The pricing logic of the computing power industry—traditionally based on resource consumption—is being rewritten. A new model combining a guaranteed minimum with Token-based excess revenue sharing has been piloted between several leading domestic computing power leasing service providers and their strategic customers. Under this model, the base guaranteed service fee is lowered, and when a customer's Token usage exceeds an agreed threshold, an additional revenue share is charged on a proportional basis. Service providers give up a portion of their fixed income in exchange for a stake in the incremental revenue from downstream inference business, achieving risk sharing: customers reduce upfront rigid expenditures, while service providers can earn higher comprehensive income as usage grows. At present, this model is being piloted only with leading strategic customers that have a high proportion of inference business and strong expected usage growth. Traditional fixed long-term leasing remains the core business and is unlikely to be fully replaced in the short term. The new model also brings challenges and risks: on the one hand, it pushes chip manufacturers to adopt metrics such as actual Token output per unit of hardware as new performance benchmarks, forcing them to strengthen their technical adaptation capabilities; on the other hand, downstream usage fluctuates, and these fluctuations are compounded by tightening risk controls in computing power project financing and the difficulty of realizing value from hardware disposal. Ultimately, the effectiveness of implementation will depend on sustained customer business growth and stable equipment operation.