Kai-Fu Lee, the chief executive of 01.AI and chairman of Sinovation Ventures, used a Fortune Daily interview to unpack three misconceptions about China’s AI. In his telling, it is wrong to think that Chinese open models cannot be monetized, that China competes as a unified entity, and that export controls will constrain the development of Chinese AI. He shared the interview on X on 11 November.
The two Chinese model makers listed in Hong Kong show fast revenuegrowth but not yet profit. Zhipu’s revenue for the six months to 30 June 2026 rose 399.7%, about five times, and MiniMax’s rose 283.1%, their interimresults show. Both reported losses, and by our arithmetic their 30 June balances covered roughly 12 and 27 months of first-half losses.
Business Pill · Open Weights Explained
An evergreen Business Engineer explainer on open weights: rent a model through an API or run it yourself. It teaches the general idea only and does not discuss Lee’s interview or these companies.
The key insight: As we read it, Lee’s first claim and the filings agree on revenue, not yet on profit: Zhipu’s first-half revenue rose about five times and MiniMax’s 3.8 times, while their 30 June balances covered roughly 12 and 27 months of first-half losses, short of the three or four years Lee describes for Chinese AI startups.
On the first point, Lee told Fortune that Chinese companies are “born to make money”, pointing to Pinduoduo, Meituan and ByteDance as companies once questioned on how they would earn it. His advice: “never doubt Chinese ability to find a way to make money”.
On the second, he said: “The Chinese companies are venture-funded, just like the American companies”. He called the idea that the government tells model companies to open-source “made-up nonsense” and traced the open-weight wave to the DeepSeek moment, which in his account left rival labs with one question: “can you be Deep Seek? If not, get out of the business”.
On the third, he said the belief that “export control is what limits the Chinese spending on GPUs. It is not.” In his account, “The Chinese companies are naturally parsimonious”, and a startup raises money when it can and spends it carefully: “Now they have money to last three or four years and we’ll see what happens”.
Lee also framed the market as two models. American labs keep their models closed and, he said, “price their tokens extremely high with very high margins”, while Chinese labs build nearly as good models “for 1/10 or even less the price for both training and inference”. In his words: “So now we’re in an iPhone and Android situation”. The quotes here come from YouTube’s automatic captions of the video, which Fortune says is a small part of a longer conversation.
Months of first-half 2026 loss covered by each lab’s 30 June 2026 balance, from the Zhipu (Z.ai) and MiniMax interim results: MiniMax’s $1,322.8 million cash balance against its $293.0 million adjusted net loss (27 months) and $358.0 million reported loss (22 months); Zhipu’s RMB3,993.7 million of cash and cash equivalents against its RMB1,964.1 million adjusted net loss and RMB2,072.0 million reported loss (12 months each). The dashed line marks 36 months, the low end of Kai-Fu Lee’s ‘three or four years’ for Chinese AI startups. The months are our arithmetic; a loss is not cash burn.
What Was Posted on X
We began this Fortune interview by unpacking 3 misconceptions about China’s AI development: that Chinese open models cannot be monetized, that China competes as a unified entity, and that export controls will constrain the development of Chinese AI. We then broadened the conversation to explore the key ideas and insights behind my new book, AI Native.
Kai-Fu Lee, whose X profile lists him as CEO of 01.AI and chairman of Sinovation Ventures, posted at 16:37 UTC on 11 November that his Fortune interview began by unpacking three misconceptions about China’s AI development: that Chinese open models cannot be monetized, that China competes as a unified entity, and that export controls will constrain Chinese AI. He links the video, which also covers his new book, AI Native. This post is the starting point for the piece, which tests his three claims against the Zhipu and MiniMax filings.
Claim One: The Revenue Is Growing
Zhipu, which trades in Hong Kong as Z.AI, reported revenue of RMB953.9 million for the half, against RMB190.9 million a year earlier, and puts it at approximately USD142 million. Its open platform and API business brought in about RMB825 million, up 2,735.7%, and made up 86.5% of revenue.
MiniMax reported revenue of $116.6 million for the half, up 283.1%, “exceeding our total revenue of US$79.0 million for the full year of 2025”. By our arithmetic, its first half alone was 148% of all of 2025, and Zhipu’s revenue was about 5.0 times its year-earlier half.
Profit is a different matter. Zhipu’s loss for the period was RMB2,072.0 million and MiniMax’s was $358.0 million. Gross margins were 26.4% at Zhipu, down from 50.0% on a change in revenue mix, and 17.9% at MiniMax, up from 12.1% on what it calls improving infrastructure efficiency. We compared the two labs with Anthropic in Anthropic Made 63x the Revenue of China’s Zhipu and MiniMax.
Lee’s own words allow for that gap: “Chinese companies don’t make nearly as much money, but they’re going to get the footprint and the market share”. Epoch AI’s revenue estimates for six Chinese AI firms, including DeepSeek, are in our piece on the $11 billion estimate.
Reported gross profit margins from the Zhipu (Z.ai) and MiniMax interim results for the six months to 30 June 2026 and the same period of 2025. Zhipu attributes its decline mainly to a change in revenue mix; MiniMax attributes its rise mainly to improving infrastructure efficiency. The revenue multiples in the title are our arithmetic.
