Alphabet’s Google is now selling its custom-designed Tensor Processing Units (TPUs) to external data center operators. This strategic pivot ends Google's practice of reserving these powerful AI accelerators for internal use. The company aims to compete directly with Nvidia in the merchant AI chip market.
The eighth-generation lineup includes the TPU v8t for model training and the TPU v8i for AI inference. Google is specifically targeting the growing demand for inference workloads and alternatives to Nvidia’s dominant GPUs.
While Google Cloud reports strong backlog growth, the move faces resistance from data center operators with deep Nvidia ties. Google is reportedly offering increased financial support to partners to encourage adoption of the new hardware.