Kai-Fu Lee
Last reviewed
Apr 30, 2026
Sources
30 citations
Review status
Source-backed
Revision
v1 · 3,987 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
Apr 30, 2026
Sources
30 citations
Review status
Source-backed
Revision
v1 · 3,987 words
Add missing citations, update stale details, or suggest a clearer explanation.
Kai-Fu Lee (Chinese: 李開復; born December 3, 1961) is a Taiwanese-American computer scientist, venture capitalist, and author who has spent four decades at the center of artificial intelligence research and industry. He is best known today as the founder and CEO of 01_ai, the Beijing-based large language model startup behind the open-source Yi model series, and as the founder and chairman of Sinovation Ventures, one of China's most influential early-stage technology investors. His earlier career took him from Carnegie Mellon, where his 1988 doctoral thesis on the Sphinx system produced the first speaker-independent, large-vocabulary, continuous speech recognition engine, through senior R&D roles at Apple Computer, Silicon Graphics, Microsoft, and Google, where he served as president of Google's China operations from 2005 to 2009.
Lee occupies an unusual position in the global AI ecosystem. He grew up between Taipei, Tennessee, and Pittsburgh; built research labs in Beijing for an American software giant; ran a Chinese division for a Mountain View internet company; wrote one of the most cited popular books on US-China AI competition; and now runs a Chinese frontier-model lab that competes directly with OpenAI and Google on the LMSYS Chatbot Arena leaderboard. By his own count he has more than fifty million followers across Chinese social media, which makes him one of the most widely followed technology figures in the Chinese-speaking world.
| Item | Detail |
|---|---|
| Born | December 3, 1961, Taipei, Taiwan |
| Father | Li Tianmin, Republic of China legislator and historian |
| Education (BS) | Columbia University, Computer Science, 1983, summa cum laude |
| Education (PhD) | Carnegie Mellon University, Computer Science, 1988 |
| Doctoral advisor | Raj Reddy ([turing_award](Turing Award) laureate, 1994) |
| Citizenship | Taiwanese (renounced US citizenship in 2011) |
| Major employers | CMU faculty, Apple, SGI, Microsoft, Google |
| Companies founded | Innovation Works / Sinovation Ventures (2009), 01.AI (2023) |
| Major books | AI Superpowers (2018), AI 2041 (2021) |
Kai-Fu Lee was born in Taipei in late 1961, the youngest of seven children. His father, Li Tianmin, was a legislator in the Republic of China government on Taiwan and later a historian of modern China; his mother was a homemaker who pushed her children hard on schoolwork. The family was originally from Sichuan province on the Chinese mainland and had moved to Taiwan in 1949 with the Nationalist government.
In 1973, when Lee was eleven, his mother sent him to live with an older brother in Oak Ridge, Tennessee, so that he could attend American schools. He stayed in Tennessee through middle school and high school, and graduated from St. Mary's High School in Memphis.
He entered Columbia University in 1979, originally on a pre-law track, and switched to computer science after discovering programming. He graduated in 1983 summa cum laude. He went directly to Carnegie Mellon University in Pittsburgh for graduate school, where the speech and language group around Raj Reddy and Alex Waibel was one of the strongest in the world.
Lee's doctoral dissertation, completed in 1988 under Raj Reddy, is the work that established his reputation in computer science. The thesis, titled Large-Vocabulary Speaker-Independent Continuous Speech Recognition: The Sphinx System, described what was then a startling result: a single recognizer that did not need to be retrained on each new user's voice, that could process continuous (not isolated word) speech, and that handled vocabularies of around a thousand words drawn from the DARPA Resource Management task. Earlier systems had typically required users to train them by reading sample sentences, or to pause between every word, or to work only with very small vocabularies. Sphinx broke all three of those constraints at once.
The system used Hidden Markov Models for acoustic modeling, with discrete vector-quantized features and triphone-based context-dependent units. Lee combined ideas from his CMU peers with new techniques for sharing parameters across triphones (so that rare contexts could borrow statistics from common ones) and for jointly modeling acoustic and language probabilities. The published version of the thesis appeared in 1989 as a Kluwer monograph, Automatic Speech Recognition: The Development of the Sphinx System, which became a standard reference in the field for years afterward.
