The idea that data can be turned into money—we have written about this several times over the past year: data assets being recorded on the balance sheet, data property rights registration, and authorized operation of public data. But most policies talk about "your data will be valuable in the future"; the corpus vouchers issued by Shenzhen this time are different—they directly provide real money, and they provide it in both directions.
Two directions, subsidies at both ends
First, subsidize the buyer. For enterprises that purchase corpora from non-related parties through data exchanges for artificial intelligence research and development, funding shall be provided at no more than 30% of the total procurement cost, with a maximum of 2 million yuan per enterprise.
Second, reward the seller. Enterprises that open eligible corpus data through the Shenzhen Public Data Open Platform will receive a maximum reward of 1 million yuan.
The amount of the reward and subsidy is not decided on a whim; it must be determined by taking into account multiple factors such as the amount of data procurement spending, the volume of open data, data quality, update frequency, and application results.
The application window is from September 1 to November 10, 2026, at 18:00, and the channels are 'iShenzhen', the Guangdong Government Service Network, and the Municipal Government Service Center. There is only one week left until the deadline, so enterprises engaged in AI R&D and with a need to purchase useful data should act quickly.
Why "corpus"?
A corpus is the training data fed to AI models. In the era of large models, high-quality datasets have become a scarce resource on par with computing power—computing power can be rented, but industry corpora cannot be rented, because the best industry data is in the hands of enterprises themselves: production records, quality data, customer service conversations, and maintenance work orders.
The logic behind Shenzhen's issuance of corpus vouchers is exactly this: the government uses subsidies to leverage data circulation—buyers dare to buy, sellers are willing to open up, and only then do AI applications have the ingredients to work with. In 2026, it will also explore cooperation with provincial and municipal data exchanges, adding corpus voucher subsidy labels to listed data and building policy outreach into the transaction process.
Two implications for enterprises
At the first level, for enterprises currently working on AI applications: if you have a budget for purchasing external industry corpora, this expense can now be reduced by 30%. Note the compliance requirements—actual operations within Shenzhen and compliant sources of corpus data are basic thresholds; procurement should go through the formal channels of the data exchange, and keep the vouchers.
The second layer, for all enterprises that view data as an asset: the corpus voucher proves that data is not just a cost item. The production, quality, and service data you have accumulated over the years, once cleaned, desensitized, and opened up, can bring both reward income and potentially give rise to new collaboration opportunities. Of course, the precondition is that the data must first be in the system, at scale, and clean enough—data scattered across Excel files on individual computers does not even meet the threshold for asset inventory.
Here we circle back to that old topic again: data governance. Master data standards, online business processes, data quality verification—these fundamentals in ERP implementation have become direct preconditions for economic value in the era of data elements.
The deadline is November 10, and companies in Shenzhen have one week left. Other cities will most likely follow suit with this type of data subsidy—the framework for the "Data Elements ×" initiative in Guangdong Province's new data elements policy was set up long ago. First, get a clear inventory of your data; when the next voucher is issued, you'll be ready to take it.
