On August 14, Guangdong's first specialized financial product for the “token economy,” dubbed “Token Loan,” launched in Guangzhou's Haizhu District, alongside a policy document titled “Several Measures for Supporting the High-Quality Development of the Token Economy in Haizhu District,” known in the industry as the “Eight Token Measures.” Bank of China, China CITIC Bank, and Bank of Guangzhou are participating; the product from Bank of China's Guangzhou branch, called “Compute Token Loan,” has already approved 28 million yuan in credit for early pilot orders.
The most concrete part of the product design is how credit lines are set: based on contracts and token consumption quotas. Collateral relies mainly on credit standing, accounts-receivable pledges, and order-based financing, with the option to combine guarantees and mortgages to raise the limit further. Newly established companies that can't show operating cash flow can apply for financing based on orders, backed by proof of continuous operations or guarantees from an existing operating entity.
Turning Call Volume Into a Risk-Control Variable
For small AI application companies, the hardest part is rarely getting access to models — it's usually the payment cycle. Clients settle accounts quarterly, while model API fees are deducted in real time as they're used, leaving companies to cover the cash-flow gap themselves. The bigger the business grows, the more they have to front, turning growth itself into a source of financial pressure.
Banks have historically struggled to read these companies' balance sheets — no factories, no equipment, servers mostly rented, and the biggest line item on the books is the monthly model bill. Token Loan flips that expense into evidence of operations: stable call volume implies a real business is running, and contracted quotas imply predictable revenue. It's the same logic as e-commerce lenders extending credit based on platform transaction volume, or SaaS companies getting credit lines based on ARR — just with the unit of measurement swapped for tokens.
The 28-million-yuan pilot scale isn't large — it could be a rounding error in the procurement budget of a single compute center. Its significance lies in establishing a replicable pricing basis. Once a bank has internally validated the “call volume to credit line” conversion, the scale of future credit can grow from there, provided the variable's stability holds up over one or two credit cycles.
The “Token Economy” Framing
Haizhu District's policy framework breaks the token economy into four stages — production, circulation, consumption, and value-add — with the eight measures covering token production efficiency, service aggregation, distribution hubs, application scenarios, cultivating “One Person Company” (OPC) ventures, industry upgrading, financial services, and support for going overseas. On July 30, the Guangdong Token Trading and Service Center was established ahead of the loan product, positioned as a compliant and trustworthy hub for token trading meant to address small businesses' high AI procurement costs and mismatched supply and demand.
Treating tokens as something tradable, priceable, and financeable runs straight into an existing industry debate. Per-token pricing has taken plenty of criticism in developer communities, with some calling it a usage-based wealth tax, on the grounds that costs are unpredictable and the room for optimization sits with the model vendors. Local government has taken a different view — precisely because tokens are measurable, they can be folded into statistics, subsidy terms, and credit models.
One reference point: China's daily AI token consumption has reportedly reached the hundred-billion scale. Once a number gets big enough to show up in local economic statistics, building trading hubs and credit products around it becomes a natural next step.
Two Things to Watch as It Rolls Out
First is whether call volume can be shielded from padding. Running scripts to fake call volume is far cheaper than falsifying accounts receivable, so banks need cross-verification using raw billing data from model vendors, not just figures companies self-report.
Second is coverage. The publicly disclosed plan currently applies to small and micro compute-related businesses in Haizhu District — a district-level pilot. The overseas-expansion support and OPC cultivation measures mentioned in the eight-point plan haven't had their specific terms fully disclosed yet, and whether the program expands citywide or provincewide will depend on how this batch of pilot loans performs.
Sources: IT Home, Jinrongjie, CocoLoop, Guangzhou Daily's Dayoo; figures on pilot credit amounts, approval methodology, and the scope of the eight token measures are cross-checked against public reporting — specific product terms from individual banks have not been fully disclosed.