Technology and Scienceupdated 16m ago2 min read
Google Launches TPU 8tx And TPU 8i Max For Agentic, Confidential AI
“also compete with Nvidia in a billion-dollar market.”
“The competition in AI infrastructure is not about building a faster processor.”
This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.
Google adds 8tx and 8i Max to TPU lineup for confidential AI and agentic workloads. TPU 8tx enables confidential computing during AI model inference.
Save coverage image
Two new TPU systems
Google expanded its eighth-generation Tensor Processing Unit portfolio with two new AI systems, TPU 8tx and TPU 8i Max, and positioned them for “agentic artificial intelligence (AI).” Google said TPU 8tx focuses on secure and confidential AI inference for enterprises and government organisations that need to run AI models and autonomous agents alongside sensitive information. Google said TPU 8i Max targets memory-intensive workloads by combining two low-latency TPU 8i chips into a single package to address memory bandwidth bottlenecks when multiple agents work together.
Google also said the TPU 8tx uses the same silicon as TPU 8t and pairs it with fifth-generation AMD EPYC CPU hosts, while the TPU 8i Max fits twice the silicon within the existing rack footprint. Google’s announcement framed the expansion as part of a broader eighth-generation lineup that includes TPU 8t for large-scale model training and TPU 8i for low-latency reasoning.
Confidential inference and memory
Google said TPU 8tx “builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use.” Google said TPU 8tx is designed to keep data and model weights protected even while they are being processed, and it supports fine-tuning models using private datasets and training AI agents through reinforcement learning without moving sensitive data between separate computing clusters. Google said TPU 8i Max “basically combines two of the inference-focused TPU 8i chips the company launched in April, doubling the memory capacity, processing power and bandwidth, but in a single package.”
Google’s materials described TPU 8i Max as a way to help processors access data more quickly and reduce the chances of large workloads slowing down because of memory constraints. Google’s broader software push accompanied the hardware, with the company introducing tools including support for PyTorch and Hugging Face models and open-source inference engines.
Competition and availability
Google’s TPU expansion arrived as the company sought to reduce its dependence on chips made by other companies for its cloud computing business, and TechCrunch said the launch broadened its chip portfolio for confidential inference and more memory capacity for large workloads. TechCrunch reported that the TPU 8tx launch came with Google’s claim that it was “geared specifically for confidentially training and running AI models,” and it described TPU 8i Max as doubling memory capacity, processing power, and bandwidth in a single package. CriptoTendencia said Google told Bloomberg that the 8tx and the 8i Max are not completely new architectures, but different configurations of silicon the company already manufactures, and it added that both systems will be available in 2027.
TradingView said Alphabet shares rose after Google unveiled the TPU 8tx and TPU 8i Max, and it reported that Alphabet said all available TPU capacity is contracted. TradingView also said Bloomberg Intelligence expects related sales to exceed $135 billion by 2028, while it reported that Oracle plans to offer them next year.