🔍 Read the full analysis: Google Adds AI Chips At A Fast Pace: What A Second Nvidia Source Means 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 specialized 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. Both are based on existing silicon and are expected in 2027; prices, regions and independent performance tests have not been published.
Google announced two new systems for its eighth-generation AI chips on November 9, adding confidential-computing and long-context options to a lineup aimed at customers seeking alternatives to Nvidia processors. Google says the TPU 8tx and TPU 8i Max will be available in 2027; neither has been released for customer use or independently tested.
The additions extend a generation Google introduced at its Cloud Next conference in April. The lineup now includes TPU 8t for training AI models, TPU 8i for inference, or running trained models to generate responses, plus the new variants for confidential inference and workloads with very long inputs. Google said this is the first year it has introduced four TPU systems.
Google describes TPU 8tx as a system for secure, confidential inference, particularly for regulated businesses and government users. It is based on the same silicon as TPU 8t and paired with fifth-generation AMD EPYC processors. Google claims it can provide up to two times lower latency than TPU 8t with near-zero performance penalty. The company says the system protects data and model weights while they are being processed, not only when stored or transmitted.
TPU 8i Max combines two TPU 8i chips in one package. Google says it doubles the base system’s compute, on-chip SRAM and high-bandwidth memory bandwidth, within the existing rack footprint. It is designed for large context windows, such as AI agents that process a codebase or lengthy video. Google has not released absolute performance figures for either new system.
Google expands its TPU lineup A second route to AI compute
Two specialized eighth-generation systems target confidential inference and very long inputs. They could give customers another option beyond Nvidia, but availability, cost and independent results remain unknown.
Two new workload options
Google says both variants reconfigure existing silicon around specific customer needs.
TPU 8tx
Uses the same silicon as TPU 8t. Google says it protects data and model weights during processing.
TPU 8i Max
Combines two TPU 8i chips in one package for workloads such as processing a codebase or lengthy video.
A broader lineup
The four systems address different stages and demands of AI computing.
| System | Designed for |
|---|---|
| TPU 8t | Training AI models |
| TPU 8i | Inference: running trained models |
| TPU 8tx | Confidential inference |
| TPU 8i Max | Workloads with very long inputs |
What a second source could mean
More choice may help customers match compute to their needs and negotiate, but switching still takes work.
- Another supplier of AI compute
- Options tailored to sensitive data or memory-intensive workloads
- Oracle plans to offer Google TPUs in its cloud from calendar year 2027
- Prices, regions and firm launch dates
- Independent performance results
- Software migration effort and real-world compatibility
What buyers can verify today
The new variants have not been released for customer use or independently tested.
- Google’s performance figures are company claims, not independent benchmarks.
- Full specifications for TPU 8tx and TPU 8i Max have not been published.
- Google’s product page lists all four eighth-generation systems as “Coming soon.”
A Second Route to AI Compute
For cloud customers, Google’s expanded lineup could offer another place to source computing capacity as demand for AI processors remains high. The potential value is not simply a different chip: customers may gain more negotiating leverage and options tailored to sensitive data or memory-intensive workloads. Oracle has said it will offer Google TPUs in its cloud starting in calendar year 2027, which could make access possible beyond Google’s own cloud services.
But an additional supplier does not make switching straightforward. Customers need software and engineering work to adapt applications, and their choice depends on performance, availability, price and support. Google points to TorchTPU, an open-source project intended to run PyTorch code and Hugging Face models on TPUs, as well as TPU support for inference tools vLLM and SGLang. Those efforts may reduce some friction, but the new systems’ real-world compatibility and cost have not yet been established.
Google’s announcement also signals a faster way of tailoring its infrastructure: the two new systems are reconfigurations of existing silicon, rather than entirely new chip designs. That can let the company respond to particular customer needs without waiting for a completely new processor generation. Whether that approach gives buyers a useful alternative to Nvidia will depend on results they can measure, not vendor claims alone.
Products Worth Considering
Apple Mac mini Desktop Computer with M2 Pro, 16GB Memory, 512GB SSD, Gigabit Ethernet
FLEXISPOT Electric Adjustable Workbench, 48 x 30 in, Maple
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz
As an affiliate, we earn on qualifying purchases.
From Two Base TPUs to Four
At its April 22 Cloud Next event, Google introduced TPU 8t for training and TPU 8i for inference. Google said TPU 8t could deliver up to a 2.7-times performance-per-dollar improvement over the prior Ironwood generation for large-scale training, and claimed up to an 80% performance-per-dollar improvement for TPU 8i inference. Those figures are Google’s own comparisons, not independent benchmarks.
Google’s published specifications show different priorities for the base systems. TPU 8t is listed with 216 GB of high-bandwidth memory and 12.6 petaflops of peak FP4 compute; TPU 8i has 288 GB of that memory and 10.1 petaflops. The newer 8i system also has more on-chip SRAM and memory bandwidth, which Google says helps with inference workloads. The company has not published comparable absolute specifications for TPU 8i Max or TPU 8tx.
Google’s chip chief Amin Vahdat told Bloomberg the new systems are not entirely new silicon designs but configurations shaped around customer requests. Google’s TPU product page lists all four eighth-generation systems as “Coming soon.” The source report also says the company previously released a new AI processor about every two years, citing Bloomberg, while the 2026 rollout introduces several systems within one year.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Prices, Testing and Access Await
Neither new system is available yet, and Google has not announced list prices, cloud regions or firm launch dates. Vahdat told Bloomberg the new variants would be available in 2027, but further timing details have not been provided in the source material.
All performance figures for the additions are Google claims. No independent benchmarks are available, and the company has published ratios rather than full specifications for TPU 8tx and TPU 8i Max. It is also unclear how the systems will perform across customer workloads, how much they will cost relative to Nvidia-based options, and what software migration will require. The supplied source notes that Google describes an environment constrained by demand, but does not provide a detailed allocation or capacity forecast for these systems.
Watch for 2027 Launch Details
The next concrete developments are expected to be Google’s launch information for the two systems, including availability, pricing, supported regions and fuller specifications. Independent testing and customer deployments will be needed to check the company’s performance claims against real workloads.
Oracle’s planned offer of Google TPUs in its cloud from calendar year 2027 is another milestone for customers considering access outside Google Cloud. Until those services and the new TPU variants are available, buyers can compare Google’s published base-chip specifications and software support, but cannot yet make a direct, independently verified comparison of the new systems with Nvidia alternatives.
Key Questions
What did Google announce?
Google added TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads with very long inputs, to its eighth-generation TPU lineup.
When will the new TPU systems be available?
Google’s Amin Vahdat told Bloomberg the two additions are expected in 2027. Google has not published exact launch dates, prices or regions.
Are these entirely new chips?
No. Vahdat described them to Bloomberg as configurations that are not completely new silicon designs. Google says TPU 8tx uses the same silicon as TPU 8t, while TPU 8i Max combines two TPU 8i chips in one package.
Have Google’s performance claims been independently verified?
No independent benchmarks for TPU 8tx or TPU 8i Max are available in the source material. The performance comparisons published so far are Google’s own claims.
What could a second source of AI chips mean for customers?
It could give customers another supplier and more options for specific workloads, potentially strengthening their negotiating position. The practical benefit remains uncertain until pricing, availability, performance and the engineering needed to move applications are clear.
Source: ThorstenMeyerAI.com
