First third-party TPU inference benchmark: Google Ironwood up to 50% better perf/$ than B200
dylan522p · x · 2026-09-08
SemiAnalysis and the InferenceX team published the first open third-party inference benchmarking of Google TPUs, running daily across many models and scenarios. Key takeaways:
- Headline result: In apples-to-apples tests, TPUv7 Ironwood delivers up to 50% better performance per dollar than B200/B300, holding an advantage across much of the Pareto curve.
- Economics: The report analyzes both Google's internal TCO and the external TCO a real customer pays.
- Context: Anthropic is the biggest TPU user, committed to over 1 million chips (400k+ direct purchases, 600k+ rented via GCP), and will surpass DeepMind's usage by 2029.
- Trend: Rapid externalization of the TPU stack (Ironwood, TPUv8i) is eroding NVIDIA's CUDA moat.
More from Venture
- Anthropic files S-1: Kuber Denis reacts to the full IPO registration statement — himanshustwts · 2026-10-26
- NBER: Only 2.9% of US consumers pay for AI; retention as low as 20%, pricing power at risk — avicgoldfarb · 2026-10-26
- General agents burn ~140M tokens/month on LinkedIn outreach; scripts do it 60x cheaper — ChrisHarpon2 · 2026-10-26
- Fireworks captures ~35k monthly clicks with model-name SEO, 9 pages for one Kimi K3 — FinanceYF5 · 2026-10-26
- Anthropic's $80.7B H1 loss is almost all non-cash fair-value swings — adjusted net income is positive — BenBajarin · 2026-10-26
- DeepSeek reportedly raised ~$15B in one of China's largest funding rounds ever — thione · 2026-10-26