TL;DR
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Google announced two systems built around its eighth-generation AI chips: TPU 8tx for confidential inference and TPU 8i Max for workloads with very long inputs. The additions bring the generation to four TPU systems in one year, but Google says the new variants will not be available until 2027. Pricing, launch regions and independent performance results have not been published.
Google announced two new systems for its eighth-generation AI chips on November 9: TPU 8tx, designed for confidential inference, and TPU 8i Max, aimed at workloads that need to process very long inputs. The additions give Google four TPU systems in one year as it seeks to offer businesses an alternative to relying on Nvidia GPUs, although the new systems are not expected to be available until 2027.
Google introduced the eighth-generation TPU family at its Cloud Next conference on April 22, initially dividing it into TPU 8t for training AI models and TPU 8i for inference, the process of running a trained model to generate answers. The November announcement adds systems tailored to other requirements. Google says TPU 8tx is intended for regulated businesses and government agencies that need to protect sensitive information while it is being processed. TPU 8i Max is designed for jobs involving very large context windows, such as an AI agent working with an entire codebase or long video files.
The two additions use existing chip designs in different system configurations. Google says TPU 8tx uses the same silicon as TPU 8t, paired with fifth-generation AMD EPYC processors. It claims the system can deliver up to two times lower latency than TPU 8t with a near-zero performance penalty. TPU 8i Max puts two low-latency TPU 8i chips on one package; Google says this doubles compute, on-chip SRAM and high-bandwidth memory bandwidth compared with TPU 8i, within the existing rack footprint.
Those performance figures come from Google, and no independent benchmarks for the new systems are available. Google’s TPU product page lists all four eighth-generation systems as “Coming soon.” Amin Vahdat, Google’s chief technologist for AI and infrastructure, told Bloomberg the two new variants are expected in 2027. Google has not announced prices or the cloud regions where customers will be able to use them.
Google adds two TPU systems to challenge Nvidia dependence
TPU 8tx targets confidential inference, while TPU 8i Max is built for very long inputs. They offer businesses another potential source of AI computing, but availability and independent performance evidence are still ahead.
Two systems for distinct workloads
Google says both variants reuse existing chip designs in different system configurations.
TPU 8tx
Designed to protect sensitive data and model weights while in use, for regulated businesses and government agencies.
TPU 8i Max
Aimed at workloads such as agents processing entire codebases or systems working with long video files.
What customers may gain
More accelerator options could help businesses plan AI capacity around specific needs.
- Another possible source of AI computing capacity
- Systems tailored to confidential data or very long inputs
- Google TPUs planned for Oracle Cloud in 2027
- Prices, launch regions and customer access
- Independent results for real workloads
- Engineering effort to move applications from other accelerators
From announcement to buying decision
Google expects TPU 8tx and TPU 8i Max in 2027; specific dates have not been announced.
Compare published pricing, capacity and cloud regions when details emerge.
Use independent benchmarks and workload-specific results to assess performance and switching effort.
Claims still need outside evidence
The performance figures published so far come from Google; independent benchmarks for the new systems are not available.
More Options for AI Infrastructure
The announcement gives cloud customers another potential source of AI computing capacity, with systems aimed at specific needs rather than a single general-purpose configuration. Companies handling regulated data may value the security design Google describes for TPU 8tx, while teams running agents over large codebases or processing long videos may be interested in TPU 8i Max’s memory and bandwidth claims. Whether those advantages hold up in practice remains to be tested outside Google.
Supply is part of the commercial case. Google says nearly 100 customers use its TPUs, and Alphabet has described a supply-constrained environment. Oracle has also said it will offer Google TPUs through its cloud starting in calendar year 2027. If customers can access capable hardware from more than one provider, they may gain another option when planning capacity. But an additional supplier does not automatically make switching simple: customers may need to adapt software, workflows and infrastructure to use a different accelerator.
