Inference profitability and GPU shortage

Building OpenCode with Dax Raad

Topic 9 of 19

Published May 27, 2026 · Clip 36:36–40:54 (4:18)

Dax explains that the floor cost of inference is electricity, and some models show ~80% margin between sticker price and what even mid-tier buyers pay; at Anthropic/OpenAI scale, margins could approach 90%. Prices have effectively risen as users migrated from Sonnet to pricier Opus defaults. He compares this to cloud computing's 'welcome secret' of profitability. He also describes the GPU shortage: exponentially growing inference demand meets linear GPU production, big tech vacuums up supply, and everyone is hoarding capacity — though historically shortages end in oversupply.