Tencent's established businesses still grew operating profit 8%; the drag comes from new AI products and capital spending that management expects to stay near peak levels through 2027.
Tencent’s spending on artificial intelligence is now large enough to pull down its headline profit. Non-IFRS operating profit fell 6% from a year earlier to RMB68.1 billion in the third quarter, even as revenue rose 8% to RMB207.3 billion, according to the company’s results release on November 18. Capital expenditure payments jumped 225% to RMB65.1 billion, and free cash flow was negative RMB69.1 billion. Tencent said free cash flow would have been a positive RMB16.9 billion without its prepayments for compute.
The pressure is concentrated in what Tencent calls new AI products: its Hy models, the Yuanbao chatbot, CodeBuddy, WorkBuddy and Xiaowei, an assistant inside WeChat. Excluding its new AI products, Tencent’s operating profit grew 8% to RMB83.3 billion, with a 40% margin that was stable from a year earlier. Including them, the margin fell to 33% from 38%. The roughly RMB15.2 billion difference between the two measures is the closest gauge Tencent offers of what the AI push costs in a quarter.
On the analyst call, President Martin Lau addressed two weak spots directly. He said US restrictions on advanced chips would slow work on Tencent’s next model, and he said Yuanbao had fallen short of the company’s hopes.
Advertising and games carried the quarter while AI costs climbed
Marketing services revenue rose 19% to RMB43.0 billion, as consumer goods and local services advertisers spent more and Tencent credited AI-driven ad and content recommendation. Domestic games revenue grew 11% to RMB47.6 billion, while international games fell 9% to RMB18.9 billion. FinTech and Business Services revenue rose 9% to RMB63.7 billion, helped by a 35% increase in cloud services revenue that chairman Ma Huateng linked to WorkBuddy, Tencent’s AI office agent, which the company says has more than 10 million daily active users.
Costs rose faster. Chief financial officer John Lo said research and development spending increased 42% to RMB32.4 billion, mainly for AI models and AI features, and selling and marketing expenses rose 6% to RMB12.1 billion, largely to promote AI products such as WorkBuddy, according to an Investing.com transcript of the call. Tencent ended September with net debt of RMB13.7 billion, compared with net cash of RMB58.2 billion at June 30. It still bought back about 15.9 million shares for roughly HK$7.0 billion during the quarter. Investors had flagged the cash drain before the report: Tencent shares fell below HK$400 in October after second-quarter capex rose 176%.
Lau says chip curbs will delay training more than inference
Asked about the latest US limits on chip access, Lau said: "I would say it’s unfortunate, but a few points." He added: "It sets everyone back in China." On model development, he said "when we look at training, then it’s going to be impacted," and that "it would take probably longer time for us to come up with the next version in the short run." He noted that training efficiency on Hy4, which Tencent says ranked first among Chinese models on the Agent Arena leaderboard at launch, is high.
Management drew a line between training and inference, the work of running models for users. Lau said domestic supply should be able to cover Tencent’s inference needs, and chief strategy officer James Mitchell said inference "can be run very effectively on a surprisingly limited number of domestic chips." According to IT Home, citing STAR Market Daily, management said recent restrictions had cut the compute actually available to Tencent, affecting both leased and self-built capacity. Earlier this month Oracle reportedly canceled a chip lease worth about US$7 billion that would have given Tencent access to up to 100,000 advanced chips in Southeast Asia.
Lau framed the long-run outcome as a shift in supply. The restrictions "will only accelerate the development of Chinese chipmakers," he said, and "all Chinese tech companies ultimately will rely on domestic chips and supply chains," Nikkei Asia reported. A separate IT Home summary, which cited Jinrong Jie for the training comments, noted that Tencent has released a major new Hy version roughly every two months on average this year.
Tencent is redirecting consumer AI spending toward WeChat’s Xiaowei
Lau was blunt about Yuanbao. "Yuanbao was not as successful as we expected," he told analysts, adding that Tencent had "reduced the spending on Yuanbao quite significantly." According to IT Home’s report, he said Yuanbao will now concentrate on improving model capability rather than chasing user numbers. The app ran a RMB1 billion red-packet giveaway during the 2026 Spring Festival and later reported more than 50 million daily active users, IT Home noted.
Xiaowei is taking its place as the consumer bet. Lau said daily active users interacting with Xiaowei have reached 30 million, and that the assistant is deliberately less powerful than rival agents so Tencent can offer it to every WeChat user at low cost. IT Home reported that its quarterly cost after a full launch is expected to be lower than what Yuanbao cost earlier this year.
Spending will stay high for now. Mitchell said "cash CapEx may bounce up and down around this peak level through the rest of 2026 and 2027," and Lau said Tencent sees "2028 to be much better and beyond," when management expects equipment prices to fall noticeably, IT Home reported. The release date of Tencent’s next Hy flagship model, after a two-month cadence this year, will show how much the chip limits are slowing it down.