According to Heilongjiang Daily, at an early-November digital-intelligence project signing event held by Harbin Economic Development Zone, Harbin Data Exchange Co. and Harbin Chaoyang Technology Development Co. signed a cooperation agreement completing Heilongjiang province's first general-purpose LLM token-service transaction.
The logic: turning model inference into a wholesale-purchasable commodity. Tokens are the standardized billing unit of LLM inference, consumed as used — buying alone means retail prices, while purchasing through the exchange works like group-buying at wholesale rates. The deal uses pay-as-you-go billing and online delivery, adaptable to software development, smart manufacturing and digital government scenarios, with expected procurement cost savings of around 40%.
The platform is the province's only licensed state-owned data trading institution, approved by the Heilongjiang provincial government and operational since 2025, handling provincial data registration, trading, supply-demand matching and asset enablement. The buyer, Chaoyang Technology, is a "specialized and sophisticated" enterprise with over a decade in government software; since 2024 it has pivoted to government AI, shipping more than ten products including AI-powered portal search, policy calculators, digital humans and digital employees — all with LLM inference as baseline demand.
This is another provincial-level replication of the token-trading model. Fujian's Minqing "token factory" adopted pay-per-use pricing when it launched, while Heilongjiang has moved the trading itself into a licensed exchange: fragmented channels, opaque pricing, high integration barriers and weak compliance are common pain points for smaller enterprises buying model services, and on-exchange matching plus filing standardizes each step. The signatories also flagged upcoming compute vouchers and data vouchers for enterprises.
Tokens becoming a tradable asset on an exchange means model inference has formally entered the "factor circulation" system — usage can be priced, matched and subsidized. For model vendors, the exchange channel brings volume orders; for the regional economy, it keeps local AI procurement spending settled within the province. Whether the 40% saving holds depends on whether on-exchange volume keeps compounding.