The focus of computing power demand shifts from training to inference, and global AI infrastructure construction is accelerating.
Recently, Oracle, a veteran technology company, shocked the global capital markets with a single-day surge of 40%. This was driven by the company's first-quarter results for fiscal year 2026 (June 1 to August 31, 2025), which showed a 359% year-over-year increase in remaining performance obligations (RPO) for its cloud infrastructure (OCI) business, reaching $455 billion. Nearly 60% of this came from a five-year, $300 billion inference computing power contract signed with OpenAI. A business representative from a leading domestic cloud computing vendor told reporters that while there are still uncertainties between the signing and actual delivery of this massive order, it sends a clear signal to the entire industry chain that computing power, especially inference computing power, is in short supply, bolstering confidence in the industry's continued investment in AI and the cloud.