Tencent says US remote-access chip curbs may slow next Hunyuan model, sees no substantial cloud revenue impact

Text
iHOLYSHITVISUAL / TECHCompany newsiHOLYSHIT / ILLUSTRATION
iHolyShit illustration

Tencent released its third-quarter results in Hong Kong, China, on November 18 and addressed new US restrictions on remote access to advanced AI chips. Management said its cloud business primarily runs inference and should avoid a substantial adverse revenue impact, while computing constraints could lengthen development of the next Hunyuan model.

Tencent Holdings released its unaudited third-quarter results for the period ended September 30, 2026, in Hong Kong, China, on November 18, and addressed new US restrictions on remote access to advanced AI chips during its earnings call. Management drew a distinction: cloud revenue is not expected to suffer a substantial adverse impact, but model training faces constraints that could lengthen development of the next Hunyuan model in the short term.

The restrictions follow an interim final rule issued on November 16 by the US Department of Commerce’s Bureau of Industry and Security (BIS). According to the agency, the rule addresses remote access to advanced super-intelligence chips and servers through data centers in third countries, seeking to prevent circumvention of existing export restrictions on physical items. In plain terms, remote access means using computing equipment in a data center elsewhere over a network, without moving that equipment to the user’s location.

Reuters reported on November 18 that Tencent Chief Strategy Officer James Mitchell told analysts its cloud business primarily runs inference, so the company “actually don't see an impact.” More efficient models can now run on a “surprisingly limited number” of domestic chips, he said. He added that Tencent did not expect its cloud revenue to be substantially adversely affected by the changes. This was management’s assessment of the business impact.

Training and inference use computing resources for different purposes. Training enables a model to learn from data, while inference uses an already-trained model to handle user requests. Tencent’s response centered on that distinction between serving requests through existing cloud services and developing its next model. The company believes its current inference business can accommodate the changes, while acknowledging tighter constraints on advanced computing resources for training.

Reuters quoted Tencent President Martin Lau as calling the tighter restrictions “unfortunate,” saying they reduce the advanced computing resources available to Chinese companies and could slow development of Tencent’s next model in the short term. Reporting on the same call, TMTPost quoted him as saying that both rented and self-built computing capacity face constraints. Model training would inevitably be affected, and releasing the next Hunyuan version could take longer.

Management also discussed efficiency and supply chains. According to TMTPost, Lau said computing efficiency in training and inference is improving rapidly each year, and the methods used to train Hy4 were already highly efficient. Reuters also quoted him as saying that Chinese companies would rely more heavily on domestic supply chains and chips over the longer term, and that the latest restrictions could accelerate that process. These are expectations about future developments, rather than a completed supply-chain transition.

Tencent’s financial results provide context for its current cloud business. Official third-quarter revenue was RMB207.3 billion, up 8% year on year. FinTech and Business Services revenue rose 9% to RMB63.7 billion. Within that segment, cloud services revenue grew 35%, which the company attributed primarily to demand for AI-related services and continued international expansion.

For cloud customers, Tencent offered an assessment of the impact on its current business and revenue. For users following Hunyuan updates, the risk identified by management concerns the development cycle of the next model. These involve different uses of computing resources, so an expectation of no substantial cloud revenue impact can coexist with potentially slower model development.

iHolyshit review

I find Tencent’s distinction between inference services and model training the most useful point. The 35% increase in cloud revenue describes the quarter’s business performance, while potentially longer development of the next Hunyuan model reflects management’s expectations about computing constraints. To me, treating those as separate signals is more accurate than describing the entire AI business as either affected or unaffected.

Share this article

What do you think of this story?

Anonymous feedback, up to 5 reactions per article.

Reader comments

Comment without signing in, or sign in to display your member name.

Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.

0/500

No comments yet. Be the first to share.

Sources