Energy in.Intelligence out.
Every AI token starts as fuel. Token Outbound measures what it takes to make intelligence, and which countries and power sources make it best.
Tokens can come from anywhere.
Wherever there is clean, affordable power, there can be a token factory. Token Outbound connects the places that make sustainable compute with the people and companies who need it.
A token is work, made from energy.
Strip away the hype and an AI token is thermodynamics. Fuel becomes heat, heat becomes electricity, electricity becomes computation, and computation becomes useful intelligence. Every step has a cost and a loss. We measure them.
Fuel
Gas, coal, uranium, water, wind or sunlight.
Power plant
Conversion efficiency sets the first and largest loss.
Grid
Transmission losses, reliability and price.
Data center
Cooling and overhead, measured as PUE.
Chip
Matrix math and statistics, billions of times a second.
Token
Useful intelligence, ready to export.
Inside the chip, it's matrix math.
A language model is a vast table of numbers. Every token comes from multiplying those numbers together and turning the result into probabilities. That arithmetic is where the electricity goes.
Training
The model learns by running trillions of matrix multiplications over enormous amounts of text, adjusting its numbers until its predictions are good. A large energy investment, made up front.
Inference
Every time the model answers, it runs those same numbers forward to predict the most likely next token, one token at a time. A small cost per token, repeated billions of times a day.
Not all tokens are made equal.
The same model, on the same chip, carries a very different footprint depending on what powers it. Choose a fuel to compare its lifecycle carbon per kilowatt-hour.
Lifecycle median estimates, IPCC Fifth Assessment Report (2014), Annex III. Grid averages vary by country.
Three numbers that explain AI.
Published token prices hide the physics. These measures put every supplier, country and fuel on the same footing.
Tokens per kilowatt-hour
How much intelligence a unit of energy produces. The core efficiency measure.
Tokens produced divided by energy consumed, including cooling.
Cost per million tokens
What intelligence actually costs to make, regardless of the price on the rate card.
Compute cost per hour divided by tokens per hour, times one million.
Carbon per million tokens
The climate footprint of intelligence, traced back to the power plant.
Grid carbon intensity times energy per million tokens.
Where the world's tokens come from.
Nobody publishes how many tokens a country produces or exports. Our first study builds transparent estimates for five countries, each powered by a different fuel, with every assumption shown.
United Arab Emirates
Gas wealth and capital aimed at becoming a regional AI hub.
Malaysia
The Johor data center boom, built largely to serve customers abroad.
Norway
Cheap, nearly carbon-free power and a cool climate. The clean benchmark.
France
A mostly nuclear grid and an existing electricity exporter.
United States
The largest producer and consumer of AI compute. The reference point.
The study is in progress. Findings will be published here and on our newsletter, country by country.
Work with us.
Two fixed-scope studies, one for each side of the market. Both are grounded in engineering, not marketing claims.
Compute and Power Cost Assessment
Find the lowest real cost of useful AI output for your workloads, and the energy and carbon behind it.
- Benchmark of three or more suppliers
- Cost per million tokens, normalized
- Energy and carbon per million tokens
- Supplier introductions where they fit
Token Production Feasibility
Find out what your power and land are worth when megawatts are turned into tokens instead of sold as electricity.
- Power supply and interconnection review
- Estimated token output per megawatt
- Cost and carbon position against global peers
- Paths to market and operating partners
Built by an engineer, not a marketer.
Token Outbound was founded by Michael, a licensed Professional Engineer with 30 years of experience in civil engineering, energy, data center and manufacturing facilities, and river and environmental policy.
The idea is simple. Power has always been the input that matters most. Now it makes intelligence too, and the countries and companies that understand that will lead.
Let's talk about your tokens.
Buying compute, producing power, or researching AI and energy? Send a note and we'll reply within two business days.