Photo: Nana Dua / Pexels
New export rules targeting overseas data centers rattle Chinese tech stocks and push the country's cloud giants toward domestic chips Washington has closed one of the last side doors Chinese AI companies were using to reach advanced chips. The fallout showed up quickly in earnings guidance and stock prices.
A run of new US measures hammered Chinese technology shares this week. Investors are now trying to work out whether the curbs will slow Beijing’s AI ambitions, or simply force China to build its own path faster.
The loophole Washington just closed #
On November 16, the Commerce Department’s Bureau of Industry and Security announced export controls aimed at a quieter workaround. Chinese firms had been tapping advanced AI chips through data centers outside China, in countries such as Malaysia, Singapore, Thailand, Indonesia, and Brazil.
Under the new rules, any data center supplying substantial computing power to Chinese companies needs a license. Similar restrictions also apply to certain chip shipments.
The workaround mattered because it involved no physical chip ever crossing into China. Remote access to computing power was enough to train and run AI models, and that is exactly what the rule now targets.
Markets reacted fast. Chinese AI stocks Z.AI and MiniMax fell 22% and 34%, respectively, after the rule announcement.
The timing is awkward for anyone who expected a thaw. Between December 2025 and May 2026, more than 879,000 Nvidia H200 and AMD MI325X chips were approved for Chinese buyers.
Alibaba trims expectations, Tencent bets on inference #
Alibaba expects cloud revenue growth to slow to between 30% and 45% next year. That compares with 53% growth in its latest quarter.
The crypto and markets news you actually need—every morning.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The company also signaled it may revise down its long-term target of 20 gigawatts of computing capacity. At the same time, Alibaba is holding firm on its goal of $100 billion in external cloud revenue by 2030.
That different hardware already exists. Alibaba’s in-house T-Head chips are in use, and mass production of next-generation versions is expected by 2027.
Tencent took a more confident tone. The company said its cloud business remains resilient for inference workloads that run on domestic chips.
The distinction between training and inference matters a lot here. Training is building the model, which demands enormous, cutting-edge computing clusters. Inference is using the finished model to answer questions or generate content, which is far less demanding.
Tencent expects delays in training its next-generation models.
Its cloud unit still posted 35% year-over-year growth, according to the research findings, despite the new restrictions.
What this means for China’s AI race #
For investors in Chinese tech, the immediate signal is reduced visibility. Alibaba’s guidance range of 30% to 45% is wide, and that width reflects real uncertainty about how much computing capacity the company can secure. A possible cut to the 20-gigawatt target is worth watching closely. If Alibaba formally lowers it, that would be the clearest sign yet that the curbs are constraining physical buildout, not just headline sentiment.
The sharp drops in Z.AI and MiniMax point to where the pain may concentrate. Smaller AI developers often lack the scale and in-house chip programs that Alibaba and Tencent can fall back on.
On the flip side, the policy may hand domestic chipmakers a captive customer base. Alibaba’s T-Head timeline, with next-generation mass production expected by 2027, is now one of the more important dates on the Chinese tech calendar.
What to watch next: whether Alibaba follows through on revising its capacity target, how quickly T-Head’s next-generation chips reach volume, and whether Tencent’s training delays turn into a measurable slowdown in product releases. Licensing decisions for data centers in Southeast Asia and Brazil will also show how strictly the new rules get enforced in practice.
Disclosure: This article was edited by John Chen. For more information on how we create and review content, see our Editorial Policy.