🔍 Read the full analysis: Can Google’s New AI Chips Help Break Nvidia’s Hold On AI Hardware? on ThorstenMeyerAI.com
Get smart everyday buys 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 new versions of its eighth-generation AI chips on November 9: TPU 8tx for confidential computing and TPU 8i Max for workloads with very long inputs. Google says they are systems built around existing silicon, not new chip designs; both are expected in 2027, and no independent tests, prices or regional availability have been published.
Google announced two new eighth-generation Tensor Processing Unit systems on November 9, adding hardware for confidential AI inference and very long inputs as it seeks to give cloud customers another option alongside Nvidia GPUs. The TPU 8tx and TPU 8i Max are not available yet: Google’s chip chief told Bloomberg they are expected in 2027, and their performance claims have not been independently tested.
Google describes the TPU 8tx as a system for secure, confidential inference, aimed at regulated businesses and government users. The company says it can keep sensitive data and model weights isolated while they are being processed. It also claims up to two times lower latency than TPU 8t, with a near-zero performance penalty. Those figures are Google’s claims; independent testing is not available.
The TPU 8i Max is intended for workloads with large context windows, such as processing an entire codebase or extended video. Google says it combines two low-latency TPU 8i chips in one package, doubling compute, on-chip SRAM and high-bandwidth memory bandwidth compared with TPU 8i. The company has not released absolute specifications for either new system.
Google says both variants use existing silicon in different system configurations rather than entirely new chip designs. The company’s chief technologist for AI and infrastructure, Amin Vahdat, told Bloomberg the versions were reconfigured in response to customer requests. Google’s TPU product page lists all four eighth-generation systems as “Coming soon.”
Can Google’s new AI chips challenge Nvidia? The 2027 test
Google’s new TPU systems target confidential inference and very long inputs. They could give cloud customers another option, but their performance and practical value remain unproven.
Two systems, two bottlenecks
Google tailored each configuration to a different workload.
TPU 8tx
Designed to isolate sensitive data and model weights during processing. Google claims up to two times lower latency than TPU 8t.
TPU 8i Max
Combines two TPU 8i chips in one package for workloads such as processing an entire codebase or extended video.
A broader TPU lineup
The two new systems join Google’s training and inference options announced in April.
| System | Stated focus |
|---|---|
| TPU 8t | Training |
| TPU 8i | Inference |
| TPU 8tx | Confidential inference |
| TPU 8i Max | Very long inputs |
What buyers need to weigh
Another source of compute may help customers manage capacity and cloud terms, but switching takes work.
- Another cloud compute option alongside Nvidia GPUs
- Specialized systems for confidentiality or long-context workloads
- Support for tools including PyTorch, Hugging Face models, vLLM and SGLang
- Prices, launch regions and precise release dates
- Independent benchmarks and real-workload performance
- Software migration effort and available capacity
The evidence gap
Google’s performance statements are vendor claims; no independent tests of either new system are available.
Google says both systems are expected in 2027, without exact dates or named cloud regions.
Buyers will need to assess pricing, capacity and performance for their own applications.
Deployment experience will show whether software and migration work make TPUs practical for more organizations.
A Second Route to AI Compute
For cloud customers, more TPU options could provide an alternative to Nvidia hardware, which Bloomberg describes as dominant in the AI processor market. A second source can give buyers another way to secure computing capacity and negotiate cloud terms, especially as demand for AI processors remains high.
But an alternative chip is not automatically an easy substitute. Companies may need to adapt software, operations and model-serving systems to use TPUs, which takes engineering time. Google is promoting support for tools including PyTorch, Hugging Face models, vLLM and SGLang, but the new hardware has not yet been released for outside evaluation. The practical value will depend on availability, pricing, performance on real workloads and the effort required to move applications.
The two variants also reflect a shift toward tailoring systems to specific bottlenecks. TPU 8tx targets confidentiality during processing; TPU 8i Max targets memory-intensive, long-context tasks. That specialization may matter to buyers whose needs do not match a general-purpose accelerator, but Google has yet to show independent results confirming its stated benefits.
