Research from CCID Consultants in the English-language special “Foresight” shows that the generative AI industry's “bigger is better” development paradigm is being challenged by commercial reality. In the structured business scenario, the performance of large models reaches the platform period at 40 to 60 billion parameters. The return on investment of large models above 100 billion parameters is only 5%, while models exclusive to the 7-30 billion parameter field can reach 35%. Inference is the main cost black hole, accounting for 70%-90% of the cost of the entire life cycle of the model; the giant monolithic model also poses compliance governance risks.

Research from CCID Consultants in the English-language special “Foresight” shows that the generative AI industry's “bigger is better” development paradigm is being challenged by commercial reality. In the structured business scenario, the performance of large models reaches the platform period at 40 to 60 billion parameters. The return on investment of large models above 100 billion parameters is only 5%, while models exclusive to the 7-30 billion parameter field can reach 35%. Inference is the main cost black hole, accounting for 70%-90% of the cost of the entire life cycle of the model; the giant monolithic model also poses compliance governance risks.