Usage-Based Pricing
Pay for what you consume
The Pattern
Usage-Based Pricing charges customers based on actual consumption: compute hours, API calls, data processed, tokens generated, messages sent. Revenue scales directly with customer success — when your customer grows, your revenue grows automatically.
AWS pioneered this at massive scale, charging per EC2 instance-hour and per GB of S3 storage. Snowflake perfected it with “compute credits” that separate storage from compute. OpenAI brought it to AI with per-token pricing. The model now dominates cloud infrastructure, communications, and AI services.
The economics are compelling: Snowflake’s net dollar retention exceeds 130%, meaning existing customers spend 30%+ more each year without any sales effort. This “expand” motion is built into the pricing model itself — as customers process more data or make more API calls, revenue grows organically.
Key Metrics & Benchmarks
Who Uses This Pattern
Strengths & Weaknesses
STRENGTHS
- Revenue scales automatically with customer success
- Perfect alignment of incentives — customers pay for value received
- Low barrier to entry — start small and grow
- Natural expansion revenue without sales effort
WEAKNESSES
- Revenue volatility — consumption drops in downturns
- Harder to forecast than subscription revenue
- Customer optimization can compress spend
- Requires sophisticated metering and billing infrastructure
How AI Is Transforming This Pattern
AI has made usage-based pricing the default for the most consequential new technology category. OpenAI charges $2.50-$15 per million tokens. Anthropic charges similarly. Google Gemini offers per-token pricing. This creates a direct, linear relationship between AI adoption and revenue.
The strategic implication: as AI usage grows exponentially across every industry, usage-based AI companies sit on exponential revenue curves. The constraint shifts from demand to supply — can you provision enough GPU capacity to serve demand? Companies like CoreWeave and Together AI are building entire businesses around providing the infrastructure for usage-based AI.
Business Engineer Insight
Usage-based pricing is the natural business model for AI because AI consumption is inherently variable and unpredictable. A customer might send 1,000 API calls one day and 1,000,000 the next. Per-seat pricing can’t capture this variability; usage-based pricing captures it perfectly. The companies that master real-time metering, cost management, and dynamic pricing for AI workloads will control the economic layer of the AI stack.
Related Patterns
Understand the strategic architecture behind this business model pattern — and how the best companies deploy it for competitive advantage.