Estimating Kimi K3's energy and water consumption per million tokens
Understand this story
Based on inference data for Kimi K3 on an 8-GPU GB300 server, an author estimates that processing 1 million tokens consumes about 0.14 kWh of electricity and 0.35 liters of water, and emits 0.05 kg of CO2. These figures are an estimate of the environmental cost corresponding to personal AI usage, not a measured general conclusion; they apply to Kimi K3 inference under that server configuration.
AI overview based on source reports · Updated 51 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIWhat's the environmental impact of my AI usage?
The author estimates the marginal energy, water and CO2 impact of running Kimi K3 on an 8-GPU server, arriving at roughly 0.14 kWh, 0.35 L of water and 0.05 kg CO2 per million tokens, then assumes impact scales with cost to estimate per-dollar figures. Applying these to about $5,871 of coding-agent spend over 6.3 months gives roughly 1,200 kWh, 3,000 L of water and 410 kg CO2, which is small compared with the author's household energy, irrigation water and flight emissions. The post warns that flat-rate subscriptions, which hide underlying token costs, could produce much larger impacts.
Activity for this story
Not enough continuous observations to show a trend yet.