Google open-sourced TorchTPU, its native PyTorch stack for TPUs, and released llm-d 1.0, a Kubernetes-native distributed inference stack co-founded with Red Hat, IBM and others, at PyTorch Conference 2026.
Nvidia holds 81% AI chip market share, but Alphabet's TPUs like Ironwood and new 8-series pose the sharpest long-term challenge, closing performance gaps while hyperscalers build alternatives amid surging inference demand.
Tensor Processing Units are a technology developed and owned by Google. While you can find GPUs in every cloud provider offer, the TPUs are currently only available through Google Cloud Platform.…
Google's eighth-gen TPUs split into 8t for training and 8i for inference, delivering 3x compute gains, 80% better inference value, and massive pods. Backed by Broadcom deals and Anthropic's million-chip bet, they challenge Nvidia on cost and scale.
