AMD: AI Inference Compute Surpasses Training, Targeting $2T TAM by 2030
ryanshrout · x · 2026-07-24
AMD CEO Lisa Su noted that AI token consumption has skyrocketed by nearly 160x over the past two years. Marking a historic shift, compute dedicated to AI inference has now surpassed training compute for the first time, even as AI agents are just beginning to launch this year.
Based on this trend, AMD projects the total addressable market (TAM) for AI compute will reach $2 trillion by 2030. This includes a $220 billion CPU market with a 50% CAGR, with a significant portion of compute shifting towards edge and local devices.
Related event: AMD Declares AI Inference Compute Surpasses Training(4 posts)→
More from Venture
- Investor says the AI buildout could leave Mag 7 winners closer than expected — JOBhakdi · 2026-07-24
- Product Hunt launch brings 4 sales and $258.95 in revenue — ayushtweetshere · 2026-07-24
- Greylock says consumer AI can escape OpenAI’s “God Box” through vertical services and networks — SethGRosenberg · 2026-07-24
- Paper raises $34 million Series A from Accel and ICONIQ Capital — hewarsaber · 2026-07-24
- Gary Marcus says AI-heavy tech stocks now trade like crypto — GaryMarcus · 2026-07-24
- Multi-agent tools that share workspace and approvals cut task cost to $1.65 — HeyNayeem · 2026-07-23