HTSC·2025-07-01 08:26Investment Rating - The report maintains an "Overweight" rating for the technology and computer sectors [5]. Core Insights - The demand for inference computing power is expected to grow faster than the decline in unit computing costs, driven by the increasing user penetration of AI applications and the continuous improvement of algorithms and hardware [1][24]. Summary by Sections Token Demand Growth - The monthly token call volume for Google increased from 9.7 trillion in April 2024 to 480 trillion in April 2025, a growth of 50 times [2][9]. - ByteDance's Volcano Engine reported a daily average token call volume of 16.4 trillion in May 2025, which is 137 times higher than in May 2024 [12][9]. Google Token Demand and Spending - Google's token call volume is projected to reach 2009 trillion in Q2 2025, representing a quarter-over-quarter growth of 223% [3][20]. - The unit computing cost for Google is expected to decrease, with a projected decline of 14% in April, 13% in May, and 13% in June 2025 [3][26]. Microsoft Token Demand and Growth - Microsoft's total token call volume is expected to reach 205 trillion in Q2 2025, reflecting a growth of approximately 100% [4][29]. - The increase in token calls is primarily driven by the rising user engagement with OpenAI's ChatGPT and the penetration of the Deep Research feature [29][30]. Capital Expenditure Projections - Google's inference computing power capital expenditure is projected to increase by 159% quarter-over-quarter in Q2 2025, driven by the significant growth in token volume [24][28]. - The report anticipates that the overall inference computing power demand will continue to rise, offsetting the cost reductions from hardware iterations and algorithm optimizations [24][26].