SemiAnalysis: AMD MI355X hits 53.6% inference margin on Kimi K3, beating GB200 NVL72's 44.3%
AccBalanced · x · 2026-09-20
SemiAnalysis's open-source AgentX agentic inference benchmark finds AMD achieves better profit margins per gigawatt than Nvidia on the 2.8T Kimi K3 model: 53.6% margin on MI355X vs 44.3% on GB200 NVL72 (also ahead of B300 and B200), driven by lower TCO and better performance.
EmbeddedLLM amplified the finding, calling it the industry's best-kept secret and crediting two engineers for optimizing the model.
Related event: AMD Mi355X Beats Nvidia on Inference Profit Per Gigawatt: SemiAnalysis(2 posts)→
More from Infra
- These 10 GPU optimization papers are why your model runs fast today — NandoDF · 2026-11-04
- One prompt to audit cross-region traffic: dev cut trans-Atlantic latency by moving a service — DanielLockyer · 2026-11-04
- SurrealDB 4 launches as multi-model ACID database for the agentic era — Scobleizer · 2026-11-04
- Fire at Taiwan Panasonic PCB materials plant may tighten CCL supply further — zephyr_z9 · 2026-11-04
- Same Claude model served by Bedrock one day and Anthropic the next, dev finds — georgos_pappas · 2026-11-04
- Self-improving software is inevitable: closing the agent-observability loop — bibryam · 2026-11-04