Cost per token
How much it costs to produce each chunk of AI output.
Ranks high (67/100) — set apart by structural importance (82) and cross-layer reach (86).
Why it matters
It is the economic translation of the whole stack into application feasibility.
Why now
Margins on AI products hinge on driving cost per token down faster than usage grows.
If Cost per token runs short
Stalled cost-per-token declines cap which applications can be profitable.
In depth · editorial + model
Cost per token is how much it costs to produce each chunk of AI output — the unit price of running a model. It is the economic translation of the whole stack: every upstream constraint, from power to memory to serving efficiency, eventually surfaces here as a figure attached to each generated token. It matters because the margins on AI products live or die on driving this cost down faster than usage grows. Physical constraints in the chain do not disappear — they reappear as unit cost.
If cost-per-token declines stall, the effect is to cap which applications can be profitable at all; use cases that need cheap, high-volume generation simply become uneconomic. It is the stack's output metric — the place where all the engineering upstream cashes out. The names most directly working this lever are the model owners and efficiency specialists such as OpenAI, Anthropic, DeepSeek, and Together AI, who compete partly on how cheaply they can serve.
Who makes Cost per token
The companies exposed to Cost per token
How to think about it
- Cost is the stack’s output metric
- Physical constraints surface as unit cost
What to watch
- Published cost-per-token trends
- Drivers: power, utilization, efficiency
Key figures
Frequently asked
What is Cost per token?
How much it costs to produce each chunk of AI output.
Why does Cost per token matter for AI?
It is the economic translation of the whole stack into application feasibility.
Who makes Cost per token?
The companies the model tags as producers or suppliers of Cost per token: DeepSeek.
Which companies are most exposed to Cost per token?
OpenAI, Anthropic, DeepSeek, Together AI — 4 companies in total are mapped to Cost per token.
What happens if Cost per token runs short?
Stalled cost-per-token declines cap which applications can be profitable.
Where does Cost per token sit in the AI value chain?
Cost per token sits in the Models layer of the AI value chain.
Go deeper on Cost per token
- The materials, geographies and policies it depends on — heat-mapped
- Substitutes, relief valves and the domino chains if it tightens
- The live tension score, momentum and news drivers
- Four levels of analysis — from plain-English to strategic
model v0.7.0 · research, not advice
Model scores are illustrative reads from our model of the AI value chain — not investment advice.