Tokens become a business metric

Tokens are becoming the unit economics of AI.

We usually discuss AI in terms of models, applications, and GPUs.

In production, the operating meter is increasingly the token.

OpenRouter says its platform processes more than 10 trillion tokens per day across 400-plus models for over 10 million developers and companies.

A recent study using OpenRouter data found agentic requests now represent more than half of total token consumption.

Humans are becoming the minority in token usage.

AI agents consume tokens continuously as they reason, retrieve context, call tools, evaluate results, and coordinate with other agents. One human request might trigger thousands of automated model interactions.

Deloitte found 61% of surveyed enterprise leaders expect their organizations to consume more than 10 billion tokens per month by 2028.

Vendors are already responding:

• Stripe agreed to acquire OpenRouter, combining economic infrastructure with model routing and token optimization
• Snowflake introduced dynamic model routing, usage visibility, quotas, and spending limits
• OpenAI and Anthropic discount cached inputs, rewarding applications designed to reuse context efficiently
• AI gateways route requests according to quality, cost, latency, and complexity

My takeawaytoken consumption is not a measure of AI success.

You need to track:

• Cost per completed workflow
• Tokens per successful outcome
• Human versus agent-generated consumption
• Cache utilization
• Frontier-model usage
• Quality and latency by route

Enterprise AI needs more than adoption dashboards.

You need unit economics for both human and autonomous consumption.

Are you prepared to manage an AI environment where agents consume more tokens than people?