TPUv7 Delivers Up to 50% Better Inference Performance per Dollar Than B200/B300

SemiAnalysis's report "The Iron(wood) Throne" argues Google TPUv7 delivers up to 50% higher inference performance per dollar than NVIDIA's B200/B300, directly challenging the CUDA moat. The SGLang team confirmed it is working with Google and that TPUv7 inference now rivals mainstream GPU setups.

2026-11-06 ~ 2026-11-06 · 2 related posts

• SemiAnalysis: Google TPUv7 Delivers Up to 50% Better Performance-per-Dollar Than B200/B300 for Inference — vllm_project · 2026-11-06
• SGLang Says TPUv7 Now Competitive With GPUs for Long-Context Agentic Inference — ying11231 · 2026-11-06