AI Model API-as-a-Service: Token Economics and Pay-as-you-go
1) API calls: Billing based on usage volume (Tokens); 2) Prepaid Tokens: Revenue collected in advance through recharge p
Key Fields
FIELD STAMPS📌 Background
In 2026, the revenue structure of major AI model companies underwent a collective shift. Zhipu AI reported H1 revenue of 954 million RMB, a year-on-year increase of 399.7%, with open platform and API revenue reaching 825 million RMB, up 2735.7% year-on-year, accounting for 86.5% of total revenue compared to 15.2% in the same period last year (based on the company's semi-annual report). DeepSeek's API gross margin reached 82.9% (based on performance disclosures). Token-based billing has transformed model invocation into a tradable commodity, making pay-as-you-go the primary channel for CaaS commercialization.
👤 Target Customers
AI developers, enterprise SaaS platforms, AI startups, and other businesses paying for model usage.
💰 Revenue Streams
1) API calls: Billing based on usage volume (Tokens); 2) Prepaid Tokens: Revenue collected in advance through recharge packages; 3) Value-added services: Custom models, dedicated SLAs, and technical support charged by project or annual fee; 4) Industry solutions: Custom development and deployment fees charged by project (opportunistic, with no specific figures available yet).
🧮 Cost Structure
Cloud computing rental fees, model R&D and iteration investment, data labeling and cleaning costs, platform maintenance and security expenses, and labor costs for professional customer service and technical support.
🛡️ Moat
Mature large-scale model technology and massive training data; a complete Token economic loop that locks in active users; standardized, multi-language SDKs and documentation that lower entry barriers; ecosystem partnerships and industry solutions.
🔑 Keys to Success
- Build a tradable Token economic loop
- Provide unified, standardized model APIs and multi-language SDKs
- Develop a high-availability, low-latency computing delivery platform
⚠️ Risks
- Profit compression due to soaring computing costs
- Competitors launching lower-priced or open-source alternatives
- Regulatory restrictions on model API usage and data security
🏢 Cases
- Zhipu AI API (API revenue share increased from 15% to 87%)
- DeepSeek API (Revenue increased 10x in the first seven months, with an API gross margin of 82.9%)
📊 SWOT Analysis
Strengths
- High API gross margins, with both Zhipu AI and DeepSeek achieving over 80%.
Weaknesses
- High reliance on computing power, with costs fluctuating alongside cloud pricing.
Opportunities
- Rapidly growing demand for plug-and-play AI capabilities among enterprises, with significant cross-industry penetration opportunities.
Threats
- Rising commission rates from cloud providers and tightening regulatory requirements for data compliance.