🔍 Read the full analysis: What Google’s Rapid AI Chip Expansion Could Mean For The Market 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 lineup on Nov. 9: TPU 8tx for confidential inference and TPU 8i Max for workloads with very long inputs. Google says both are due in 2027, but neither has a published price or independent benchmark, and all four eighth-generation systems remain listed as coming soon.
Google announced two additions to its eighth-generation Tensor Processing Unit lineup on Nov. 9, expanding its AI chip systems for confidential computing and workloads that process very long inputs. The new TPU 8tx and TPU 8i Max are not available yet; Google’s Amin Vahdat told Bloomberg they are expected in 2027, leaving buyers without prices or independent performance tests for now.
The new systems extend two TPUs Google introduced in April: TPU 8t, designed for training AI models, and TPU 8i, designed to run trained models and produce responses, a process known as inference. Google says the 8tx is built on the same silicon as the 8t, with additional system features aimed at protecting data and model weights while they are being used. It is intended for regulated businesses and government workloads.
Google says the 8tx can deliver up to twice the latency reduction compared with the 8t, with a near-zero performance penalty. The 8i Max combines two 8i chips in one package. Google says it doubles the 8i’s compute, on-chip memory and high-bandwidth memory bandwidth, while fitting within the existing rack footprint. Its target workloads include AI agents handling large codebases or lengthy video inputs. These are Google’s claims; independent testing has not been reported.
All four eighth-generation systems are listed by Google as “Coming soon.” Google has not announced prices, cloud regions, or absolute specifications for the two new variants. Vahdat told Bloomberg the new systems are reconfigurations of existing silicon rather than entirely new chip designs, an approach Google says can respond to customer needs faster than designing new chips.
Google expands its TPU lineup What it could mean for buyers
Google announced two specialized AI chips, but availability, pricing and independent test results are still missing. Their market impact will depend on whether customers can access and use them at scale.
Two systems, distinct jobs
The additions extend Google’s TPU lineup beyond training and general inference.
TPU 8tx
Adds protections for data and model weights in use, targeting regulated businesses and government workloads.
TPU 8i Max
Combines two 8i chips in one package for workloads such as AI agents working with large codebases or lengthy video.
TPU 8t
The existing eighth-generation system designed to train AI models.
TPU 8i
The existing system designed to run trained models and produce responses.
Potential upside and practical hurdles
A new supplier option matters only if the systems fit customer workloads and are available at a competitive cost.
- Another chip option could give cloud customers more choice and negotiating leverage.
- Confidential computing and added memory may suit specialized workloads.
- Oracle plans to offer Google TPUs in its cloud from calendar year 2027.
- Prices, cloud regions and available capacity have not been announced.
- Independent benchmarks for the two new systems are not available.
- Software adaptation and migration from Nvidia GPUs may take work.
What buyers should verify
Google’s claims are not a substitute for customer testing and deployment details.
Look for specific launch dates, cloud regions and capacity beyond the expected 2027 timing.
Wait for independent results on latency, long-input workloads and security benefits.
Evaluate Google’s PyTorch tools, TorchTPU, vLLM and SGLang against existing systems and migration needs.
More Options Beyond Nvidia GPUs
Google is seeking to make its own chips a more credible option for organizations buying large amounts of AI computing. A second supplier could give cloud customers more choice and negotiating leverage, particularly as demand for AI processors remains high. But an announced alternative does not immediately resolve supply constraints: the new TPUs are not yet available, and Google has not provided capacity or pricing details.
The specialized systems also reflect a market that is moving beyond a single measure of chip speed. Confidential inference may appeal to organizations handling sensitive data, while additional memory and bandwidth may help with long prompts and large inputs. Whether those features deliver practical benefits at competitive cost will depend on real-world performance, availability and software support, none of which can yet be fully assessed for these two systems.
