Google announced two new variants of its in‑house tensor processing units (TPUs) on Monday: the TPU 8tx and the TPU 8i Max. These chips expand the company’s eighth‑generation TPU family, introduced in April, to cover confidential inference scenarios and workloads that require more memory capacity.
The April generation of TPUs focused on increasing inference throughput and supporting large training demands; the two new models build on those capabilities and target more specific use cases.
TPU 8tx: hardware‑level security for confidential model training and inference
According to Google’s blog post, the TPU 8tx is based on the TPU 8t chip and was developed in collaboration with AMD. It is designed specifically for confidential training and inference of AI models. The chip incorporates security directly into the hardware layer so that sensitive data and model weights remain protected and isolated even while they are actively used.
TPU 8tx also includes a custom inferencing network that enables reasoning to run alongside training. That allows the chip to both distill and train smaller models on the fly, which can be useful for organizations that want to train models on proprietary data without moving that data between devices.
TPU 8i Max: doubled memory and bandwidth in one package
The TPU 8i Max effectively combines two inference‑focused TPU 8i chips that Google launched in April, doubling memory capacity, processing power and bandwidth within a single package. This design lets models handle a broader range of information during queries, benefiting applications that need larger context windows or more data available at inference time.
Context and significance
Google’s launch occurs amid a broader industry trend in which hyperscalers and AI labs are developing their own AI‑specific semiconductors and supporting software to reduce reliance on dominant external suppliers. Companies pursuing in‑house or tailored chips include Microsoft, Amazon, and AI organizations such as OpenAI and Anthropic.
The announcement arrives at a moment of increased scrutiny for Nvidia after reports that Nvidia CEO Jensen Huang admitted over the weekend to an affair involving a person some have alleged could be a Chinese spy. That episode has drawn additional attention to competitive dynamics and supply‑chain risks in the chip industry.
Taken together, the TPU 8tx and TPU 8i Max demonstrate Google’s effort to round out its internal chip portfolio with targeted capabilities for enterprise needs such as data confidentiality and larger‑memory inference workloads.