Sphinx caused a stir well beyond academia. Business Week selected it as the "Most Important Innovation" of 1988. The CMU lineage of Sphinx continued for decades through Sphinx-2, Sphinx-3, and Sphinx-4, releases that were used widely in research and in commercial products.
While at CMU, Lee also collaborated with Sanjoy Mahajan on a Bayesian-learning-based program called Bill that played Othello. In 1989, Bill won the United States national computer Othello tournament, and an earlier version had defeated a then-world-champion human player. Lee stayed at CMU as an assistant professor for two years after his PhD before deciding that industry, not academia, was where speech and natural-language interfaces would actually reach users.
Lee joined Apple Computer in 1990 as a principal research scientist and rose to become director of advanced technology, then a vice president of the multimedia laboratory. His brief was to take research-quality speech and human-interface technology and put it into Macintosh products. He led teams that built PlainTalk, Apple's speech recognition and text-to-speech infrastructure for the Macintosh, and Galatea, a high-quality concatenative text-to-speech system. He also worked on QuickTime VR, an early panorama-and-object-movie technology that let users navigate 360-degree photographic scenes from their desktops, a precursor in spirit to today's spatial computing on devices like the [apple_vision_pro](Apple Vision Pro).
The most public of his Apple projects was Casper, an experimental speaker-independent voice assistant for the Mac. On March 2, 1992, Lee appeared on ABC's Good Morning America with Apple CEO John Sculley and demonstrated Casper live, asking it to schedule meetings and read mail aloud. The demo predated siri, which Apple later acquired in 2010, by nearly two decades, and predated alexa and modern [apple_intelligence](Apple Intelligence) by even longer. Casper itself never shipped as a consumer product, but several of its components found their way into PlainTalk and into Apple's later voice work.
In 1996, Lee left Apple to join Silicon Graphics (SGI) as vice president of its web products division, then took over as president of Cosmo Software, the SGI multimedia and 3D-internet subsidiary. The era was the height of the VRML boom, when SGI was betting that 3D worlds would become a standard layer of the early consumer web. The bet did not pan out commercially, and Lee spent less than two years at the company before being recruited by Microsoft.
In July 1998, Bill Gates and Microsoft research head Rick Rashid asked Lee to set up a new Microsoft research lab in mainland China. He moved to Beijing in late 1998 and opened Microsoft Research China, which was renamed Microsoft Research Asia (MSRA) in 2001 to reflect a broader regional mandate. Lee served as the founding managing director from 1998 to 2000.
MSRA grew quickly into one of the most respected computer science research laboratories in Asia. Lee hired heavily from top Chinese universities (Tsinghua, Peking University, USTC, Shanghai Jiao Tong) and from the Chinese diaspora returning from US PhDs, including Harry Shum, who took over as managing director in 2004 and later led Microsoft's global AI and Research group, and Hsiao-Wuen Hon, who served as MSRA chairman until his retirement in 2022. MIT Technology Review in a 2004 cover story called MSRA "the world's hottest computer lab." For broader context, see [microsoft_research](Microsoft Research).
The lab is sometimes nicknamed the "Whampoa Military Academy of AI" in Chinese press, after the 1924 military academy that produced many leaders of both the Nationalist and Communist armies. The MSRA alumni list includes founders and CTOs at Baidu, Alibaba, Tencent, ByteDance, Lenovo, Huawei, Megvii, SenseTime, and 01.AI itself.
In 2000, Lee returned to Redmond as a Microsoft corporate vice president, leading the Natural Interactive Services Division. He oversaw speech recognition, search, and natural-language interfaces, and was responsible for parts of what became Windows Vista's voice features. He stayed in this role until July 2005.
In July 2005, Lee accepted an offer from Google to become founding president of Google China and a vice president of the parent company. The day Google announced his hiring, Microsoft sued both Lee and Google in a Washington state court, arguing that the move violated his non-compete agreement.