The timing also limits what buyers can conclude now. The new systems are not available, and the claims lack independent benchmarks and published prices. Businesses weighing them against Nvidia GPUs or earlier Google TPUs will need concrete information on cost, availability and performance for their own workloads before making a comparison.
Popular right now
Ninja CREAMi 7-in-1 Ice Cream Maker, NC299AMZ
Dyson Airwrap i.d. Multi-Styler for Straight and Wavy Hair
Dyson V8 Cyclone Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
How Google Split Its TPU Line
Google’s April announcement introduced TPU 8t for large-scale training and TPU 8i for inference. Google said TPU 8t is designed for superpods of 9,600 chips. Its published specifications for the base systems list 216 GB of high-bandwidth memory, 128 MB of on-chip SRAM and 12.6 petaflops of peak FP4 compute for TPU 8t. For TPU 8i, Google lists 288 GB of high-bandwidth memory, 384 MB of SRAM and 10.1 petaflops of peak FP4 compute. These are manufacturer specifications, not independent measurements.
The differences reflect distinct workload priorities: training involves building models, while inference serves requests from models already trained. Google’s published figures give TPU 8i more memory and bandwidth than TPU 8t, despite lower peak compute. The new TPU 8i Max extends that memory-focused approach for especially long inputs. TPU 8tx applies the TPU 8t design to confidential inference, which aims to protect data and model weights while they are actively in use.
Google says it can develop these variants by changing packaging, memory, networking and security around existing silicon, a process it describes as taking months rather than the years needed to design a new chip. Vahdat told Bloomberg that the new systems are not completely new silicon designs and reflect tasks customers asked Google to address. The expansion marks a quicker release pace than the roughly two-year intervals Bloomberg reported for Google’s earlier AI processor releases.
““This is the first time we have introduced four TPU systems in a single year.””
Prices, Access and Performance Remain Open
Prices, availability by region and customer access details have not been announced for the new systems. Google has pointed to 2027 availability through Vahdat’s comments to Bloomberg, but has not specified launch dates. Independent testing has also not established whether the claimed latency, compute, memory and bandwidth advantages translate into improvements for real customer workloads.
It is also unclear how readily customers can move existing applications from Nvidia GPUs or other accelerators to Google TPUs, and what engineering work that would require for particular workloads. Google has highlighted software developments including TorchTPU, support for the vLLM and SGLang inference engines, and the general availability of Cloud Pathways on Google Kubernetes Engine. Those offerings may help with adoption, but their effect on deployment effort and performance will depend on customer use cases. The available information does not establish whether the new systems will ease supply constraints or how they will compare on total cost.
The 2027 Launch Will Test Demand
The next concrete milestone is the expected 2027 availability of TPU 8tx and TPU 8i Max. Google has not named specific launch dates, prices or regions. Oracle’s planned introduction of Google TPUs in its cloud in calendar year 2027 could also widen access, though details on which systems it will offer have not been provided.
When the systems become available, buyers and cloud customers will be able to compare published pricing and capacity with independent tests and workload-specific results. Until then, Google’s stated performance figures remain vendor claims, and the practical value of the new options for customers is not yet established.
Key Questions
What are TPU 8tx and TPU 8i Max designed to do?
Google designed TPU 8tx for confidential inference, with a focus on protecting data while it is being processed. TPU 8i Max is aimed at inference workloads with very long inputs, such as agents working across large codebases or long videos.
When will the two systems be available?
Google’s Amin Vahdat told Bloomberg they are expected in 2027. Google has not announced specific launch dates, cloud regions or prices.
Are the performance claims independently verified?
No independent benchmarks for TPU 8tx or TPU 8i Max are available in the source material. The reported latency, compute and memory comparisons are claims from Google.
How do the new systems fit into Google’s TPU lineup?
Google introduced TPU 8t for training and TPU 8i for inference in April 2026. The November additions bring the eighth-generation family to four systems, with TPU 8tx focused on confidential inference and TPU 8i Max on very long inputs.
Source: Thorsten Meyer AI
