An inflection point for AI inference is fueling a boom in China's computing power leasing market, where prices are rising and capacity is selling out.
A surge in demand for artificial intelligence inference has triggered a high-growth cycle for China’s computing power leasing market, with prices for both GPU leases and cloud services rising sharply in the first half of 2026. A recent report from Guohai Securities points to a nearly 40% jump in the cost of leasing Nvidia's H100 GPUs, a sign of tightening supply as AI workloads shift from training to inference.
"The AI inference turning point has arrived, and computing power leasing has entered a high-prosperity cycle of rising volume and price," analysts at Guohai Securities said in an April 21 report. The firm identified the sector as a core track with strong certainty in the current AI industry chain.
The data supports a clear trend of coordinated price hikes. The one-year lease price for an H100 GPU surged from a low of $1.70 per hour in October 2025 to $2.35 in March 2026. In that same period, major Chinese cloud providers including Tencent Cloud, Alibaba Cloud, and Baidu Cloud announced price increases ranging from 5% to over 400% for various AI computing services.
This price surge directly benefits the revenues of computing power providers but will increase operational costs for the AI model developers who are driving the demand. The trend could accelerate market consolidation as smaller players struggle to absorb the higher costs, while well-capitalized lessors are poised for significant long-term profit growth after their initial hardware investments fully depreciate.
Inference Boom and Overseas Growth Drive Demand
The key driver for the price hikes is a fundamental shift in AI workloads. While the past two years were dominated by the computing-intensive process of training large models, the industry is now moving into a new phase focused on inference—the process of using trained models to generate text, images, or predictions. According to Nvidia CEO Jensen Huang, this inference phase is far more extensive, with computing demand growing 10,000-fold in the last two years.
Data from Chinese companies shows explosive growth. The Doubao large model from ByteDance saw its daily token volume double in three months to surpass 120 trillion by March 2026, a 1,000-fold increase since its launch. This surge in token demand, driven by high-frequency applications like AI agents and multimodal generation, is creating a positive feedback loop for domestic models.
Chinese AI companies are also finding significant success overseas. OpenRouter data shows that for six consecutive weeks in early 2026, the top AI models with the highest global call volume were all from China. For Kimi, overseas revenue had already surpassed domestic revenue by February 2026, while about 73% of MiniMax's revenue comes from international markets.
Lessors Raise Billions to Expand Capacity
In response to the booming demand, Chinese computing power providers are aggressively raising capital to expand their infrastructure. These companies operate on a model of heavy upfront investment in data centers and servers, followed by stable, long-term rental income. Contracts are typically 2-5 years long, providing predictable cash flow.
Several firms have announced major financing and expansion plans in 2026.
- Xccelerated Data (协创数据), which has applied for a total of 50 billion yuan in credit lines, is also pursuing a listing on the Hong Kong Stock Exchange.
- HongJing Technology (宏景科技) is seeking 60 billion yuan in credit and plans to raise an additional 1.29 billion yuan, with the majority earmarked for building out its intelligent computing clusters.
- Other players like Shengshi Technology (盛视科技) and Zhiwei Intelligence (智微智能) have also secured credit lines of 23 billion yuan and 14 billion yuan, respectively.
According to the Guohai report, companies with stable procurement channels and significant financing can rapidly deploy new computing capacity, locking in long-term contracts that offer strong visibility for future revenue and profit growth.
This article is for informational purposes only and does not constitute investment advice.