Claim Two: Who Funds the Labs
The filings show where the two listed labs got their money in the half. MiniMax says it funded its cash requirements “mainly from cash received from financing activities”. Zhipu says its cash rose “mainly attributable to the receipt of proceeds from the listing and an increase in bank loans”.
Zhipu’s bank loans reached RMB2,224.8 million at 30 June, up from RMB689.5 million at the end of 2025, “primarily due to funds borrowed to meet operational needs”. By our arithmetic, the loans equal 56% of its RMB3,993.7 million of cash and cash equivalents. The filings we read do not settle Lee’s broader point about how China’s labs compete; they show listing proceeds, financing and bank debt.
Claim Three: Export Controls and the Cash Clock
Lee’s “three or four years” describes Chinese AI startups in general. The two listed labs give a yardstick. MiniMax’s cash balance, which includes short-term investments, was $1,322.8 million at 30 June; its adjusted net loss for the half was $293.0 million. By our arithmetic, that balance covers about 27 months at the first-half pace, or about 22 months against the reported loss.
Zhipu held RMB3,993.7 million of cash and cash equivalents against an adjusted net loss of RMB1,964.1 million for the half. By our arithmetic, that covers about 12 months at the same pace. Losses are not the same as cash burn, and neither figure counts money raised after 30 June.
Zhipu’s filing also describes how compute limits shaped its spending. It says it now integrates domestic chips as a primary inference resource, running inference across a cluster of over 100,000 domestic chips, with unit token inference costs down 80% since the start of the year. “Given the persistent global supply constraints on compute, this path ensures that the Company’s cost curve remains within its own control”, the filing says.
Chinese demand for Nvidia capacity is still on record. Reuters reported on 12 November that Chinese AI companies are trying to rent Nvidia-powered computing capacity at data centres outside China, as we covered in Chinese AI Firms Rush for Nvidia Cloud; Bill Waits 304 Days. Nvidia’s 10-Q puts customers headquartered in China, including Hong Kong, at 8.2% of its revenue in the quarter to 26 July, by our arithmetic.
As we read it, the business lever is price. Lee’s case is that Chinese labs win share by charging a fraction of American rates. The filings show revenue up three to five times in a year, gross margins of 18% to 26%, and losses larger than revenue; they attribute the margin moves to revenue mix and infrastructure efficiency, not to price. As we read it, the open question is less whether these labs can sell than whether their cash lasts as long as the price gap.
The Structural Read
As we read Lee’s argument, open weights are a distribution choice that trades margin for footprint, which is how he frames the iPhone and Android comparison.
As we read Zhipu’s filing, export controls show up less as a spending cap than as a design constraint: the company moved inference onto more than 100,000 domestic chips and cites persistent global supply constraints on compute.
“Chinese companies don’t make nearly as much money, but they’re going to get the footprint and the market share.”
Three Implications
FOR REVENUE Zhipu’s open platform and API revenue rose 2,735.7% to about RMB825 million, 86.5% of the total; MiniMax’s first half exceeded its full-year 2025 revenue of $79.0 million.
FOR MARGINS Zhipu’s gross margin fell from 50.0% to 26.4% on a change in revenue mix, while MiniMax’s rose from 12.1% to 17.9% on improving infrastructure efficiency.
FOR FUNDING Zhipu’s bank loans rose to RMB2,224.8 million from RMB689.5 million, and MiniMax funded its cash requirements mainly from financing activities, the filings say.
The framework opens: “Every technology stack has layers. Some are scarce, some are abundant.” Its rule is to close the scarce layer and open the abundant one.
As we read it, Lee describes Chinese labs opening the model layer to win share while American labs keep it closed and charge for it; the Hong Kong filings show the revenue that strategy has produced so far. The framework predates the interview and does not discuss these companies.
What Is Not Established
The quotes come from YouTube’s automatic captions, which can mistranscribe words; Fortune says the full conversation will be published later. Lee’s three-to-four-year remark referred to Chinese AI startups in general, not to Zhipu or MiniMax, and 01.AI was not among the filings we read.
The months-of-cash figures use first-half losses as a pace, and both labs’ revenue and costs are changing fast. The interim results are unaudited. We did not contact Lee, 01.AI, Fortune, Zhipu or MiniMax for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Kai-Fu Lee argues that Chinese AI labs will find ways to make money, are venture-funded rather than state-directed, and are held back more by thrift than by export controls. The Hong Kong filings back the revenue growth: Zhipu’s rose about five times and MiniMax’s 3.8 times. At the first-half pace, their 30 June balances covered roughly 12 and 27 months of losses, by our arithmetic.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. We read Kai-Fu Lee’s post on X of 11 November and his X profile, and the full YouTube auto-generated captions of the Fortune Daily video uploaded on 10 November; we did not use the host introduction’s passage on the Nvidia board matter. We read the Zhipu (Z.ai) interim results announcement of 31 August 2026 and the MiniMax interim results announcement of 26 August 2026.
The Reuters report and Nvidia 10-Q figures are as read for our piece of 12 November. The revenue multiples, months of cash and percentages are ours. We did not contact Lee, 01.AI, Fortune, Zhipu or MiniMax.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
