A new research report highlights that AI computing power has fully entered a new cycle dominated by inference, with explosive growth in high-token-density application scenarios driving a surge in demand. On the supply side, the market is showing a clear divergence: high-end computing power remains persistently scarce, while mid-to-low-end capacity is moving toward structural saturation. Geopolitical restrictions and rising costs across the entire supply chain are further elevating the barriers to entry for premium computing resources.
The computing power leasing sector is accelerating its growth trajectory, experiencing simultaneous increases in both volume and pricing, and is now entering a golden period of hundred-billion-scale prosperity. Market size is projected to exceed 260 billion yuan by 2026, as the industry transitions from broad, unrefined expansion to a phase defined by premium pricing and meticulous operational management. The Matthew effect is driving resources to concentrate rapidly among leading players.
Meanwhile, the business model is evolving from bare-metal leasing toward integrated Token-based revenue sharing and Token factory services. Combined with long-term contract locking and financial leverage support, this transformation is reshaping industry certainty and broadening the growth curve for market participants.
Where opportunities lie: Supply-demand gap and structural divergence coexist, with demand shifting from 'training inventory' to 'continuous computing consumption'
On the demand front, application growth is pushing Token consumption to new heights. Agent workflows and code-related tasks now dominate Token usage, driving AI applications to upgrade from single-round interactions to complex execution scenarios with high Token density. This shift is moving AI computing demand from a training-driven model into a resilient new growth cycle led by inference workloads.
On the supply side, the computing market exhibits a structural polarization: high-end capacity remains in persistent shortage while mid-to-low-end resources face oversupply. Tightened geopolitical controls, rising costs across the semiconductor supply chain, and delays in advanced packaging capacity expansion are collectively widening the gap between intelligent computing supply and demand. At the same time, older generations of mid-to-low-end general-purpose computing capacity are under dual pressure from idle capacity and declining prices.
The growth trajectory: Computing power leasing accelerates with rising volume and pricing, entering a hundred-billion-scale high-prosperity phase
As of the end of June, China's total intelligent computing power had grown strongly to 2,185 EFLOPS, with the national average rack utilization rate jumping to 71.4%. The Chinese intelligent computing leasing market is demonstrating high-growth characteristics, with market size expected to surpass 260 billion yuan by 2026, according to data from the China Academy of Information and Communications Technology.
Overall, the computing power leasing industry is transitioning from the late growth stage into the early maturity phase. It is shifting from the unrefined expansion of 'claiming territory' to a structural move toward premium pricing and refined operations. The competitive landscape is also evolving from a fragmented market into a phase where resources concentrate among top players, following the Matthew effect.
What's changing: Deep business model evolution with long-term contracts, service-oriented transformation, and financing innovation reshaping industry certainty
In terms of service models, the industry is moving from bare-metal leasing toward new value-sharing models represented by Token-based revenue sharing and integrated Token factory services, accelerating the opening of a second growth curve. In contract structures, long-term agreements, duration segmentation, and TaaS flexible pricing are breaking the cyclical bottlenecks of capital-intensive assets. On the financing side, computing infrastructure construction is entering a phase of financialization and high leverage, where capital access and supply chain innovation have become core competitive barriers.
Key risks to monitor
The report also outlines several key risks: (1) the possibility that AI application deployment and computing demand fall short of expectations; (2) rapid chip technology iteration leading to significant asset impairment; (3) high-leverage financing and debt maturity mismatches; (4) delays in power supply and data center delivery; and (5) geopolitical tensions and supply chain restrictions.
Comments