Gunjo · Business Intelligence for the AI Era
← Sticker Wall MODEL · DETAIL

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

MODEL

Key Fields

FIELD STAMPS
IndustryAI / LLM
RegionGlobal
ScaleMid-size
ChannelOnline

📌 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.