Google LLC today introduced two new specialized silicon chips for artificial intelligence to its eighth-generation Tensor Processing Unit architecture family, aimed at higher compute density and secure inference.
At Google Cloud Next 2026 in April, Google introduced two specialized TPUs in the same generation, 8t for large-scale training and 8i designed for real-time reasoning. At the time, the company said it released the chips to handle demanding AI workloads across two primary tasks: building the models and running them.
Today’s expansion includes two new specialized chips across additional systems: TPU 8i Max for massive context windows and the highest possible compute density per rack, and TPU 8tx designed for secure and confidential inference.
TPU 8i Max, when memory is the bottleneck
Google said TPU 8i Max aims to scale compute and memory, delivering double the compute, double the SRAM and double the high-bandwidth memory of TPU 8i. This gives processors as much memory as possible to handle data without swapping it out.
Additionally, the company added that 8i Max fits twice the silicon in the same rack footprint, lowering the floor space it takes per square foot and per watt in data centers.
This allows the new eighth-generation TPUs to provide customers with a purpose-built accelerator for larger models and the industry’s current push toward ultra-large context windows.
A context window acts as an AI model’s working space: how much data it can consider while answering a question. A larger window lets it work across a longer document, conversation history, audio file, video or collection of code. But expanding it comes at a cost. Processing and repeatedly consulting more information puts greater demands on chips to swap data in and out, requiring more on-chip memory and stronger connections that move data between processors.
Adding more on-chip memory reduces the need to move data between processors, lowering latency and reducing the power an AI model needs to “think” about information.
Google said it developed 8i Max alongside pushing massive context windows for its Gemini frontier model, which supports the 1 million-token industry standard. That represents roughly 750,000 words or 1,500 pages of text. Although flagship models tackle 1 million tokens, smaller models often support narrower window sizes between 128,000 and 256,000 tokens.
TPU 8tx, for confidentiality
Google said it optimized TPU 8tx for secure, confidential inference, designing the chip for regulated enterprises and governments that want to run frontier or proprietary models and autonomous agents alongside the most sensitive data.
Normally, data is encrypted when stored or sent over a network, but when an AI model processes that same information, it must be decoded. Built on the same architecture as the TPU 8t, with custom AMD central processing unit hosts, the 8tx builds in a secure enclave directly into the hardware layer that keeps sensitive data and model weights protected and isolated.
The company said the architecture pairs training-grade compute with fast inference, meaning organizations can run and customize models on a single confidential system. The combination allows regulated services and industries to fine-tune models using private data, distill large models into smaller ones and train agents through reinforcement learning.
All of this can be done without migrating data between separate clusters, which could expose sensitive information to leaks.
Image: Google
Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.
15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more
11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network
SiliconANGLE Media is a recognized leader in digital media innovation, uniting breakthrough technology, strategic insights and real-time audience engagement. As the parent company of SiliconANGLE, theCUBE Network, theCUBE Research, CUBE365, theCUBE AI and theCUBE SuperStudios — with flagship locations in Silicon Valley and the New York Stock Exchange — SiliconANGLE Media operates at the intersection of media, technology and AI.
Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.
Source credits: