TF Securities: In the second half of the year, the demand for computing power for inference and training at the data center end will increase synchronously. Focus on core hardware targets benefiting from AI inference acceleration.

Tianfeng Securities pointed out that with the continuation of high Capex prosperity, the model iteration will catalyze a new round of inference demand release, and the computing power end will continue to show a high prosperity trend. Tech giants such as Microsoft, Google, Meta, etc., have successively raised their annual capital expenditure guidance, reflecting the continuous acceleration of AI infrastructure construction. In the second half of the year, the inference and training computing power demand at the data center end will increase simultaneously, driving the continuous improvement of the prosperity of core links such as GPUs, ASICs, optical communication, and liquid cooling. It is recommended to focus on core hardware targets benefiting from AI inference acceleration.