According to a research report from CITIC Securities, there's been a notable uptick in demand for overseas inference and training computing capabilities lately. This is underscored by Amazon Web Services and Google Cloud successively raising their prices.

In the realm of inference, the swift adoption of Agent products like MoltBot (previously known as ClawdBot) and Claude Code has substantially driven up the need for cloud computing resources. Since the start of 2026, Token calls have witnessed a swift surge, growing rapidly for two to three consecutive weeks.

On the training front, models such as Grok-5 and Veo4 are undergoing continuous refinement. The industry is relentlessly pushing the boundaries of Scaling to fulfill the burgeoning demands for training computing power. Although the large-scale commercialization prospects of current AI applications are still up in the air, CITIC Securities projects that over the next three to six months, with the rapid deployment of AI applications on the inference side and the ongoing model iterations on the training side, the demand for computing power is poised to escalate further.

Should the earnings season in the U.S. stock market confirm the uptick in computing power demand, it would ease the prevalent market concerns about a 'computing power bubble' in the computing power sector. This, in turn, could herald a fresh wave of growth for the computing power industry chain.