🔍 Read the full analysis: How Google’s Two New AI Chips Fit Into Its Broader AI Strategy on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Google announced two additions to its eighth-generation TPU systems: TPU 8tx, designed for confidential inference, and TPU 8i Max, designed for workloads with very long inputs. Google says both will be available in 2027, but has not released prices or independent performance results. The announcement reflects a broader strategy of offering specialised alternatives to Nvidia GPUs and making Google’s own chips easier to use.
Google announced two new eighth-generation AI chip systems on November 9: TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads with very long inputs. The additions expand Google’s effort to offer customers alternatives to Nvidia-based computing, but neither system is available yet; Google says they are due in 2027.
The new systems build on the TPU 8t and TPU 8i, which Google introduced at its Cloud Next conference on April 22. TPU 8t is designed for training AI models, while TPU 8i is intended for inference—the computing used to generate responses from a trained model. Google says TPU 8tx is based on the same silicon as TPU 8t, paired with fifth-generation AMD EPYC processors and configured to protect sensitive data and model weights while they are being used.
Google says TPU 8tx can deliver up to twice the lower latency of TPU 8t with a near-zero performance penalty. TPU 8i Max combines two low-latency TPU 8i chips on one package. Google says it doubles the base system’s compute, on-chip SRAM and high-bandwidth memory bandwidth, while fitting twice the silicon within the existing rack footprint. The company is targeting tasks such as processing large codebases or hours of video.
Those figures are Google’s claims, not independent test results. Google has not published absolute specifications for either new system, or prices and regional availability. The company’s TPU 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 would be available in 2027.
Google’s two new AI chips fit a broader strategy
TPU 8tx and TPU 8i Max add specialised options to Google’s eighth-generation systems. They aim to widen customer choice while making Google’s own chips easier to use.
Two systems, distinct jobs
Google is tailoring systems around different AI workloads.
TPU 8tx
Uses TPU 8t silicon with safeguards for sensitive data and model weights during use.
TPU 8i Max
Combines two TPU 8i chips on one package for tasks such as large codebases or hours of video.
From two base chips to four
The new variants extend systems Google introduced in April.
TPU 8t is designed for AI model training.
TPU 8i is intended for inference, generating responses from trained models.
TPU 8tx adds confidential computing; TPU 8i Max targets very long inputs.
Hardware needs a software path
Google is also working to reduce the effort of moving AI workloads onto TPUs.
What buyers still need
Businesses cannot yet judge cost, access or real-world performance.
- Both systems expected in 2027
- Specialised designs for confidentiality and long context
- Google’s performance and latency claims
- Prices, launch regions and exact dates
- Independent performance and security results
- How widely customers can access the systems
Google’s Bid for More AI Workloads
The announcement shows Google broadening its chip offering through specialised systems built around customer needs, rather than presenting one processor as a fit for every AI job. Confidential computing may appeal to regulated businesses and government agencies that need safeguards for data during processing. The higher-memory configuration of TPU 8i Max is aimed at models that must handle unusually large amounts of context.
More options could give cloud customers another source of AI computing alongside Nvidia GPUs. That matters in a market where demand for processors has outstripped supply, according to the source material. Alphabet has described a supply-constrained environment, and Google says nearly 100 customers use its TPUs. A second provider can give buyers more choice, though it does not automatically make switching easy or cheaper.
Google is also working on the software and distribution around its chips. It points to TorchTPU, which it open-sourced, and support for the vLLM and SGLang inference engines. Cloud Pathways, the runtime Google DeepMind uses behind Gemini, is generally available on Google Kubernetes Engine. These steps address a practical barrier: customers need their software and workflows to run on TPUs before hardware specifications alone can persuade them to move.
Products Worth Considering
Bose Lifestyle Ultra Wireless Home Speaker, White Smoke
Bose Smart Soundbar 700 with Alexa, Black
Samsung Galaxy Tab S7 11-inch Android Tablet 128GB Wi‑Fi with S Pen Mystic Black
As an affiliate, we earn on qualifying purchases.
From Two Base Chips to Four
When Google introduced TPU 8t and TPU 8i in April, it divided the generation between model training and inference. Google said TPU 8t could deliver up to a 2.7-fold performance-per-dollar improvement over its previous-generation Ironwood chip for large-scale training, and claimed up to an 80% improvement for TPU 8i in inference. Those comparisons are also vendor figures, not independent assessments.
Google’s published specifications show different priorities in the two base systems. TPU 8t is listed with 216 GB of high-bandwidth memory, 128 MB of SRAM and peak FP4 compute of 12.6 petaflops. TPU 8i has 288 GB of high-bandwidth memory and 384 MB of SRAM, with peak FP4 compute of 10.1 petaflops. Google also lists greater memory bandwidth for TPU 8i. That configuration reflects the demands of serving models, where moving data can be as important as raw computing power.
Google says it developed TPU 8tx and TPU 8i Max by changing the systems around existing silicon, including packaging, memory, networking or security features. Vahdat told Bloomberg the new variants are not completely new silicon designs. The company says adapting a system can take months rather than the years involved in designing a new chip.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Performance, Prices and Access
Google has not released prices, launch regions or exact availability dates for TPU 8tx and TPU 8i Max. Although the company has stated that both are expected in 2027, it has not given a more specific schedule. Independent benchmarks are also unavailable, so customers cannot yet compare the new systems’ performance, cost or security with alternatives using outside test results.
Google has not published absolute compute, memory or bandwidth figures for the two new variants. Its comparisons describe increases relative to the base TPU 8t and TPU 8i systems, but do not establish how the systems perform across particular models or customer workloads. Nor is it yet clear how many customers will be able to access them when they launch.
Switching workloads from Nvidia GPUs to TPUs may require engineering work, even as Google expands software support. The source material also says Oracle plans to offer Google TPUs through its cloud beginning in calendar year 2027, but does not provide further details about capacity, pricing or which TPU systems will be available there.
The 2027 Availability Test
The next major milestone is the expected 2027 launch of TPU 8tx and TPU 8i Max. Google’s cloud listing currently says the systems are coming soon. Pricing, regional access and published specifications would help potential customers assess whether the specialised configurations match their needs.
Independent performance and security testing will also be needed to check Google’s claims in real workloads. Customers will be watching for details on access through Google Cloud and Oracle Cloud, as well as how well existing AI software runs on the systems. Until then, the announcement signals Google’s direction, but does not show how widely buyers will adopt the new TPUs.
Key Questions
What are Google’s two new AI chip systems?
TPU 8tx is designed for confidential inference, while TPU 8i Max is designed for workloads involving very long inputs, such as large codebases or lengthy video.
When will TPU 8tx and TPU 8i Max be available?
Google’s Amin Vahdat told Bloomberg the systems are expected in 2027. Google has not announced exact release dates or regions.
Has Google’s performance data been independently verified?
No independent benchmarks are included in the source material. The performance and latency comparisons published so far are Google’s own claims.
How do the new systems relate to TPU 8t and TPU 8i?
Google says TPU 8tx uses the same silicon as TPU 8t, with a configuration for confidential inference. TPU 8i Max places two TPU 8i chips on one package and is intended to provide more compute and memory resources for long-context workloads.
Can customers buy the new systems now?
No. Google lists all four eighth-generation TPU systems as “Coming soon”. Prices and access details for the two new variants have not been announced.
Source: ThorstenMeyerAI.com