The case, Microsoft Corp. v. Lee and Google Inc., was followed closely as a bellwether for non-compete enforcement in the technology industry. On July 28, 2005, Superior Court Judge Steven Gonzalez issued a temporary restraining order limiting what Lee could work on while at Google. On September 13, the judge ruled that Lee could move to Google but barred him from working on search, speech recognition, and budgeting and recruiting decisions for Google China until a January 2006 trial. On December 22, 2005, Microsoft and Google announced a confidential settlement that ended the case before trial; Lee began full duties at Google soon after.
As president of Google Greater China, Lee built the company's mainland presence from a small team into an organization of around seven hundred employees, launched the Chinese-language Google.cn search service in January 2006, and managed the company's complex relationship with Chinese government regulators. Google.cn complied with mainland censorship requirements, a decision that drew sharp criticism in the West and that Google itself reversed in 2010 after the "Operation Aurora" cyberattacks, when it stopped filtering Chinese-language results and re-routed users to its Hong Kong service.
Lee left Google in September 2009, before the company's eventual departure from the mainland search market. His last day at Google was September 4, 2009.
On September 7, 2009, three days after leaving Google, Lee announced Innovation Works, a $115 million early-stage venture fund and incubator based in Beijing. The original investors included Foxconn's Terry Gou, YouTube co-founder Steve Chen, and US-based VC firms. The thesis was that China's mobile internet boom needed an American-style early-stage VC firm willing to also act as a startup studio, providing legal, recruiting, and product help to founders.
In 2016, Innovation Works rebranded to Sinovation Ventures (创新工场 in Chinese), reflecting both a wider geographic mandate and a shift toward a more conventional venture model. The firm now manages a series of dual-currency funds (US dollar and renminbi) totaling roughly three billion US dollars, with offices in Beijing, Shanghai, Shenzhen, and (until 2019) Silicon Valley. The Silicon Valley office closed in 2019, citing the difficulty of investing across the US-China trade war.
Sinovation has backed more than four hundred companies. Notable portfolio names include:
| Company | Sector | Notes |
|---|---|---|
| Meitu | Photo apps | Hong Kong listing 2016 |
| VIPKid | Online English tutoring | One-time decacorn |
| Megvii (Face++) | Computer vision | Pioneer in face recognition |
| Bitmain | Bitcoin mining and AI chips | One of the largest crypto-mining ASIC vendors |
| Mobike | Bicycle sharing | Acquired by Meituan in 2018 |
| TuSimple | Autonomous trucking | Listed on Nasdaq 2021 |
| Knowbox | K12 ed-tech | Math homework app |
| Zhihu | Q&A community | NYSE IPO 2021 |
| WeRide | Autonomous driving | Listed on Nasdaq 2024 |
| 4Paradigm | Enterprise AI platforms | HKEX IPO 2023 |
| 01.AI | Foundation models | Lee's own startup, Sinovation is a backer |
In 2017 Sinovation also launched the Sinovation Ventures AI Institute, an internal applied-research team that productized image recognition, speech, and recommender models for portfolio companies. Lee runs both the fund and the AI Institute, with co-founder Wang Hua handling day-to-day fund operations.
In March 2023, weeks after the global ChatGPT shock, Lee founded 01.AI (零一万物, Língyī Wànwù; "zero-one, ten thousand things") as a dedicated foundation-model lab. He has often described 01.AI as his fourth company-building act after MSRA, Google China, and Sinovation, and the only one where he is personally on the founder's track again rather than the investor's.
01.AI is headquartered in Beijing and incubated inside Sinovation. It crossed unicorn valuation (one billion US dollars) by November 2023, eight months after launch. Backers reportedly include Alibaba and other Chinese strategics. The company's research and product strategy has centered on the open-weights Yi model family, which is named after the Chinese character 一 ("one") and is the inspiration for the company's name. Many of 01.AI's checkpoints are released on Hugging Face under the 01-ai organization.
