Token-based Computing Power Reselling

📌 Background

In 2026, the computing power leasing market shifted from selling GPU hours to selling Tokens to lower the barrier to entry for customers and improve asset turnover. Calculations by Fenghe Intelligence indicate that the effective output of most intelligent computing centers is only 20% to 35% of their theoretical capacity. By increasing the effective output rate from 30% to 65%, a cluster of 1,000 B200 GPUs can be equivalent to an additional 1,167 GPUs. Through nine-layer architecture optimization, inference throughput increases by over 200% and time-to-first-token (TTFT) latency decreases by approximately 80%; this efficiency gap is the source of the resale gross margin.

👤 Target Customers

AI application developers, SME algorithm teams, and research institutions

💰 Revenue Streams

1) Token-based billing: Charging customers based on Token consumption, reselling upstream GPU cluster computing power at a premium; 2) Volume-based agreements: Signing long-term volume commitments with customers to lock in cash flow through periodic settlements; 3) Idle capacity leasing: Renting out idle computing power at dynamic prices; 4) Scheduling system licensing: Licensing proprietary computing power scheduling and Token billing systems to third parties (an opportunistic revenue stream, not yet validated).

🧮 Cost Structure

GPU cluster leasing or procurement costs, data center electricity and bandwidth expenses, operations team labor, and R&D for the computing power scheduling platform

🛡️ Moat

Bargaining power for bulk procurement from upstream IDCs or NVIDIA channels, proprietary computing power scheduling and Token billing systems, and customer stickiness through long-term volume agreements

🔑 Keys to Success

• Securing stable and low-cost GPU cluster resources
• Developing proprietary, high-efficiency computing power scheduling and Token billing systems
• Locking in enterprise customers with long-term volume commitments

⚠️ Risks

• Intense competition in the computing power leasing market leading to thin margins for intermediaries
• High volatility in upstream GPU pricing and supply

🏢 Cases

• Xunce Technology TokenCloud
• Xingyun Technology Computing Power Leasing Agreement
• Fenghe Intelligence Nine-Layer Architecture Token Billing

📊 SWOT Analysis