TECV proposes pricing each GPU-hour from physical facts instead of markets: how well the machine sheds heat, what local electricity costs and how dirty it is, and how much slower a job runs if it lands on a machine with the wrong memory profile. Combine the three into one score, turn the score into a dollar price per hour, and use it to decide where agent workloads should run.
The proposal calls the score a "weighted product" but gives no weights, units or conversion to dollars. This calculator fills those gaps with stated assumptions so you can see how the idea behaves. The machines and figures below are illustrative, not quotes.
Price calculator
Example data · illustrative nodes
Node
Rent $/hr
Heat eff. (0–1)
Power $/kWh
gCO₂/kWh
Latency penalty
TECV score
TECV $/hr
vs rent
Elasticity: price change from a +10% move in one factor
Assumptions this calculator adds
Each factor is normalized 0–1 where 1 is best, and the score is the weighted geometric mean.
TECV $/hr = energy cost (kW × $/kWh + kW × carbon price) ÷ score, where the numerator also includes an assumed hardware cost per hour for each node ($2.50, $1.20, $0.30, $0.45).
Heat efficiency replaces the proposal's "Celsius-degree-hours per watt", which isn't a standard unit.
Latency penalty is the slowdown versus the best-matched node for an example LLM inference job.
Claimed by the proposal, not verified
A "sovereign compute-pricing layer" for enterprises. Nobody sells or buys at TECV prices today, so it's a valuation model, not a price.
That it improves on "existing real-time inference elasticity indices". None are named, so there is nothing to compare against.
That renters can measure thermal efficiency per GPU. Cloud and VPS customers usually can't see chassis temperatures or grid mix at the rack level.
What to try
Set the carbon price to $200. The coal-heavy node's TECV price climbs further above its rent, while the solar-backed node barely moves.
Put all the weight on latency penalty. The node with the right memory profile wins regardless of power cost.
Compare the "vs rent" column. Where TECV and rent disagree is where the framework claims to find value; for a small fleet on fixed monthly VPS plans, the rent is the price you actually pay.