AMD's Mi355X Beats Nvidia GPUs on Per-Gigawatt Margins Running Kimi K3, SemiAnalysis Finds
m2saxon · x · 2026-09-19
SemiAnalysis reports that running the 2.8T-parameter Kimi K3, AMD's Mi355X delivers 53.6% profit margins per gigawatt, beating GB200 NVL72 (44.3%), B300, and B200, thanks to lower TCO and better performance.
The numbers come from AgentX, their open-source agentic inference benchmark, with optimizations contributed by two engineers on the 2.8T model. If it holds up, this directly challenges the assumption that AMD can't compete in the inference compute market.
More from Infra
- Qwen 27B Hits 1253 tok/s on 2x RTX 5090, 8-10x Cheaper Than API Pricing — me_broke · 2026-09-19
- India hosts ~20% of global chip design engineers, and the real share is likely higher — bookwormengr · 2026-09-19
- Micron and the AI memory cycle: HBM heads to $100B by 2027, 5-year deals reshape the trade — Beth_Kindig · 2026-09-19
- Redditor builds ROI calculator matching local LLMs to hardware by memory bandwidth — SnoobieJunes · 2026-09-19
- NVIDIA's AIPerf benchmarks LLM inference at scale with multiprocess load testing — NVIDIAAI · 2026-09-19
- Dev builds daily-driver AI coding environment on AWS Lambda MicroVMs with S3 persistence — blaizedsouza · 2026-09-19