UBS China software and telecom analyst Wang Xinyi predicts that the demand for data centers in China will accelerate from the second half of this year, benefiting from the increase in domestic GPU production capacity. She mentioned that China's super-large cloud factories, Internet companies, and leading AI companies have maintained a high level of commitment to AI capital expenditures. She expects the data center industry to maintain a growth rate of around 25% in 2026, with a net IT load increase of about 4GW. Wang Xinyi pointed out that token and cloud service prices have risen this year, reflecting strong downstream AI demand, but high-end computing power supply remains tight. After a large-scale tender was released in early January this year, data center customers continue to release data center orders in a rolling manner, with delivery schedules extending to the second half of 2027. The shipment and deployment progress of domestic GPUs is in line with expectations. For example, DeepSeek has stated that the price of its V4 Pro model will decrease significantly in the second half of 2026 with the mass production of the Ascend 950 super node. UBS stated that the construction of an AI data center currently requires tens of billions of RMB without IT equipment. Therefore, one of the main challenges is that operators need higher financing capabilities. It is expected that leading companies with quality assets will be able to accelerate capital turnover, further increase capital expenditures, and obtain more orders. UBS China's Internet research director Fang Jinceng predicts that in 3 to 5 years, the profit margin of Chinese mainland cloud service providers can return to a level of 20%, benefiting from the higher profit of AI clouds, the transparency of token prices leading to increased pricing power, and reduced market vicious competition. Fang Jinceng also pointed out that looking at the current pace of the industry, the development of AI agents will increase the usage of tokens by several orders of magnitude, with the next step being communication between agents, further increasing the usage of tokens by another multiple. As for the introduction of paid versions of consumer models in mainland China, Fang Jinceng believes that due to the large number of alternative consumer models in China, it may be difficult to require consumers to pay. Furthermore, the distinction between consumer and enterprise users is becoming increasingly blurred, and he estimates that in the future, consumer models may be charged based on results.