Favorable policies continue: Local support details accelerate rollout
In September 2026, several provinces and cities across China issued detailed policies to support the deployment of AI large model applications. A major eastern city introduced a special subsidy program for large models used in government services, healthcare, and education, offering rewards of up to several million yuan for enterprises that complete filing and actual deployment. Meanwhile, a central province released computing vouchers and model evaluation subsidies to encourage small and medium-sized enterprises to access public computing platforms for fine-tuning and inference deployment. These policies have generally lowered the initial cost for enterprises to use large models and simplified the application filing process, creating an institutional environment for accelerated penetration into vertical industries.
Vertical large models launch intensively: From healthcare to finance sectors
Since September, multiple AI companies have announced that their medical large model products have passed national filing and are now in practical diagnostic assistance applications. These models focus on image recognition, medical record structuring, and medication recommendations, and have launched pilot collaborations with several top-tier hospitals. Additionally, the financial sector has also welcomed the launch of multiple large model products covering intelligent investment advisory, risk compliance review, and customer service. Notably, these vertical models generally adopt a “base model + domain knowledge base” architecture, maintaining general capabilities while significantly improving specialized accuracy for specific tasks. At the same time, a batch of industrial large models for the manufacturing and energy sectors have entered small-scale client trials, signaling that large models are extending beyond dialogue and content generation into more physical scenarios.
Investment and financing trend shifts: Inference optimization and edge deployment favored
September 2026 saw a clear structural shift in the AI investment and financing market: funding enthusiasm for foundational model training cooled, while startups focused on inference optimization and edge deployment attracted significant capital. Multiple teams specializing in model compression, quantized inference, and edge computing hardware adaptation announced new funding rounds, each reaching hundreds of millions of yuan. Investors generally believe that as large model applications enter the phase of large-scale deployment, reducing inference costs and improving privacy protection have become critical bottlenecks. Furthermore, lightweight large model solutions for smart terminal devices also received multiple rounds of industrial capital backing, with some products already pre-installed in smartphones and IoT devices. This funding cycle indicates that the industry is shifting from “competing on parameters” to “competing on deployment,” with technological dividends spreading into engineering and infrastructure areas.