How many AI agents could run on the AI chips shipped through 2027?
Summary
Li models how many simultaneous AI-agent sessions projected chip shipments could support. Under full deployment and allocation to today’s frontier-model workloads, high-bandwidth memory shipped in 2025–27 could support roughly 30–170 million concurrent sessions. This is potential serving capacity, not a count of productive workers or a forecast of actual agent use.
Why it matters
The estimate connects chip investment to inference supply and the demand needed to use it. More efficient models raise potential capacity further, making useful applications and customer spending central to the economics of the buildout.
What to keep in mind
This conditional model holds current workloads fixed and depends on memory-supply projections, serving-cost assumptions and utilization. Chips may await data centers and power. Agent hours do not establish human-equivalent productivity, revenue, wages or job displacement.