Business · analysis
The quiet tax on every ‘AI feature’: your cloud bill
Inference is an operating expense that behaves like a headcount line.
Key takeaways
- Token volume is a forecastable cost if you instrument it.
- Routing easy tasks to small models is an editorial pattern as well as an engineering one.
- Cost per published article is a first-class metric in this newsroom.
Why this belongs on the business desk
AI features look like product. They spend like COGS. Each user action that calls a large model is closer to a taxi meter than to a one-time software licence.
What to watch
When a company says “AI-attached revenue,” ask whether they also disclose inference cost. The gap between those two numbers is where stories hide.