Hardware
TPU Inference Externalization Full Steam Ahead - InferenceX
Google has begun allowing third‑party customers to acquire or rent its TPU chips—starting with the TPUv7 “Ironwood” generation. In comparative tests against NVIDIA Volta (B200/B300), Ironwood shows up
Google has begun allowing third‑party customers to acquire or rent its TPU chips—starting with the TPUv7 “Ironwood” generation. In comparative tests against NVIDIA Volta (B200/B300), Ironwood shows up to 50 % higher performance per dollar across most operating points, and Google’s own total cost of ownership analyses confirm similar gains for external users such as Anthropic. These results illustrate a significant shift toward wider TPU adoption outside Google’s internal workloads.
Related
- OpenAI Jalapeño: Better Than Nvidia Blackwell
- Vera Rubin NVL72 vs GB200 NVL72? Inference TCO & Architecture Analysis
- DeepSeekV4 1.6T Day 0 to Day 43 Performance Over Time - Huawei, GB300 NVL72, MI355X, B200
Source: SemiAnalysis | 2026-09-07