In August 2026, the global AI computing power market is undergoing a profound structural change. As the demand for large model training gradually saturates, the demand for inference computing power becomes the core driver of market growth, and the supply shortage of NVIDIA H200 chips is pushing AI computing power leasing prices up by 25%, triggering a chain reaction in the upstream and downstream industries. This change not only reshapes the supply and demand structure of the computing power market but also brings new layout opportunities for investors.
## I. AI Computing Power Shifts from "Training King" to "Inference King"
Over the past two years, the focus of the AI computing power market has been on training computing power. Large model training, represented by models like ChatGPT and GPT-4, requires massive computing power support, leading to insufficient supply of training chips (such as A100, H100) and continuously rising leasing prices. However, since 2026, as the pace of large model iteration slows down, the implementation of enterprise AI applications has become a new growth point, and the demand for inference computing power is starting to explode.
According to the latest forecast from IDC, the global inference server market size will grow by 45% year-over-year in 2026, reaching $120 billion, while the training server market growth is only 15%. The core reason for this shift is the enterprise demand for AI application implementation: from intelligent customer service and image recognition to autonomous driving and drug discovery, inference computing power is becoming the key infrastructure for enterprise digital transformation.
"Inference computing power is the last mile for AI to move from the laboratory to commercialization," stated the CEO of a leading AI leasing company. "In the past, enterprises focused more on model accuracy; now they focus more on application efficiency and cost, and the demand for inference computing power naturally rises."
## II. H200 Supply Shortage Causes Leasing Prices to Surge by 25%
Against the backdrop of exploding inference demand, the supply shortage of NVIDIA H200 chips has become the key factor driving up leasing prices. The H200 is NVIDIA's dedicated inference chip launched in 2023, offering a 30% improvement in inference performance and a 20% reduction in power consumption compared to the H100, making it more suitable for enterprise applications.
However, due to adjustments in the H200 production line and capacity ramp-up issues, its supply volume is far below market demand. According to supply chain news, the H200 capacity can only meet 60% of global demand, forcing leasing companies to increase prices to balance supply and demand.
"The supply shortage for H200 is more severe than for H100," pointed out a chip analyst. "H100 is mainly used for training, while H200 is for inference. Inference demand is more dispersed, covering more small and medium enterprises, so the impact of the supply shortage is greater."
Data shows that in August 2026, the leasing price for H200 chips increased by 25% from the beginning of the year, reaching $0.8 per hour, while the leasing price for H100 only increased by 10%. This gap reflects the tightness of the inference computing power market.
## III. Inference Demand Takes Over: The Inevitable Result of Enterprise AI Application Implementation
The shift in inference demand taking over training demand is essentially the result of AI technology transitioning from "technology-driven" to "application-driven." In the past, the development of AI mainly relied on model training by research institutions, but now enterprises are beginning to apply AI to actual business scenarios, requiring a large amount of inference computing power support.
For example, a major e-commerce giant stated that its intelligent recommendation system needs to process billions of inference requests every day, while an autonomous driving company needs to process real-time sensor data, all requiring high-efficiency inference computing power. As enterprise AI applications become more widespread, the demand for inference computing power will continue to grow. The demand for computing power is inelastic. An operator of a data center stated, "Enterprises will not stop using AI because chip prices rise; instead, they will seek more efficient solutions, such as optimizing models or adopting inference-specific chips, which further drives demand for H200."
## IV. Industry Chain Opportunities: Leasing Companies, Chip Suppliers, and Data Center Operators
The soaring prices of AI computing power leasing present new opportunities for enterprises in the industry chain. First are leasing companies, such as CoreWeave and Lambda Labs, which gain higher profit margins by providing H200 leasing services. CoreWeave's Q2 2026 earnings report showed that its inference computing revenue proportion reached 40%, a significant increase from 20% in the same period last year.
Second are chip suppliers. NVIDIA, as the manufacturer of H200, benefits from price increases and growing demand. Its Q2 2026 earnings report showed that inference chip revenue grew by 35% year-over-year, becoming the main driver of the company's growth.
Finally are data center operators, which meet enterprise demand by building inference-specific data centers. For example, an operator in North America built multiple inference data centers using H200 chips, attracting a large number of enterprise customers.
## V. Investment Logic: Focusing on Leading Stocks in the Inference Computing Chain
For investors, the rise in AI computing power leasing prices and the explosion in inference demand mean that attention should be paid to leading stocks in the inference computing chain. First are NVIDIA, as the chip leader; the supply shortage of H200 will continue to drive its earnings growth. Second are leasing companies, such as CoreWeave, whose increasing inference computing revenue proportion will improve profit margins. Finally are data center operators, such as Digital Realty, whose growing demand for inference data centers will drive rental income.
"Investors should focus on companies that can seize inference computing opportunities," stated a fund manager. "These companies benefit not only from price increases but also from growing market demand, possessing long-term investment value."
## VI. Risk Warning: Supply Shortages and Intensified Competition
Although the prospects for the inference computing market are broad, risks must be noted. First, the potential supply shortage of H200 may continue, leading to further increases in leasing prices and affecting enterprise costs. Second, intensified competition, with more enterprises entering the leasing market, could lead to price wars. Finally, technological iteration, such as the launch of new inference chips (like AMD MI400X), could divert demand for H200.
"Investors need to monitor changes in supply shortages and shifts in competitive landscape," pointed out a securities analyst. "If H200 capacity increases, leasing prices may fall, and new chips entering the market could change the market structure."
In summary, the soaring prices of AI computing power leasing and the explosion in inference demand mark a new phase for the AI computing power market. Investors who seize this trend and position themselves in the leading stocks of the inference computing chain will gain long-term investment returns.
Disclaimer
The content of this article is for market analysis, investment knowledge popularization, and personal opinion sharing only, and does not constitute any investment advice. Stock market investment involves risks, please invest cautiously.
