KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
Mining
Neutral

Human Brain Uses 20 Watts While AI Demands Power Plants

Published September 27, 2026 1:29 PM · 8 views
Human Brain Uses 20 Watts While AI Demands Power Plants

The human brain operates at approximately 20 watts. China’s 'LineShine' supercomputer consumes 42.2 million watts. While this comparison appears valid, most viral social media posts misrepresent data from a single paper that has been widely cited incorrectly for years.

Reputable sources like Epoch AI estimate ChatGPT uses about 0.3 watt-hours per query as of early 2025, aligning with findings in the journal Joule. Both figures vary significantly based on workload—longer responses require more energy. Biological systems and silicon-based computing share a key insight: both operate at low precision levels.

Neuroscience researchers note that brain neurons do not use 32-bit precision, while semiconductor companies like DeepSeek trained models with 8-bit precision, and NVIDIA used 4-bit for smaller models. The brain also processes memory and computation in the same physical space—a feature digital systems lack, leading to significant energy inefficiencies when fetching operands from memory.

Neuromorphic hardware attempts to mimic this but has not yet achieved commercial success. Intel’s Hala Point prototype with 1.1 billion artificial neurons remains a research project. Meanwhile, companies like BrainChip report minimal revenue—$700k in Q1 2025—with high operating costs of $5.3M.

Bitcoin miners are already addressing the power bottleneck. They’ve secured land, power contracts, and grid connections for years, with existing facilities requiring $3-4 million per megawatt to repurpose. New construction costs $10-12 million per megawatt. Major mining firms have signed over $70 billion in AI-related contracts, though actual operational capacity as of Q2 2026 stands at about 750 megawatts across the industry.

Despite efficiency gains, global data center energy use has surged from 58 terawatt-hours in 2014 to 176 terawatt-hours by 2023. While historical trends absorbed increased computing demand through efficiency, AI’s growth now faces physical power constraints rather than computational limits.

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: 뇌, 20와트 사용…AI는 발전소 필요