# HBM demand — X 热门讨论 (2026-09-24 12:52 UTC)

## @moninvestor (mon) · 09-24 08:00 · ♥44 ↻4 💬6 We are still early.

If AI agents eventually scale to hundreds of millions of users, each agent will require infrastructure around it to actually work.

• Memory: More agents means more DRAM and storage on top of the HBM already required for AI inference.

• CPUs: Agents need CPUs to run browsers, code, tools and background tasks. This adds another major source of server CPU demand for companies like $INTC and $AMD.

• Neo-clouds: All of this requires more compute, more servers and more data center capacity, benefiting companies like $IREN and $NBIS.

And underneath everything is power.

AI agents → more GPUs + CPUs + memory → more data centers → more electricity. > 引用 @firstadopter: Imagine if one billion people used a personal AI agent. That's a lot of CPUs and memory https://t.co/ibozUi3a07 https://x.com/moninvestor/status/2103031863950410195