For customers, changing chip platforms is not just a purchasing decision. Teams may need to adapt software, workloads and infrastructure, and the cost of that work can affect whether a nominally cheaper or faster chip is worthwhile. Google is promoting tools for running PyTorch and open-source models on TPUs, but customers will need to test how well those tools fit their existing systems.
Products Worth Considering
Bose TV Speaker (2nd Gen) Compact Soundbar
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black
Sony WH-1000XM5 Premium Noise Cancelling Wireless Headphones – Midnight Blue
As an affiliate, we earn on qualifying purchases.
From Two TPUs to Four Systems
Google introduced the TPU 8t and TPU 8i at its Cloud Next conference on April 22, 2026. The company said the 8t is designed for large-scale training and can be arranged in superpods of 9,600 chips. The 8i is aimed at inference, where memory capacity and the speed of moving data can matter as much as raw computing power.
Google’s published specifications for those base systems give the 8t 216 gigabytes of high-bandwidth memory and 12.6 petaflops of peak FP4 compute. The 8i has 288 gigabytes of high-bandwidth memory and 10.1 petaflops of peak FP4 compute, alongside more on-chip memory and memory bandwidth. Google has also claimed performance-per-dollar gains over the prior generation, Ironwood; those figures are vendor comparisons, not independent benchmarks.
The November announcement marks a quicker product cadence than Google’s earlier pattern, which Bloomberg described as releasing a new AI processor about every two years. Google says it can create different systems by changing packaging, memory, networking and security around existing silicon. The company is also developing its software ecosystem, including TorchTPU and support for inference engines vLLM and SGLang. Separately, Oracle plans to offer Google TPUs in its cloud from calendar year 2027, according to the source material.
““This is the first time we have introduced four TPU systems in a single year.””
— Google, in its Nov. 9 announcement
Prices and Test Results Still Missing
Availability and cost remain unknown. Google has not published prices or regions for the new systems, and the company’s TPU page lists all four eighth-generation systems as coming soon. The 2027 timing for the 8tx and 8i Max comes from Vahdat’s comments to Bloomberg; a more specific launch date has not been announced.
Google has not released absolute performance specifications for either new variant, beyond relative claims such as doubling certain 8i measures or reducing latency compared with the 8t. No independent benchmarks are available for the new systems. The security properties of the 8tx and the practical benefits of the 8i Max for long-context workloads will require evaluation beyond the company’s descriptions.
It is also unclear how much capacity Google will make available, whether customers can use the systems outside Google Cloud, and how easily existing workloads can be moved from Nvidia GPUs. The source material reports that demand for AI processors exceeds supply, but does not provide a quantified comparison of Google’s TPU capacity with Nvidia’s supply.
Availability Expected During 2027
The next concrete milestone is Google’s planned release of the TPU 8tx and TPU 8i Max in 2027. Buyers will be watching for launch dates, pricing, cloud-region availability and independent comparisons with Nvidia GPUs and earlier TPUs. Oracle’s planned Google TPU offering in 2027 could also give customers another route to access the chips.
Until those details emerge, customers can assess Google’s software support and compare the existing TPU generation where suitable data is available. The new systems’ market impact will depend on whether Google can deliver them at scale, demonstrate the claimed performance and security benefits, and make migration practical for customers.
Key Questions
What did Google announce?
Google announced TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads with very long inputs. They join the TPU 8t training system and TPU 8i inference system announced in April.
When will the new TPUs be available?
Google’s Amin Vahdat told Bloomberg the two new variants are expected in 2027. Google has not given a more specific date, and its TPU page lists all four eighth-generation systems as coming soon.
Has Google’s performance claims been independently tested?
No independent benchmarks for the 8tx or 8i Max are included in the source material. Claims about latency and doubled compute, memory and bandwidth are Google’s own figures.
Could the announcement reduce reliance on Nvidia?
It could give some cloud customers another chip option, but the announcement alone does not show how much capacity Google will offer or whether customers can move workloads easily. Prices, availability and real-world comparisons will shape the competitive effect.
Source: ThorstenMeyerAI.com
