brionv / Wikimedia Commons / CC BY-SA 2.0
MOUNTAIN VIEW, Calif. – Google announced two new artificial-intelligence computing systems Monday, one designed to protect sensitive information while models process it and another built to handle larger workloads without running into memory limits.
The additions, TPU 8tx and TPU 8i Max, expand Google’s eighth-generation Tensor Processing Unit lineup to four systems. They are new configurations built around existing chips, not two entirely new chip architectures.
In its November 9 announcement, Google said the systems address different needs as businesses move AI agents into production: protecting proprietary information and giving software enough memory to work across large amounts of data.
TPU 8tx is aimed at regulated businesses and government customers running models alongside sensitive data. Google said it uses the same silicon as TPU 8t, paired with fifth-generation AMD EPYC processor hosts, to add hardware-level protection for data and model weights while they are in active use.
The distinction matters because information that is encrypted in storage or during transmission normally has to be unlocked in memory for processing. Google says its confidential-computing design protects and isolates that information during the work itself.
The second system, TPU 8i Max, combines two TPU 8i chips in a single package. Google says this doubles computing capacity, on-chip SRAM memory and high-bandwidth-memory bandwidth compared with TPU 8i. The goal is to keep large AI workloads supplied with data rather than leave processors waiting for memory.
Google described uses such as groups of AI agents working across an entire software codebase or hours of video. Those tasks can require much larger context windows, the amount of information a model can consider during its work.
The company also outlined software changes intended to make TPUs easier to use with PyTorch and open-source inference tools. One system, Agent Compute Engine, is designed to suspend and resume idle AI-agent sessions instead of keeping computing resources occupied while an agent waits for a database response or human feedback.
For businesses choosing AI infrastructure, the announcement separates security and memory requirements rather than treating every workload as the same computing problem. Google’s performance comparisons are company claims, not independent test results.
The announcement did not provide a price or a customer rollout timetable for the two new systems. Those details will determine how the technical changes translate into actual computing costs for customers.
JBizNews Desk | Mountain View, Calif.
© JBizNews.com. All rights reserved. Unauthorized reproduction or redistribution is strictly prohibited.
