Normalized by intelligence delivered, AI inference is already more energy-efficient than the human brain
FlorianGallwitz · x · 2026-09-14
Countering the popular claim that current LLMs on GPUs are less energy-efficient than the human brain, the author offers a rough calculation: normalizing as IQ points per watt (adjusting for serving speed and parallel requests), humans score 5 while AI scores 7–40. He argues the original comparison wrongly pitted one brain against an entire data center; larger models mostly mean more GPUs serving requests in parallel, not worse per-request efficiency. On this metric, AI already outperforms brains for equivalently intelligent work.
More from Infra
- SK hynix completes HBM4 internal qualification, ushering in custom base die competition — blaizedsouza · 2026-09-14
- One architectural change cuts KV cache 8x: how GQA works, explained with Llama 3 70B — blaizedsouza · 2026-09-14
- A complete breakdown of HBM system architecture, from DDR roots to GDDR7, PIM and HBF alternatives — blaizedsouza · 2026-09-14
- Nvidia paper shows transformer LLMs can be sparser, faster, and lighter without losing accuracy — YesThisIsLion · 2026-09-14
- Musk: 10 million tons to orbit per year needed for terawatt of space compute — elonmusk · 2026-09-14
- Dual NVIDIA Spark agent setup: heat is the killer, headless mode saves 2-3GB RAM — jasonkneen · 2026-09-14