On one hand, the intelligent computing center project is queuing up for construction, while on the other hand, GPU rental prices have skyrocketed by 50% in six months. The demand for computing power has surged by 417% year-on-year, while the supply growth rate is only 128% - this ice and fire of supply-demand imbalance is reshaping the underlying logic of the entire industry.

High end cards become scarce assets

In the first half of 2026, the Chinese computing power market will present two completely different images: local governments will queue up to approve data center projects, with signed investment amounts often reaching tens of billions; On the other hand, small and medium-sized enterprises cannot find computing power with money, and having value but no market has become the norm. According to IDC's prediction, the scale of China's intelligent computing power will exceed 1117.4EFLOPS by 2027.

The price data confirms the degree of supply-demand tension. The average on-demand rental price of NVIDIA B300 has increased by over 50% compared to November 2025; Baseten disclosed that its cloud service provider plans to increase the B200 lease renewal price by approximately 94%. The one-year lease of H100 will increase from $1.70/card · hour in October 2025 to $2.35 in March 2026. Starting from July 12th, Kingsoft Cloud's AI computing power products in China will increase in price by 15% to 50%, followed by Tencent Cloud, Alibaba Cloud, and Baidu Cloud. The delivery cycle of high-end thousand card clusters has been extended to 12 to 15 months.

According to data from the China Academy of Information and Communications Technology, the domestic demand for AI computing power in the first quarter of 2026 surged by 417% year-on-year, while the supply growth rate was only 128%.

Tier 1 is fully loaded, Tier 2 is idle

In early July, Meta was exposed to be planning to sell idle AI computing power to external parties, causing a panic like plunge in global technology stocks. But the truth is far more complex than stock price fluctuations. Industry executives generally believe that the sale of excess capacity by Meta and xAI is an exception, as the overall industry's demand for computing power far exceeds supply.

The computing power market has formed a clear stratification: the supply and demand of Tier 1 professional intelligent computing centers are basically balanced, and prices are returning to rationality; Tier 2 investment projects are scattered and lack unified scheduling capabilities, with utilization rates hovering at a low point of 30% to 50% for a long time.

The solution at the national level is clear: the national integrated computing power dispatch platform is accelerating its implementation. The National Data Administration has clarified that the next step will focus on four major directions: the East West Computing Project, integrated computing power network monitoring and scheduling, computing power collaboration, and edge computing power. The national super computing Internet core node has been launched, which can provide more than 100000 domestic AI computing power. In this wave of computing infrastructure frenzy, who will swim naked after the tide recedes? The answer will be given in 2026.Editor/Cheng Liting