Ask HN: Carbon Sequestration Recommendations
We are GameTorch, a pre-launch startup that lets you generate 2d video game assets and publish them to the Creative Commons.
We use image models that were energy intensive to train. We care about the earth and have made a promise to be carbon positive at the inference level. It would be extremely ideal if the sum of all AI driven companies were so carbon positive that we also outweigh the costs of training these models.
We need to make good on our promise. Does HN have any recommendations for carbon sequestration companies?
Separately, any feedback or advice on the following analysis of our energy usage and carbon emissions would be greatly appreciated. We really truly care about making a positive impact on the Earth and are eager to hear about any flaws in our thought process. It's not about PR, it's about doing things the right way. Here's our current analysis:
Energy Use, CO₂ Emissions, and Cost per Image Generation
Approximate energy use per image generation: ~0.5 Wh (watt-hours)
CO₂ emissions per image (US grid average): ~0.00008 kg (0.08 grams)
Total for 1,000 images: 0.5 kWh, 0.08 kg CO₂
Context: A typical US domestic flight (e.g., NYC to Chicago) emits about 300 kg CO₂ per passenger—over 3 million times more than a single image generation.
Context: Boiling water for one pour-over coffee (about 12 oz/350 ml) uses ~50 Wh and emits ~0.02 kg (20 grams) CO₂—about 100× more energy and 250× more CO₂ than generating one image.
Cost to sequester CO₂ from 1,000 image generations: At scale, permanent carbon removal is about $100/tonne. Sequestering 0.08 kg (from 1,000 images) costs about $0.008 USD.