Chinese banks are beginning to package credit around the operating needs of AI companies rather than treating artificial intelligence as a generic technology label. New lending products described in recent reporting target chip design, computing-power procurement, intelligent-computing centers, model development, and AI applications. The effort reflects a basic financing problem: many AI companies have high research and infrastructure costs but relatively few conventional assets to pledge as collateral.
A China Business Journal report published on August 15 said AI-related lending products are commonly unsecured credit loans with limits up to 10 million yuan. The report cited regional regulatory guidance in Shanghai and Guangdong encouraging banks and insurers to tailor products to the AI industry chain. It also described examples from bank sources showing how lenders are attempting to assess AI firms through business operations, order flow, and specialized credit models.
The development is more specific than a policy statement about supporting innovation. Regulators are encouraging products for different parts of the AI stack, from upstream chip design to downstream model development and applications. Banks, in turn, are trying to translate those policy goals into interest rates, maturity terms, loan limits, and underwriting rules that can work for asset-light companies.
Shanghai and Guangdong Push Lenders Toward AI-Specific Products
China Business Journal reported that the Shanghai financial regulator recently urged banks and insurers in the city to expand the financial supply available to the AI sector. The guidance included differentiated products for distinct needs along the industry chain. “Chip loans” were identified for upstream chip design, while “computing-power loans” were aimed at downstream AI development and application work.
This division is useful because an AI company’s financing needs change depending on where it sits in the value chain. A chip-design company may face long research cycles and tape-out costs. A model developer may need to purchase or rent large amounts of computing capacity. An application company may need working capital to integrate a model into a product, serve customers, or pay recurring inference bills. A single standard technology loan may not fit all three cases.
The report also described Guangdong measures that call on banks, insurers, and funds to support large models, computing power, “AI plus” applications, key AI companies, and innovation projects. The Guangdong approach includes encouragement for intellectual-property-backed financing and technology insurance. That matters because a young AI company may hold software, algorithms, and patents rather than land, equipment, or a long history of cash flow.
China’s effort to expand AI finance has been developing alongside a major increase in computing demand. EastFrontier previously reported that China’s AI computing power reached 962 EFLOPS. Financing mechanisms that help companies purchase or rent capacity are becoming more relevant as model training, inference, and deployment draw larger operating budgets.
Unsecured Credit Up to 10 Million Yuan Targets AI Companies
The most concrete lending detail in the China Business Journal report is the 10 million yuan ceiling cited for many AI-related unsecured credit products. The report said one source at a state-owned bank in northern China described a newly launched computing-power loan supporting computing procurement, intelligent-computing centers, and large-model research and development. The source said the product could provide up to 10 million yuan after an assessment of operations and orders.
That same source described an annual interest rate between 2.4% and 2.6%, while noting that customer inquiries had not been especially numerous. The rate and the level of customer interest are not universal market figures. They are source-reported examples from one bank, and the report does not identify the institution. Still, they show the kinds of commercial terms lenders are testing as they seek to serve AI companies without relying only on traditional collateral.
A second local-bank source told the newspaper that its AI-focused product was aimed at artificial-intelligence and technology companies. That source described a 2.6% to 3.0% annualized rate, a three-year term, and possible renewal at maturity, again with a maximum limit of 10 million yuan. The source said the bank had developed a model specifically for AI companies that could provide more credit and longer terms than its earlier assessment approach in comparable cases.
The distinction between an ordinary loan and an AI-specific model is important. A lender may decide that a company with recurring customer contracts, confirmed orders, or stable operating activity deserves a different assessment from a startup that only has a research plan. The details of those models remain largely undisclosed. China Business Journal reported the broad approach, not the full credit criteria used by each bank.
AI Finance Must Balance Policy Support With Credit Discipline
The banking push carries clear opportunities for AI startups, but it also brings conventional credit risks into a fast-moving sector. A company can have a promising model, high usage, or a large compute budget without having predictable revenue. Lenders still need to understand repayment capacity, customer concentration, contract quality, and the cost of the computing resources a borrower plans to use.
That challenge is especially clear for computing-power loans. A model developer may need expensive capacity before it has a mature product. An AI application company may face recurring inference costs that rise as customers use its service. Financing can help bridge those needs, but a lender must distinguish between sustainable demand and a temporary wave of subsidized or experimental use.
The article’s sources point toward several ways Chinese banks may try to manage that risk. The state-bank source cited operating conditions and orders. The local-bank source described a dedicated assessment model. Guangdong’s measures include intellectual-property financing and technology insurance. These tools indicate that lenders are trying to combine traditional credit indicators with industry-specific evidence rather than base a decision only on a company’s AI branding.
The trend also connects to the large funding flows already moving into embodied AI and model development. EastFrontier’s review of China’s world-model startups showed the scale of equity capital pursuing physical AI. Bank lending serves a different function. Equity investors can finance longer-term experimentation in exchange for ownership. Banks require defined repayment paths and cannot treat every AI company as a venture-style bet.
That difference will shape whether AI loans become a meaningful part of the sector’s financing mix. A 10 million yuan unsecured facility can be useful for a company buying computing services, funding an application rollout, or supporting a defined project. It is unlikely to replace the large equity rounds needed for frontier model training or full-scale hardware manufacturing.
The next step is likely to be more specialization. Shanghai’s chip-loan and computing-power-loan language already separates different AI-industry needs. Guangdong’s support for models, computing, applications, and intellectual property points in the same direction. The August 15 report shows Chinese banks moving from broad encouragement toward specific products. Their long-term impact will depend on whether lenders can price AI risk accurately and whether borrowers can turn advanced technology into durable revenue.