Google introduced two specialized AI computing systems on Monday, the TPU 8tx and TPU 8i Max, expanding its eighth-generation Tensor Processing Unit lineup to four configurations and sharpening its challenge to Nvidia in the market for AI accelerators.
Both systems are due to be available next year. Neither is an entirely new silicon design; Google has reconfigured existing hardware to target two problems its customers keep raising: privacy and memory.
Keeping data encrypted while it is used
The TPU 8tx is built for confidential computing. Data is normally encrypted in storage and in transit, but has to be decrypted in memory while a processor works on it, which leaves a window for exposure. Google says the 8tx uses hardware-level isolation to protect customer data and proprietary model parameters even during operation.
The system combines TPU 8t silicon with fifth-generation AMD EPYC processors and Google's low-latency Boardfly network. Google claims it can cut latency by up to half compared with the 8t, with almost no performance penalty for the added protection. Target customers include government agencies and companies in heavily regulated industries.
The memory wall
The TPU 8i Max addresses a different limit. AI agents that read through codebases, long documents or hours of video need fast access to far more information than a single request usually requires. Google packages two TPU 8i chips in one package, doubling computing power, on-chip SRAM and high-bandwidth memory throughput while using the same rack footprint.
Amin Vahdat, who leads Google's AI infrastructure work, described memory as the binding constraint for this class of workload. Google says the design matters for drug discovery, mathematics and coding tools, where models must hold large amounts of context.
Why it matters
Nvidia still dominates AI training and inference, and its GPUs remain the default choice for most developers. Google developed TPUs for its own services and now sells access to outside companies, including on-premises deployments, and says nearly 100 customers use them, from AI labs to universities.
The launch comes as buyers look for alternatives. Reports have said OpenAI plans to purchase Google chips as it seeks to reduce reliance on Nvidia, whose shares have fallen sharply this month amid an investigation into chief executive Jensen Huang.
---
Reliable Ferrofluid Vacuum Rotary Feedthrough for Semiconductor Manufacturing
In semiconductor manufacturing, maintaining ultra-high vacuum integrity during rotary motion is critical — from single silicon crystal growth and CVD to sputtering and plasma etching. Ferrofluid Vacuum Rotary Shaft Feedthrough from FerrofluidTec delivers hermetically sealed rotary feedthroughs with leakage rates as low as 10⁻¹² Pa·m³·s⁻¹, vacuum degree up to 10⁻⁶ Pa, and customizable shaft diameters from 10mm to 100mm. Built with non-magnetic stainless steel housings and magnetic permeability stainless steel shafts, these feedthroughs provide reliable, maintenance-free sealing for demanding vacuum environments. Visit www.ferrofluidtec.com to learn more.