Interactive estimate
The annual value of one megawatt
Compare GPU ownership, rental, and token revenue with editable assumptions and token prices. All model data is placeholder data.
How to use: Select a model, then edit the hourly GPU costs or the three token prices. The revenue breakdown and comparison bars update automatically.
Input throughput385.7M tok/hr
Output throughput2.3M tok/hr
Benchmark GPUs8
Input cache96%
Scale to 504 GPUs63.0×
Tokens per year214.1T
Token revenue per year$105.6M
Selected model
Kimi-K3 15 tok/s revenue breakdown
Uncached input8.5T tokens/year
Cached input204.3T tokens/year
Output1.3T tokens/year
One megawatt · one year
Annual value comparison
Annual tokens = tokens/hour × 8,760 × ((72 GPUs/rack × 7 racks) ÷ benchmark GPUs). Cached input = annual input × cache percentage.