Google broadens TPU lineup with two new specialized chips for confidential training and bigger memory
TechCrunch AI · rss · 2026-11-10
Google has expanded its AI hardware portfolio with two new specialized tensor processing units (TPUs), targeting more use cases beyond general training and inference: one chip is aimed at confidential model training, and the other focuses on larger memory capacity. The move deepens Google's push into custom AI silicon, competing with AWS Trainium and Nvidia's GPUs for cloud AI workloads.
More from Infra
- Ray creator shows KubeRay joining CNCF as Ray powers MAI, Cursor, NVIDIA and more — PyTorch · 2026-11-11
- Epoch AI chart: AI cost falling ~13x a year, faster than any tech in history — sebkrier · 2026-11-11
- CoreWeave details its open source contributions to vLLM, llm-d and Kubernetes at KubeCon — PyTorch · 2026-11-11
- turbopuffer ships CJK tokenizers; dev shares hard-won Chinese/Japanese/Korean search tuning tips — bo_wangbo · 2026-11-11
- Visualized: GLM 5.3 decode offload reads 13GB of weights plus experts per token — HankYeomans · 2026-11-11
- Anthropic chart shows 12x YoY revenue growth and ~$81B run-rate with 4x compute-efficiency gains — firstadopter · 2026-11-11