The Shenzhen Municipal Science and Technology Innovation Bureau has officially opened applications for its 2026 "Computing Power Voucher" program. The initiative is designed to provide financial support to enterprises, universities, and research institutions that rent intelligent computing power for the training and inference of large artificial intelligence models. Eligible entities can submit their applications by June 5th, with a maximum annual subsidy of 10 million yuan per applicant.

This policy follows the "Several Measures for Shenzhen to Build a Pioneer City in Artificial Intelligence," introduced at the end of 2024, which proposed issuing such vouchers to help innovators reduce the costs associated with AI model development and training.

Applicants can request voucher quotas from the Shenzhen Municipal Science and Technology Innovation Bureau. To qualify, an organization must sign a service contract with a pre-approved service provider to rent intelligent computing power for AI model tasks.

According to the "2026 Shenzhen Computing Power Voucher Application Guide," support is granted when an organization's annual expenditure on renting non-affiliated intelligent computing power reaches 500,000 yuan or more. A single voucher can offset up to 50% of the service contract value. The total annual voucher amount for any single applicant is capped at 10 million yuan. Organizations with multiple service contracts must submit a consolidated application within this total quota.

Notably, the program offers enhanced support for startups. Companies registered for less than one year that meet the 500,000-yuan annual spending threshold can have their voucher offset ratio increased to up to 60% of the contract value.

Applicants must submit standardized service contracts through the Shenzhen Science and Technology Business Management System. All applications must be filed directly by the applying entity; submissions through intermediaries are not permitted. When redeeming the vouchers, applicants will be required to provide supporting documentation, such as details of the AI model training/inference conducted and its application scenarios, to prove the computing services were completed.