Currently, with the deepening application of AI agents in various industries, the number of token calls has maintained a rapid growth trend. How to further reduce the cost of computing power in the future has become the focus of efforts by all parties. Industry insiders explained that they are using a computing power platform to coordinate the use of multiple major mainstream models to comprehensively reduce the cost of the term. At the same time, it is also speeding up the construction of large-scale computing power clusters across the country. However, large-scale computing power clusters supported by domestic computing power chips are viewed by the industry as the key to the next step in cost reduction. Furthermore, in the next three to five years, the implementation of new, more cutting-edge technologies such as optical chips is also expected to become another exploration direction for reducing computing power costs. “Compared to electronic chips, the core of an optoelectronic chip is that it has lower computational latency and lower power consumption. The cost per token unit will be reduced by 50%, or more.”

Zhitongcaijing · 2d ago
Currently, with the deepening application of AI agents in various industries, the number of token calls has maintained a rapid growth trend. How to further reduce the cost of computing power in the future has become the focus of efforts by all parties. Industry insiders explained that they are using a computing power platform to coordinate the use of multiple major mainstream models to comprehensively reduce the cost of the term. At the same time, it is also speeding up the construction of large-scale computing power clusters across the country. However, large-scale computing power clusters supported by domestic computing power chips are viewed by the industry as the key to the next step in cost reduction. Furthermore, in the next three to five years, the implementation of new, more cutting-edge technologies such as optical chips is also expected to become another exploration direction for reducing computing power costs. “Compared to electronic chips, the core of an optoelectronic chip is that it has lower computational latency and lower power consumption. The cost per token unit will be reduced by 50%, or more.”