| Model | Release | Notes |
|---|---|---|
| Yi-6B and Yi-34B | November 2023 | First Yi releases; English and Chinese; topped early Hugging Face open LLM leaderboard for base models |
| Yi-VL-6B and Yi-VL-34B | January 2024 | First open-source 34B vision-language model worldwide |
| Yi-9B | March 2024 | 48-layer up-scaled model continued from Yi-6B on 0.8 trillion tokens |
| Yi-1.5 (6B, 9B, 34B) | May 13, 2024 | Stronger coding, math, and reasoning |
| Yi-Coder (1.5B, 9B) | September 2024 | 52 programming languages; 128K context |
| Yi-Lightning | October 2024 | API-only flagship; topped LMSYS Chatbot Arena rankings for non-US labs |
Yi-34B's November 2023 release drew immediate attention. Tested on standard benchmarks, it outperformed Meta's Llama 2 70B at less than half the parameter count, and Hugging Face listed it as the top open-source pre-trained base LLM at the time. The release was unusual in that 01.AI made the weights, not just the API, freely available for both research and (with a separate license grant) commercial use. The Yi technical report, Yi: Open Foundation Models by 01.AI, was posted to arXiv in March 2024.
Yi-Lightning, released in October 2024, was the model that put 01.AI in direct conversation with frontier US labs. On the LMSYS Chatbot Arena (since renamed lmarena.ai), Yi-Lightning's overall Elo placed it at sixth in the world on its first appearance, and joint-third by company alongside xAI's Grok-2, behind only OpenAI's o1-preview and GPT-4o variants and Google's Gemini-Exp. Lee publicly framed the result as evidence that the gap between Chinese and US frontier models had shrunk to about five months. Yi-Lightning was also notable for cost: 01.AI priced API inference at roughly fourteen cents per million tokens, well under what competing US flagship models charged at the time. Tom's Hardware reported that the Yi training stack was built on roughly two thousand GPUs at a reported total cost of around three million US dollars, several times less than commonly cited US training budgets for comparable models.
In late 2024 and early 2025, Chinese press reported that 01.AI was reorganizing around enterprise applications and partnering more closely with Alibaba's cloud infrastructure. Lee denied selling the pre-training team outright, but the company did pull back on competing at the absolute frontier of base-model training in favor of applications built on top of Yi and other strong open-source bases such as DeepSeek's models.
Lee has written or co-written more than ten books, in both Chinese and English. He uses publishing as a tool for shaping public discussion of AI in a way that is unusual for a working CEO. The English-language books are listed below; he has also published several Chinese-language autobiographies and essay collections, including A Walk Into The Future (2006), Be Your Personal Best (2005), the autobiography Making A World of Difference (2009), and Seeing Life Through Death (2015), which is a meditation on his cancer experience.
| Title | Year | Publisher | Notes |
|---|---|---|---|
| Automatic Speech Recognition: The Development of the Sphinx System | 1989 | Kluwer Academic | Monograph based on his 1988 PhD |
| Readings in Speech Recognition (edited with Alex Waibel) | 1990 | Morgan Kaufmann | Standard reference reader |
| Making A World of Difference (世界因你不同) | 2009 | Citic Press (Chinese) | Autobiography |
| AI Superpowers: China, Silicon Valley, and the New World Order | September 2018 | Houghton Mifflin Harcourt | International bestseller; 272 pages; ISBN 9781328546395 |
| AI 2041: Ten Visions for Our Future (with Chen Qiufan) | September 2021 | Currency / Penguin Random House | Ten near-future short stories with technical commentary |
AI Superpowers is the book that made Lee a household name in English-language policy circles. Its core argument is that deep learning rewards data, compute, and engineering scale more than scientific novelty, and that on those dimensions China has structural advantages: a larger pool of engineers, a more aggressive startup culture, lower friction around personal data, and a state that treats AI as a strategic priority. He frames the future as a US-China duopoly rather than a single-power race, and devotes the second half of the book to labor displacement and to a personal argument, shaped by his cancer diagnosis, about the kinds of work that will remain meaningful for humans. The book was a New York Times, USA Today, and Wall Street Journal bestseller. Foreign Affairs praised its on-the-ground reporting of Chinese tech but criticized it for understating Chinese state intervention.
