2month, ChinaAIThe model usage surged, exceeding that of the United States for the first time. Chinese model manufacturers are occupying the global market with rapid iteration and cost advantages, and domestic computing power demand is experiencing exponential growth.tokenThe call cost is generally lower than that abroad. What are the reasons behind this? Has the decrease in power costs or lower technical costs enabled Chinese large models to have stronger market competitiveness?AI tokenWhat will the future profit-making model look like?AI tokenFee model and legacySaaSWhat is the biggest difference between the charging models? With technological progress,AIWill the pricing of services gradually move towards high customization and flexibility?
Frost & SullivanfrostSullivan,hereinafter referred to asLi Qing, China director of Frost & Sullivan, was interviewed by NBD to discuss the development of large language models in ChinatokenThe technical root cause and commercial evolution path of cost advantages.
QChinese Large ModeltokenThe call cost is generally lower than that in foreign markets. What are the reasons behind this?
Top Chinese ModelsDeepSeekGeneral Meaning of a Thousand Questions3.5-PlusSparse Hybrid Expert (MoE)”architecture. Although this design has a huge total number of parameters, only a small number are activated during actual inference. This directly reduces memory usage.60%, reasoning throughput improved19times, reducing computing power waste at the source. At the same time, with Alibaba's “Tongyi-cloud-Taking the 'core' system as an example, algorithms, infrastructure, and chips work in deep collaboration. This vertical integration solves the pain point of model-hardware adaptation by maximizing resource utilization through extreme computing power scheduling, significantly reducing the cost of underlying computing power.
QDoes the reduction in electricity costs or lower technology costs enable Chinese large models to have stronger market competitiveness?
The cost advantage is translating into absolute market competitiveness, but this is not just about electricity costs.tokenThe cost core is indeed computing powerGPUDepreciation) and electricity (data center operations). Although China has certain advantages in electricity costs, its greater advantage comes from technical efficiency and the adaptation of domestic computing power.
Q: AItokenWhat will the future profit-making model look like?
SaaSWhat is sold is the 'right to use the tool', andAI tokenWhat is sold is 'fuel to solve problems'. The core difference is whether it sells directly application results or services, or underlying computing resources.
QWith the progress of technology,AIWill the pricing of services gradually move towards high customization and flexibility?
This is an inevitable trend.agentWith the advent of the era, a single pricing model will no longer be able to cover complex business needs. Pricing will evolve towards multi-dimensional dynamic pricing. High-cost operations such as computing consumption, call frequency, task complexity (for example, whether it involves multi-step reasoning, planning, reflection, etc.) will all affect the pricing model.
*This interview has been published in the Daily Economic News, with reporter Song Xinyue. The original title was:2Moonlit well eruption! ChinaAIThe usage volume has exceeded that of the United States for the first time, with four large models ranking among the top five globally. The demand for domestic computing power is experiencing exponential growth.