Products Worth Considering
Apple Mac mini (2026), M6 12-core CPU and GPU, 24GB Memory, 1TB SSD
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and 12MP Center Stage Camera (2026, Silver)
Dell XPS Tower 8960 Desktop, Intel Core i7-13700, NVIDIA GeForce RTX 4060, 16GB DDR5, 512GB SSD
As an affiliate, we earn on qualifying purchases.
Google’s Eighth-Generation TPU Lineup
Google introduced the first two eighth-generation systems, TPU 8t for training and TPU 8i for inference, at its Cloud Next conference on April 22, 2026. The November announcement adds 8tx and 8i Max, bringing the total to four systems in one year. Google said it had not previously introduced four TPU systems in a single year; Bloomberg reported that the company had earlier released a new AI processor about every two years.
Google’s April specifications show different priorities in the two base systems. TPU 8t is listed with 216 GB of high-bandwidth memory, 128 MB of on-chip SRAM and 12.6 petaflops of peak FP4 compute. TPU 8i is listed with 288 GB of high-bandwidth memory and 384 MB of SRAM, alongside 10.1 petaflops of peak FP4 compute. These are Google-published specifications, not independent benchmark results.
Google has also claimed performance-per-dollar improvements for the base chips over its prior generation, Ironwood: up to 2.7 times for TPU 8t in large-scale training and up to 80% for TPU 8i in inference. Those comparisons are Google’s figures and concern the earlier systems, not independent tests of the new variants.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Testing, Pricing and Availability
Several details needed to compare the systems with Nvidia hardware remain unknown. Google has not announced prices, launch regions or precise release dates for TPU 8tx or TPU 8i Max. The company has published relative claims for the new systems but no absolute specifications for them, and no independent benchmarks are available.
It is also unclear how much work customers will need to move existing AI applications to the new hardware, or how the systems will perform across different models and real-world workloads. Google says nearly 100 customers use TPUs, while Alphabet has described a supply-constrained environment; those statements do not establish how much capacity the new variants will provide. Google’s article was drafted with assistance from Anthropic’s Claude, and the author disclosed that Anthropic is a Google TPU customer. Google and Anthropic did not review or sponsor the article, according to the disclosure.
The 2027 Availability Test
Google’s next stated milestone is making the two new systems available in 2027, according to Vahdat’s comments to Bloomberg. The company has not given a more specific date or named the cloud regions where customers will be able to use them. Oracle has said it will offer Google TPUs in its own cloud beginning in calendar year 2027, another possible route for customers seeking access.
Once the systems are available, buyers will be able to assess pricing, capacity and performance against their own workloads. Independent measurements and deployment experience will help show whether the claimed latency and memory advantages translate into useful results—and whether the software and migration work make TPUs a practical alternative for more organizations.
Key Questions
What did Google announce?
Google announced TPU 8tx, aimed at confidential inference, and TPU 8i Max, aimed at workloads with very long inputs. Both are additions to its eighth-generation TPU lineup.
When will TPU 8tx and TPU 8i Max be available?
Google’s Amin Vahdat told Bloomberg the two systems are expected in 2027. Google has not published exact launch dates or cloud regions.
Are the new systems new chip designs?
Google says they are configurations built around existing silicon, rather than entirely new chip designs. TPU 8tx is based on the same silicon as TPU 8t; TPU 8i Max combines two TPU 8i chips in one package, according to the company.
Have the performance claims been independently verified?
No independent benchmarks for TPU 8tx or TPU 8i Max are available in the supplied material. Google’s statements about lower latency and doubled compute, SRAM and memory bandwidth remain vendor claims.
Could Google’s TPUs replace Nvidia GPUs?
The announcement does not establish that TPUs will replace Nvidia GPUs. They could offer some cloud customers another option, but availability, price, workload performance and the engineering needed to move software remain open questions.
Source: ThorstenMeyerAI.com
