9月，01.AI转型继续：2025年订单约5亿元，收入2.5亿元；到2026年5月合同订单超15亿元，目标15-20亿元并盈亏平衡。准备新一轮融资，拆除海外结构，计划2027年港股上市；推出万策AI和一号位AI。未经验证：订单转收入率低、ARR占比、交付可复制性、以及客户获取是否依赖李开复个人品牌。
资料来源
发现哪里写错了？告诉我们。
轮到你了
你刚读完一家。说说你在做什么，看看谁在赌同一件事。
免费账号 · 3 次免费提问 · 不用绑卡