Towards Machine Economies: What Machines Consume and How That Consumption Will Reshape the World

Defining the terms

A machine economy is an economic system where machines are the primary actors in resource allocation decisions - determining what is exchanged, when, with whom, and at what price. The ultimate objectives may be human-defined, but the decisions are not.

This is distinct from a machine augmented economy, where humans remain the decision-makers and machines serve as tools for execution. Most of what we call "automation" falls into this category.

The distinction matters because the dynamics are different. In augmented economies, machines amplify human intent. In machine economies, emergent behavior arises from machine-to-machine interaction - outcomes that no individual participant explicitly programmed.

Measurement

Machines consume energy and information. The computation is the economic activity - there are fewer layers of abstraction between physical inputs and economic outputs. This makes machine economies more directly measurable.

A concrete example: the AI accelerator rack

Consider a B200 server rack. The value chain:

• Capital cost: ~$2M per rack
• Operational costs: electricity, cooling, housing, security
• Rental revenue: priced to cover costs plus margin
• End-user value: the economic activity enabled by inference

AI compute offers unique ground for measurement. Unlike most economic activity where value is abstract and layered, inference has a traceable chain from physical inputs to economic outputs. The watts become FLOPS, the FLOPS become tokens, the tokens become services or decisions, the services become revenue.

Each step has a price signal. We can ask: what is the total economic value generated by a rack over its lifetime? Not just the rental revenue, but the downstream value created by the inference it provides.

More to come...

Home