Five essays on the hidden carbon cost of agent infrastructure. Compute carbon varies 10x by region. Renewable availability shifts hourly. Scope 3 dominates operational emissions. Temporal and geographic arbitrage changes the economics. Most model operators are not pricing any of this yet.

The tool behind them is written up as a project. It is offline now; the essays stand on its April 2026 data.

  1. The Same Model Costs 10x More Carbon in Virginia Than the Netherlands The core finding: identical inference, wildly different carbon. What the data shows across 85 models, 6 providers, 9 regions.

  2. Scaling in the Wrong Direction LeCun’s architecture argument meets carbon data. What happens when a more efficient architecture runs on a clean grid.

  3. Grid Arbitrage: Why Region Choice Matters More Than Model Choice A bigger model on a clean grid beats a smaller model on a dirty grid. The free optimization nobody does.

  4. The 24-Hour Carbon Curve Wind at 3am, gas at 6pm. Scheduling batch inference for the cleanest hours costs nothing.

  5. Scope 3 and the API Call You Can’t See Every inference call is a scope 3 emission. The cloud abstraction hides the carbon. Regulation is catching up.