Google unveiled two new tensor processing units on Monday, expanding its in-house AI chip lineup to reduce reliance on external semiconductor suppliers for cloud computing.
TPU 8tx, designed with AMD, is based on the TPU 8t chip and targets confidential AI model training and inference. The chip "builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use," Google wrote in a blog post. The chip features a custom inferencing network that runs reasoning alongside training, allowing companies to train models on proprietary data without moving data between devices.
TPU 8i Max combines two inference-focused TPU 8i chips released in April, doubling memory capacity, processing power and bandwidth in a single package. This enables AI models to process larger volumes of information during queries.
Both chips are part of Google's eighth-generation TPU line introduced in April. The launch reflects a broader industry trend where hyperscalers like Microsoft and Amazon, along with AI labs such as OpenAI and Anthropic, are developing proprietary semiconductors and software to reduce dependence on Nvidia.