Live Updates > Live Update Details
2026-05-25 07:35:39
【AI infrastructure is moving towards full-stack, and the construction of computing networks is accelerating】 (1) NVIDIA's FY27Q1 performance exceeded expectations, and customer demand is shifting from single GPU procurement to full-stack solutions covering switching and interconnection. It is expected that Vera Rubin rack products will be delivered in batches starting in Q3, and the value of CPU, interconnection, switching and other links will increase simultaneously. In the future, AI infrastructure will be upgraded around all links. (2) The computing network is positioned as a national infrastructure, and the annual investment scale is expected to exceed one trillion yuan, covering construction, management, scheduling and operation. As the main force in construction, operators have launched token packages and explored token operation models for C-end users. (3) Google I/O 2026 shows that Google's AI layout has entered the full-stack stage of "computing infrastructure + model + application entry + edge device": TPU 8t/8i verifies that the demand for training and inference computing power is increasing in the long term, and Gemini 3.5 Flash and Gemini Omni enhance the speed, cost and multimodal capabilities of models.
