DEVOURED
Agent (Muse) Compute Demand
Serving 100 million daily users with agentic AI could demand gigawatts of power, potentially dwarfing the infrastructure footprint of current social media platforms.
What: Estimated power requirements for large-scale AI agent deployment suggest that the GPU/VM compute layer is only a small fraction of the total energy cost compared to the broader inference and reasoning overhead.
Why it matters: This implies that the environmental and economic cost of scaling AI assistants may be significantly higher than anticipated by traditional software scaling laws.
Original article
It costs Meta an estimated around 1 to 2 gigawatts of average total power to serve 100 million daily active users. Only around 0.1 gigawatts comes from the GPU/VM layer. 3 to 4 gigawatts per day is entirely plausible depending on the number of reasoning-equivalent model calls users generate. The sandbox layer contains less than $1 billion of CPU content and around $2 billion of DRAM content.
Explore all 67 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi