🔍 Read the full analysis: A New AI Chip Option? Google’s Latest Launches Put Nvidia In Focus on ThorstenMeyerAI.com
Get audio and creator gear 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 additional versions of its eighth-generation AI chips: TPU 8tx, aimed at confidential computing, and TPU 8i Max, designed for workloads with very long inputs. Google says both are due in 2027; pricing, availability by region and independent performance results have not been published. The expanded lineup gives cloud customers another potential option alongside Nvidia GPUs, but its commercial impact remains untested.
Google announced two new versions of its eighth-generation tensor processing units on November 9: the TPU 8tx for confidential computing and the TPU 8i Max for workloads with very long inputs. The systems expand Google’s options for customers seeking alternatives to Nvidia GPUs, but neither 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. The 8t is intended for training AI models, while the 8i is designed for inference: running a trained model to produce answers. Google described the November additions as specialized systems using existing silicon, rather than entirely new chip designs. Google’s AI and infrastructure chief technologist, Amin Vahdat, told Bloomberg that the variants were reconfigured in response to customer requests.
Google says the TPU 8tx is designed for secure and confidential inference, including use by regulated businesses and governments. The company says it protects data and model weights while they are being processed, and claims up to two times lower latency than the TPU 8t with a near-zero performance penalty. Google says the system uses the same silicon as the 8t and is paired with fifth-generation AMD EPYC processors.
The TPU 8i Max is intended for applications that need to process large amounts of input, such as an AI agent working across a codebase or analyzing hours of video. Google says it combines two TPU 8i chips on one package, doubling compute, on-chip SRAM and high-bandwidth memory bandwidth compared with the 8i, within the existing rack footprint. Those figures and performance claims come from Google; independent testing has not been reported.
Google expands its TPU lineup Nvidia remains in focus
Two specialized eighth-generation TPUs target confidential inference and very long inputs. Their commercial impact remains untested.
Two systems, different needs
The additions extend Google’s eighth-generation lineup alongside the TPU 8t and TPU 8i.
TPU 8tx
Designed for secure inference, including workloads for regulated businesses and governments.
TPU 8i Max
Targets tasks such as working across a codebase or analyzing hours of video.
What buyers could gain
A second supplier may expand choice and negotiating leverage for cloud customers.
- Hardware tailored to different workloads
- More options alongside Nvidia GPUs
- TorchTPU support for PyTorch and Hugging Face models
- Pricing and regional availability
- Software compatibility and migration effort
- Measured performance against Nvidia
Google’s stated specifications
The company describes the new systems as specialized configurations using existing silicon.
| System | Google’s claim | Evidence status |
|---|---|---|
| TPU 8tx | Up to two times lower latency than 8t, with near-zero performance penalty | Google claim; no independent results reported |
| TPU 8i Max | Two 8i chips on one package, doubling compute, on-chip SRAM and memory bandwidth | Google claim; full specifications not published |
What happens next
Availability and third-party results will determine whether the new options fit real workloads.
Google says both systems are due in 2027; exact dates, prices and regions are not announced.
Oracle plans to offer Google TPUs through its cloud beginning in calendar year 2027.
Independent benchmarks and customer results can test performance, security and adoption claims.
A Wider Choice for AI Cloud Buyers
The expanded lineup matters because companies building AI services have faced strong demand for computing capacity, with Nvidia GPUs remaining a leading option. A second supplier could give cloud customers more choices and negotiating leverage, particularly if Google’s systems deliver the performance and availability it promises.
That does not mean buyers can switch without cost. Moving workloads between GPU and TPU systems can require engineering work, and performance depends on software support as well as hardware. Google points to its open-source TorchTPU support for PyTorch and Hugging Face models, plus TPU backends for the vLLM and SGLang inference engines. These tools may ease adoption, but the announcement provides no independent comparison showing how easily a particular customer could move its applications.
The two variants also target different constraints. Confidential computing is aimed at customers concerned about protecting data during processing, while the 8i Max addresses workloads that need more memory and bandwidth for long contexts. If the systems become available as described, customers may be able to choose hardware around their workload rather than relying on one general-purpose configuration. Whether that becomes a meaningful alternative to Nvidia will depend on cost, access, software compatibility and measured performance.
Products Worth Considering
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
Jackery Solar Generator 1000 V2 with 100W Solar Panel, 1024Wh LiFePO4 Battery
Jackery Explorer 300 Portable Power Station with SolarSaga 100W Solar Panel
As an affiliate, we earn on qualifying purchases.
Google’s Fourth-Generation TPU Push
Google introduced its eighth-generation TPU family in April, splitting the generation between training with the 8t and inference with the 8i. The company said the 8t is designed for superpods containing 9,600 chips. For the two base systems, Google published specifications including 216 GB of high-bandwidth memory for the 8t and 288 GB for the 8i. The 8i has more on-chip SRAM and memory bandwidth, a design choice Google links to the demands of inference workloads.
Google also reported performance-per-dollar gains over the previous generation: up to 2.7 times for large-scale training on the 8t and up to 80% for inference on the 8i. These are Google’s own comparisons, not independent test results. Google has not published absolute specifications for the 8tx or 8i Max, offering ratios for the latter and a latency comparison for the former.
The November announcement marks a faster release cadence than Google’s historical pattern. Google said it introduced four TPU systems in one year; Bloomberg reported the company previously released a new AI processor about every two years. Google says changes to packaging, memory, networking and security can be made faster than designing new silicon. The company also says nearly 100 customers use TPUs. Separately, Oracle plans to offer Google TPUs through its cloud beginning in 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 November 9 announcement
Pricing, Access and Test Results
Prices, rollout regions and specific customer availability have not been announced for either new system. Google’s TPU page lists the eighth-generation systems as coming soon, and Vahdat told Bloomberg the two additions are expected in 2027. The announcement does not give a more precise release date.
There are also no independent benchmarks for the 8tx or 8i Max, and Google has not released their full specifications. Its claims about latency, doubled resources and performance-per-dollar have not been independently tested in the material available. It remains unclear how the systems will compare with current Nvidia offerings on representative workloads, what they will cost, or how much engineering customers will need to run existing applications on them. The article source also discloses that it was drafted with assistance from Anthropic’s Claude; Anthropic is a Google TPU customer, and the source says neither company reviewed or sponsored the article.
The 2027 Availability Test
The next milestone is Google’s planned 2027 release of the TPU 8tx and TPU 8i Max. Buyers will need details on pricing, cloud regions, capacity and full specifications to judge whether either system fits their workloads. Independent benchmarks and early customer results would help test Google’s performance and security claims.
Oracle’s planned Google TPU offering, also expected to begin in calendar year 2027, could widen access beyond Google’s own cloud. Until the systems are available and tested, the announcement shows how Google is extending its TPU lineup, but does not establish how much business it can draw from Nvidia or how quickly customers will adopt the new configurations.
Key Questions
What did Google announce?
Google added TPU 8tx, aimed at confidential inference, and TPU 8i Max, aimed at workloads with very long inputs, to its eighth-generation TPU lineup.
When will the new Google TPUs be available?
Google’s Amin Vahdat told Bloomberg that the two new systems are expected in 2027. Google has not given an exact launch date or listed rollout regions.
Are the new TPUs independent alternatives to Nvidia GPUs?
They are another AI-computing option from Google, but their performance against Nvidia hardware has not been independently established. Buyers would also need to weigh price, availability, software compatibility and the work required to move applications.
What is the difference between TPU 8tx and TPU 8i Max?
The 8tx is designed for confidential computing, with Google saying it protects data while in use. The 8i Max is designed for long-context workloads and, according to Google, doubles several resources of the TPU 8i.
Source: ThorstenMeyerAI.com
