To further support innovation and development in artificial intelligence, Jiangsu has officially implemented a provincial-level AI innovation support tool and implementation plan.
On October 9, five departments including the Jiangsu Provincial Development and Reform Commission, the Department of Finance, and the Data Bureau jointly issued a notice on the "Implementation Plan for Supporting Artificial Intelligence Innovation and Development in Jiangsu Province," introducing three AI subsidy vouchers: Token Vouchers, Corpus Vouchers, and Model Vouchers. These use fiscal funds to reduce the costs of large model research and development and computing power invocation for market entities, with Token Vouchers providing a maximum subsidy of up to 5 million yuan per individual entity.
According to reports, the Token Vouchers issued by Jiangsu are not traditional paper consumption vouchers but fiscal subsidies paid on a "reimbursement based on actual expenses" basis. They are aimed at enterprises registered and operating in Jiangsu, universities and research institutes, scientific research institutions, and OPCs (One-Person Intelligent Agent Companies), supporting these entities in purchasing intelligent computing power leasing and token invocation services for large model training, inference, and intelligent agent development and deployment.
Under the subsidy rules, the total combined subsidy from Jiangsu province and its cities shall not exceed 50% of the actual computing power payment, with a cap of 5 million yuan per individual entity. Funds are borne by the provincial finance department at 50% of the total reimbursement amount for each locality, with the remaining portion provided by prefecture-level cities.
Regarding application thresholds, computing power procurement expenditures must be no less than 10,000 yuan; the large models invoked must have completed filing with the Cyberspace Administration of China for generative AI; there must be no affiliated relationship between the purchaser and the computing power supplier; and entities must have good credit standing with no major violations or records of non-compliance in the past three years.
The same computing power expenditure follows the principle of "higher amount without duplication" and cannot be stacked with similar AI subsidies within the province; provincial funds are subject to total volume control, and reimbursement ratios will be lowered when budget quotas are tight.
The new policy covers computing power procurement contracts with a start date of July 1, 2026 (inclusive) or later, and the policy is valid until June 30, 2027. Enterprises submit contracts, invoices, payment vouchers, and token measurement materials online through the Jiangsu Integrated Computing Power Monitoring and Scheduling Platform for application review, with settlement based on actual expenses; rebates, refunds, and other amounts are not included in the subsidy base.
At the same time, the Token Vouchers released this time are accompanied by Corpus Vouchers and Model Vouchers, forming a complete "support combination." Corpus Vouchers subsidize the procurement of compliant training datasets and data annotation expenses; Model Vouchers support enterprises in developing their own vertical industry large models, covering the full chain of AI research and development costs from data and models to computing power invocation.
Prior to this, Yangzhou had already piloted municipal-level Token Vouchers, and the implementation of this provincial policy marks Jiangsu's promotion of the pilot experience across the entire province.
Industry insiders say that Jiangsu's subsidy mechanism based on token measurement breaks away from traditional computing power subsidy approaches by reimbursing based on actual invocation volume, which is more aligned with the current pay-per-use business model of large models and explores a new paradigm for fiscal support of China's AI industry.