Model-as-a-Service
Foundation model access via API
The Pattern
Model-as-a-Service provides access to AI foundation models via API, charging per token, per request, or via subscription tiers. This is the defining business model of the AI era. OpenAI crossed $5B+ ARR selling access to GPT-4. The provider handles training ($100M+ per frontier model), infrastructure, and scaling — customers simply send requests and receive intelligence.
This model is in its “AWS 2008” era — the market leader has first-mover advantage, but the market is expanding so rapidly that multiple winners can coexist.
Key Metrics & Benchmarks
Who Uses This Pattern
Strengths & Weaknesses
STRENGTHS
WEAKNESSES
- Potential race-to-bottom on pricing as models commoditize
- Massive capex required for training and inference infrastructure
- Regulatory uncertainty around AI liability and safety
- Customer concentration risk with large enterprise deployments
How AI Is Transforming This Pattern
This IS the AI-native business model. The key question: will Model-as-a-Service commoditize (race to zero on pricing) or maintain differentiation? Current evidence suggests specialization will win — different models excel at different tasks. The market is bifurcating: general-purpose frontier models (OpenAI, Anthropic, Google) vs. specialized models (Harvey for legal, Bloomberg for finance).
Business Engineer Insight
Model-as-a-Service faces the classic infrastructure question: will value accrue to the model layer or the application layer? History (cloud computing) suggests infrastructure captures significant value. But unlike cloud — where AWS/Azure/GCP provide commodity compute — AI models have genuine differentiation in quality, safety, and capability. This suggests MaaS margins will remain higher than cloud compute margins.
Related Patterns
Understand the strategic architecture behind this business model pattern — and how the best companies deploy it for competitive advantage.