Bank of China's first 'token loan' ties credit to AI usage
China's banks are starting to lend against something with no physical form: the tokens an AI company burns through. Bank of China has reportedly issued the country's first compute "token loan," and the product signals where AI financing is headed — past the data center and into the application layer.
Bank of China has extended what's reported as China's first compute "token loan," sizing credit lines for AI companies by how many tokens their products actually consume. Guangdong launched the product on August 14 in Guangzhou's Haizhu district, developed by the bank's Guangzhou branch. Unlike traditional lending that leans on fixed-asset collateral, the loan determines credit limits from a company's contracts or its token consumption, with guarantees spanning credit, accounts-receivable pledges, order financing, and combined structures — new companies can apply under certain conditions. The bank has already extended about 28 million yuan in pilot credit, according to local coverage.
The product rests on new token-economy infrastructure: the Guangdong Token Trading and Service Center, set up in late July, prices models, data, and compute in unified token units, aiming to make AI resources a standardized, tradable commodity.
Why it matters: AI financing is drifting down the stack. Inference is now a recurring operating cost for application companies rather than a one-time hardware purchase — yet those firms are asset-light, so conventional banks struggled to lend to them. Tying credit to token usage lets lenders follow the actual activity of AI applications, a shift from contract-based lending (invoices, service agreements) toward usage-based lending (measurable consumption as evidence of future revenue). Bank of China is already deep in this territory: it launched its sci-tech compute loan program in May 2025 with roughly 8 billion yuan in planned cooperation, pledged 1 trillion yuan of support for the AI industry chain in January 2025, and by June 2026 counted more than 5,200 partner companies across the chain.
The caveats are real. The borrower is unnamed, the terms are undisclosed, and the bank has not published an announcement — so treat the loan as a reported pilot, not a proven asset class. Token volume is also a noisy credit signal: it can balloon from free trials, inefficient prompts, or bot traffic, and it says nothing about margins. The test is whether the bank can turn usage telemetry into auditable evidence rather than a vanity metric.
What to watch: whether the pilot becomes repeat lending to other application companies, and whether usage-based credit spreads beyond Guangdong.
If a bank sized your company's credit line off its API bill, would that be a fairer deal — or a dangerous one? Tell us in the comments.
