Brain Runs on 20 Watts: Can Neuromorphic Computing Make AI Less Power-Hungry?
burny_tech · x · 2026-09-20
A Scientific American explainer on neuromorphic (brain-inspired) computing and the energy-efficiency gap between brains and AI:
- The human brain packs 86 billion neurons and runs on roughly 20 watts—about a lightbulb—yet today's AI systems need massive compute infrastructure to replicate even a fraction of its capabilities.
- Two features make the brain so efficient: energy is spent only where needed (neurons fire discrete spikes once input crosses a threshold), and memory and computation are intertwined in synaptic strength, so the brain never pauses to fetch data.
- Conventional chips separate processor and memory physically, so every data transfer costs time and energy.
- Neuromorphic computing aims not at incremental efficiency gains but at fundamentally rethinking how computers work, potentially bringing AI closer to brain-level power consumption.
More from Infra
- Musk confirms every Starlink V3 satellite will carry an Nvidia Vera Rubin NVL72 — 100k sats, 25GW — ns123abc · 2026-09-20
- SpaceX's 100,000-satellite Starlink V3 constellation accepted by the FCC — ns123abc · 2026-09-20
- jevcache memoizes model decisions by (model, schema, state) — repeat calls cost $0 and return in ~0ms — JiliJeanlouis · 2026-09-20
- Key takeaways from the 2026 AI Infra Summit with 9,000+ attendees — karlfreund · 2026-09-20
- Ternary 2-bit Bonsai-2-27B GGUF lands on Hugging Face trending — dealignai · 2026-09-20
- Pedro Domingos: ASML Is a Single Point of Failure for the AI Supply Chain — pmddomingos · 2026-09-20