According to model estimates, of every $100 in AI capital expenditure, semiconductor chips account for $50, with accelerators accounting for $25 and memory chips for $15, forming the largest expenditure source; network equipment accounts for $15, and the two together account for 65% of the total investment. Power systems account for $20, with grid connection and independent power generation accounting for $10, highlighting the importance of stable power supply for data center expansion. Cooling and infrastructure construction each account for $7.50, totaling 15%. Overall, AI capital expenditure is highly concentrated on chips and network equipment, but power, cooling, and infrastructure also constitute significant investment needs. As the scale of computing clusters expands, grid connection time, power supply capacity, and cooling efficiency may become key factors restricting construction progress.
