--- title: "Kai-Fu Lee: China's open-source AI models just hit their global moment" date: 2026-09-22 category: Builders site: NeuroAI canonical: https://neuroai.site/a/na-people-kaifulee-global-ai language: en --- # Kai-Fu Lee: China's open-source AI models just hit their global moment > At the 26th China International Fair for Investment & Trade, 01.AI founder Kai-Fu Lee argued Chinese open-source models have their best window to go global—if firms stop "wasting" AI on trivial tasks. At a trade fair in Xiamen this month, one of China's best-known AI investors sounded less like a salesman and more like a coach with a stopwatch. Kai-Fu Lee—founder and CEO of **01.AI (零一万物)** and chairman of Innovation Works—told the room the uncomfortable part out loud: most companies are using AI wrong. Lee spoke on 10 September 2026 to Securities Times (证券时报) on the sidelines of the **26th China International Fair for Investment & Trade (CIFIT)**, held 8–11 September. His thrust was not "AI is coming" but "the window is open right now, and you are late." ## The "potted plant" problem Lee's sharpest line is about waste. Many firms, he said, "are using AI for very simple, basic tasks"—customer service, bookkeeping, contract review. He compared that to arranging a few **"potted plants (盆景)"** that will never grow into a **"forest (森林)."** "On perhaps 98% of the problems in the world, AI is already smarter than me," Lee said. So deploying it as a chatbot or a layoff tool is, in his words, "the biggest under-use in human history." The real job, he argued, is to hand AI the hard problems: new strategy, new products, finding existing operational leaks and risks. ## The money math he keeps repeating Lee is unsentimental about ROI. A company that uses AI well, he said, does not cut costs by 5–10% but by **50–80%**—especially digital-native firms, which also open new product lines. He used coding as the cleanest example. A piece of enterprise software a customer might pay **50,000 yuan (about US$7,000 / HK$55,000)** for used to cost **1,000,000 yuan (about US$141,000 / HK$1.1 million)** to build the traditional way. With AI writing the code, he said, the same product can be produced for roughly **500 yuan (about US$70 / HK$550)**—and still sold for 50,000. That spread, he argued, is the real profit pool. His test for any AI project is blunt: "Which number on your financial report did AI improve? How much did it move your stock?" If you cannot answer, he said, you have not done it right. ## Why open source is his ace The global claim is the headline. "The whole world is now embracing our open-source models," Lee said; open source is "a particularly successful strategy." His logic is practical, not ideological: - Chinese open-source models are **fully downloadable and privately deployable**—no data leaves the buyer's border. - They cost roughly **one-third to one-tenth** of U.S. closed models. - U.S. closed models lead by only about **6 months**, in his estimate—a gap he called small. Outside the U.S.–China contest, Lee argued, almost every country "is far behind" and suffers a huge information gap—many have barely heard of agents (智能体). That gap is the opening. He cited 01.AI helping some countries build national AI-education systems and apply AI to energy and cultural tourism. ## Why his voice carries weight Lee is not an obscure commentator. A 40-year AI veteran, former Google China head, and founder of both Innovation Works and the model-maker 01.AI, he straddles Silicon Valley credibility and China's domestic **large model (大模型)** race. When he tells Chinese firms the export window is open, he is also signaling where 01.AI is placing its own bets—abroad. ## The valuation warning he refuses to skip Lee is not a cheerleader without caveats. He admitted current AI investment is "somewhat overheated" and invoked the image of a party: "every carnival has its last song." When growth slows from 10x to 3x to 1.2x, he warned, valuations can fall as fast as they rose. His close: every business must rest on a sound model—how it earns, subscribes, and profits—or "when the last song ends, there may be no seat." ## Honest limitations This piece is reconstructed from a single primary source: Lee's 10 September 2026 interview with Securities Times at CIFIT, as published by the newspaper. It is a media interview, not a signed speech transcript, so wording is the reporter's rendering of his remarks. - All quotes are attributed to that one interview; we found no simultaneous English-primary transcript (e.g., Bloomberg) to cross-check phrasing, so the Chinese original should be treated as authoritative. - The 50–80% cost-saving and 1/3–1/10 price claims are Lee's stated estimates, not independently audited figures; they describe his view, not market consensus. - The "98% of problems" and "6 months behind" lines are rhetorical framings from the interview, not benchmarks we can verify. - Currency conversions use approximate 7.1 RMB/USD and 0.91 RMB/HKD rates and are rounded; the underlying yuan amounts are Lee's own examples. - We did not verify 01.AI's specific overseas contracts; the "national AI-education systems" reference is Lee's own claim in the interview. ## What readers can do now - **If you run a company,** run Lee's test this quarter: pick one hard problem—new market, hidden cost leak—and put AI on it, then report the before/after number on the P&L. - **If you build or buy AI,** price a privately deployable open-source model against a closed API over 12 months; for data-sovereignty markets, the 1/3–1/10 gap is the whole argument. - **If you invest in AI,** separate revenue trajectory from narrative—re-read Lee's "last song" line before any multiple that assumes 10x growth forever. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-people-kaifulee-global-ai Free to quote with attribution and a link to the original.