> ## Content Index
> Fetch the complete content index at: https://adjacent.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI's Power Demand Will Hit Physical Limits by 2027
- URL: https://adjacent.media/signals/ais-power-demand-will-hit-physical-limits-by-2027/
- Published: 2026-11-05T00:17:08.000Z
- Updated: 2026-11-05T00:17:08.000Z
- Description: Data center electricity consumption is on track to demand 80 gigawatts by 2027—equivalent to doubling all compute infrastructure ever built in the U.S. within a single year—creating a hard constraint on AI scaling that no amount of efficiency gains can fully overcome.
- Author: Jonathan Greene
- Tags: #signal, theme-connected, energy, data centers, infrastructure

Source: [Tomasz Tunguz](https://tomtunguz.com/watts-in-store-for-2027-inference/?ref=adjacent.media)

Data center electricity consumption is on track to demand 80 gigawatts by 2027—equivalent to doubling all compute infrastructure ever built in the U.S. within a single year—creating a hard constraint on AI scaling that no amount of efficiency gains can fully overcome. The bottleneck isn't chips or talent but grid capacity and power generation, forcing cloud providers and chip makers to relocate infrastructure, invest in nuclear energy, and make trade-offs between model size and geographic reach. AI deployment is becoming a physical infrastructure problem, not just a software one, determining which companies can afford to train and deploy the largest models.