AI 2041, written with Chinese science-fiction author Chen Qiufan, takes a different format: ten short stories set in 2041, each followed by a non-fiction essay from Lee that explains the AI technology underlying the story. Settings range from a Mumbai teen rebelling against a paternalistic insurance algorithm to a Tokyo idol-worshipping fan in a mixed-reality world to a San Francisco worker whose job has been re-allocated by an automation insurer. The book was named a best book of the year by the Wall Street Journal, Washington Post, and Financial Times.
Lee is also a regular speaker on the conference circuit. His TED talk "How AI can save our humanity," delivered in April 2018 at TED2018 in Vancouver, has been viewed several million times. He has appeared on 60 Minutes, BBC HARDtalk, and the World Economic Forum, and he co-chairs the Forum's AI Council.
In September 2013, Lee announced on Weibo that he had been diagnosed with stage IV follicular lymphoma at the age of fifty-one. He moved back to Taipei for treatment, withdrew from most public-facing work for more than a year, and chronicled the experience in widely shared Weibo posts and in the 2015 book Seeing Life Through Death. By June 2015, he reported that his tumors had disappeared.
Lee has said that the diagnosis was a turning point. He had been a famous workaholic, and after recovery he reduced his work hours and prioritized time with his wife and two daughters. The closing chapters of AI Superpowers tie this argument to AI policy: a world in which machines do more economic work should also be a world in which humans spend more time on care, family, and relationships.
Lee's wife, Shen-Ling Hsieh, is also from Taiwan; the couple has two daughters.
By 2024, Lee was being framed in the international press as one of the central figures of the Chinese frontier-model wave. Yi-Lightning's strong showing on Chatbot Arena, alongside releases from DeepSeek, Moonshot AI, MiniMax, Zhipu AI, Baichuan, and Alibaba's [tongyi_qianwen](Tongyi Qianwen) family, made the case that the Chinese AI field had a depth of independent labs comparable to the US. Lee has argued publicly that US export controls have not stopped Chinese model progress, and that algorithmic efficiency is catching up with raw compute advantages.
He has also continued to use Sinovation as a vehicle for applied AI investments, including humanoid-robotics, AI chips, and vertical SaaS. His public commentary in 2024 to 2026 has emphasized practical AI deployment over scaling-only narratives, a position that aligns with 01.AI's pivot toward enterprise applications.
Lee's recognized honors include:
| Honor | Year | Body |
|---|---|---|
| Business Week "Most Important Innovation" | 1988 | for the Sphinx system |
| IEEE Fellow | 2002 | Institute of Electrical and Electronics Engineers |
| Time 100 Most Influential People | 2013 | Time magazine |
| WSJ All Things Digital Asia D Conference speaker | multiple years | News Corp / Dow Jones |
| Asia House Asian Business Leader | 2018 | Asia House |
| Wired 25 Icons | 2018 | Wired |
| Honorary Doctorate | various | Carnegie Mellon University, City University of Hong Kong |
| Time 100 AI | 2023 | Time magazine |
He is a member of the Committee of 100, a non-partisan group of prominent Chinese-American leaders, and has co-chaired the World Economic Forum's AI Council. The published references to membership in the US National Academy of Engineering are inconsistent and not confirmed in the academy's online directory, so this article does not list it among his honors.
Two threads run through Lee's career. He has consistently bet on the human-machine interface (Sphinx, Casper, Google China search, the Yi chat models) rather than on infrastructure or pure research. And he has consistently moved between cultures: Taipei, Tennessee, New York, Pittsburgh, Cupertino, Beijing, Mountain View, and back to Beijing.
The most lasting institutional legacy is probably Microsoft Research Asia, which trained an entire generation of senior researchers and CTOs at Chinese tech giants and at 01.AI itself. The most consequential bet of the last few years is 01.AI: if Yi continues to track close to US frontier systems on Chatbot Arena and on real-world enterprise tasks, Lee will end up having helped seed both the Chinese applied-AI ecosystem (through Sinovation) and one of its leading model labs. If Yi falls behind, Sinovation's broader portfolio and his books will likely remain his more important contributions.