The Economics of Machine Intelligence
WATTTOTOKEN RESEARCH
The Economics of Machine Intelligence
Independent research into how energy, silicon, compute, and algorithms become intelligent systems and economic value.
Explore the research agenda Our research standards →
An independent research initiative in development.
Energy · Silicon · Infrastructure
Compute · Data · Models
Training · Inference
Intelligence · Applications · Value
THE CENTRAL QUESTION
What does it take to turn computation into useful intelligence?
We study how physical resources, engineering choices, and capital shape AI systems—and how efficiently those systems translate into measurable economic value.
Six lenses. One connected system.
A / INFRASTRUCTURE
The physical cost of compute
Electricity, accelerators, facilities, financing, and the constraints on computational capacity.
B / TRAINING
Building model capability
Scaling, data, architecture, utilization, and the economics of model development.
C / INFERENCE
Serving useful intelligence
Throughput, latency, memory, energy, and cost per request and successful task.
D / BUSINESS
Following the capital
Pricing, margins, depreciation, competition, and returns on infrastructure investment.
E / INTELLIGENCE
Measuring economic value
Reliability, human oversight, automation, and the relationship between capability and productivity.
F / MEASUREMENT
Making claims testable
Transparent assumptions, comparable benchmarks, uncertainty, and reproducible methods.
RESEARCH IN DEVELOPMENT
A foundation for rigorous inquiry
Our initial program will connect infrastructure costs, inference engineering, and useful economic outcomes. Research releases will follow evidence review and reproducibility checks.
No research publications released yet.
Planned flagship working paper: From Watt to Token: The Economics of Machine Intelligence.