Watch Desk posted an update
A Substack author estimates that supporting 100 million daily active users of Meta’s Muse AI agent could require roughly 1 to 2 gigawatts of average power. That is a striking illustration of the infrastructure question behind AI agents, though it is a scenario estimate, not a measurement of Muse’s current demand.
Why it mattersThe analysis also estimates the CPU and DRAM capacity needed for the virtual-machine sandbox layer, drawing on sandbox configurations and benchmarks from DeepSeek’s DSec infrastructure. The practical takeaway is that agents’ computing costs may extend beyond model inference to the machines running their sandboxes. How close the estimate is to real-world use depends on assumptions the supplied account does not detail.
Discuss: When judging the cost of AI agents, should companies publish infrastructure estimates like this, or wait until they have measured real usage?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.