What happened
Bloomberg News reports that a China Telecom Research Institute report, cited by China Central Television, describes a strategic shift in China’s AI industry from competing primarily on large models and computing power toward deploying and commercializing AI agents. The report projects nearly tenfold annual growth in China’s computing demand over the next two to three years and says inference could represent 80% of the country’s computing-power market by 2029.
Bloomberg published the report on September 12, 2026, attributing its account to Bloomberg News. It says the China Telecom Research Institute report was cited by China Central Television on Saturday. The report characterizes the industry’s direction as a move away from primary emphasis on large models and computing-power competition and toward the deployment and commercialization of AI agents.
The reported forecast says AI agents could drive nearly tenfold annual growth in China’s computing demand over the next two to three years. It also projects that inference computing will account for 80% of China’s computing-power market by 2029, overtaking training-related demand. The article does not identify specific products, deployment customers, launch dates, access terms or prices.
Source details: bloomberg.com ↗
Why it matters
If the projections are accurate, China’s AI infrastructure priorities could move toward inference capacity and the systems required to operate agents in production, rather than focusing mainly on training increasingly large models. That would affect cloud providers, telecom operators, enterprise software companies and government planners. However, the figures come from a reported industry forecast, not independently verified measurements, and the source provides no methodology, company-level commitments or evidence that the shift has already occurred across the market.
The forecast points to a potentially important change in where AI infrastructure investment is concentrated. Inference is the computing used to run models for users and applications, while training creates or updates models. A market increasingly shaped by inference would place greater importance on reliable operational capacity, latency, distribution and the software needed to connect models with tools and workflows.
The practical significance remains conditional. Bloomberg is reporting the research institute’s projections rather than independently confirming them, and the source text gives no methodology, baseline figures or named commitments from Chinese companies. It therefore supports treating the figures as a strategic signal, not as established market data.
What to watch next
The key developments to watch are concrete agent deployments, new inference infrastructure, and disclosed spending or policy decisions that test the report’s projections. It is also unclear which sectors or companies the report covers, how it defines AI agents, and whether the forecast refers to total computing demand or demand attributable specifically to agent workloads.
Evidence of the projected shift would include disclosed increases in inference capacity, commercial agent deployments, telecom or cloud infrastructure investments, and revenue tied to agent-based services. Company announcements or regulatory plans could show whether the trend extends beyond research and policy messaging.
Important unknowns include the report’s definition of an AI agent, the industries included in its forecast, the current size of China’s training and inference markets, and whether the nearly tenfold growth estimate applies to total computing demand or only demand associated with agent workloads. No access or pricing information is relevant because this is not a product announcement.