Every AI model is paid for twice: once to train it, and forever after to run it. Training was the headline. Inference is the bill that never stops — and every time an agent replaces a click, that bill multiplies. The industry just signed up to run billions of digital workers on infrastructure that turns most of its electricity into heat.

The tell is already public: token prices falling, compute spending rising, margins moving the wrong way. This report explains why — and what ends it.

Inside: why agent economics break on electron-based inference; where the true cost of a token comes from (heat and conversion, not intelligence); how logic that runs on light rewrites cost-per-inference; and who is positioned to own the infrastructure of the agent era.

This isn't a prompt guide or an AI hype piece. It's an industry map for the largest infrastructure buildout in computing history — written for builders who want to see it before their industry does.

Your $197 purchase includes one full year of First Light General Membership: member letters and chat, monthly Q&A, and lifetime updates to this report as the numbers move.