GPU Computing Power and IDC Resale/Leasing

📌 Background

With the explosion of AI inference demand in 2026, the computing power leasing model has shifted from selling hardware to selling tokens and revenue sharing based on usage. The semi-annual report shows that Litong Electronic's revenue from computing power and related services reached 1.274 billion yuan, a year-on-year increase of 162.38%, accounting for 60.63% of total revenue (based on the company's semi-annual report). Xiechuang Data's revenue from intelligent computing products and services reached 3.216 billion yuan, a year-on-year increase of 163.46%, with its share of total revenue jumping from 2.04% to 25.68%. Rack utilization rates and gross margins for top-tier players are significantly higher than those of small and medium-sized intelligent computing centers.

👤 Target Customers

AI startups, large model fine-tuning teams, university laboratories, and SMEs requiring elastic GPU resources.

💰 Revenue Streams

1) On-demand computing rental: Charging AI teams and universities for computing services based on GPU hours or task cycles; 2) Monthly leasing: Charging fixed monthly service fees for basic GPU resources; 3) Second-hand GPU subleasing: Profiting from the rental price spread by subleasing recovered second-hand GPUs; 4) Computing operations, maintenance, and cluster optimization: An opportunistic revenue stream, with the specific contribution yet to be disclosed.

🧮 Cost Structure

GPU server procurement or financial leasing costs, data center electricity and bandwidth expenses, maintenance labor and facility depreciation, and resource procurement costs from upstream cloud providers or carriers.

🛡️ Moat

High capital barriers; top-tier players possess proprietary computing power and customer lock-in capabilities, while some companies leverage green energy or regional policies to establish cost advantages.

🔑 Keys to Success

• Secure high-quality enterprise clients to maintain high rack utilization rates
• Reduce upfront capital pressure through financial leasing or subleasing models
• Gradually transition to token-based usage sharing to increase revenue per card

⚠️ Risks

• Oversupply of computing power may lead to a decline in rental prices
• Rapid GPU obsolescence poses high risks for hardware residual value

🏢 Cases

• Litong Electronic's computing service revenue reached 1.274 billion yuan, up 162% year-on-year
• Xiechuang Data's intelligent computing product and service revenue reached 3.216 billion yuan
• Rongze Technology's AIDC business revenue reached 1.995 billion yuan

📊 SWOT Analysis