Policy drivers and investment recovery
In the third quarter of 2026, multiple ministries jointly issued the "Measures for Promoting High-Quality Development of the AI Large Model Industry," explicitly supporting affordable computing power and reduced model invocation costs. During the same period, several AI startups completed new funding rounds totaling over ten billion yuan, with investment focus shifting from foundation models to applications and infrastructure. Industry analysts believe that the combined push from policy and capital is accelerating large models from "showcasing" to "practical use."
Token wholesale strategy and cost optimization
In mid-October, a major domestic AI platform (often referred to in the industry as AI Hub) officially announced a token wholesale plan. The plan offers tiered pricing packages for enterprise users, with bulk purchases eligible for the lowest discounts, hailed as one of the "cheapest token" schemes. At the same time, the platform opened an AI API relay station interface, allowing developers to cache and reuse through an intermediate layer, further reducing duplicate call costs. According to the platform’s announcement, under the new strategy, the average API call cost for enterprises dropped by about 35%. Many small and medium developers indirectly accessed models through the token relay station, effectively avoiding high initial deployment costs.
Channel agent ecosystem fully open
Alongside the token wholesale, the platform announced the launch of an AI channel agent recruitment plan. Third-party service providers can become AI large model agents through the token relay station, helping small and medium enterprises with model integration and operations. The platform provides technical support and incentive sharing, forming a closed loop of "platform-channel-end user." Multiple SaaS vendors have signed up as AI channel agents, expected to cover over 5,000 enterprise customers within the year. Industry experts say the token agent model is likely to solve the "last mile" problem of large model deployment, allowing more vertical industries to flexibly invoke the most suitable model capabilities.
Industry deployment cases and results
For example, an intelligent customer service company accessed the platform’s large model through the AI API relay station and, with the help of a token agent, achieved customized fine-tuning. Its monthly call volume grew tenfold while costs increased by only 60%. Another education technology company used the token wholesale scheme to deploy AI-assisted teaching tools across its national campuses, with student feedback accuracy exceeding 90%. These cases show that lowering the barrier to invocation is releasing substantial long-tail demand, and the commercial value of AI large models is spreading from leading enterprises to small and medium customers.
Outlook: Democratization accelerates the AI-native era
As more platforms follow suit with token wholesale and channel agent models, the business loop for AI large models is gradually taking shape. At the policy level, multiple local governments have set up special AI funds to subsidize small and medium enterprises in purchasing computing services. In terms of investment, market focus is shifting toward profitability and application penetration. It is foreseeable that 2027 will see an explosion in large-scale deployment of large models, with infrastructure like AI relay stations and token wholesale serving as key supports.