🔍 Read the full analysis: Google’s AI Chip Lineup Doubles Down: Why More Nvidia Alternatives Matter on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
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 will be available in 2027, but neither has been independently tested, and prices and cloud regions have not been announced.
Google announced two new systems in its eighth-generation TPU lineup on Nov. 9, adding a variant for confidential computing and another designed for very long inputs. The systems, called TPU 8tx and TPU 8i Max, broaden Google’s effort to give cloud customers an alternative to Nvidia GPUs, but Google says they will not be available until 2027.
Google’s November announcement builds on the two base systems it introduced at its Cloud Next conference on April 22. TPU 8t is designed for model training, while TPU 8i is aimed at inference: running a trained model to generate responses. Google says the new 8tx and 8i Max systems adapt those existing designs for specific workloads rather than introducing entirely new silicon.
TPU 8tx targets confidential inference, including use by regulated businesses and governments. Google says it protects sensitive data and model weights while they are being used, and claims up to two times lower latency than TPU 8t with “near-zero” performance penalty. The system uses the same silicon as TPU 8t and is paired with fifth-generation AMD EPYC processors.
TPU 8i Max is aimed at large context windows, such as systems processing a codebase or hours of video. Google says it combines two low-latency TPU 8i chips in one package, doubling compute, on-chip SRAM and HBM bandwidth relative to TPU 8i. The company has not published absolute specifications for either new system; the performance descriptions are Google’s claims, not independent test results.
Google’s TPU lineup expands with two specialized systems
TPU 8tx targets confidential inference, while TPU 8i Max is designed for very long inputs. Both could give cloud customers another option alongside Nvidia GPUs.
Two systems, distinct workloads
Google says the new variants adapt existing chip designs for specific needs.
TPU 8tx
Designed to protect sensitive data and model weights during use, including for regulated businesses and governments.
TPU 8i Max
Combines two TPU 8i chips in one package, doubling compute, on-chip SRAM and HBM bandwidth relative to TPU 8i.
A broader lineup
The new variants join the base systems Google introduced in April.
Built for model training.
Aimed at inference: generating responses from a trained model.
Task-specific variants built around the existing designs.
What buyers need to weigh
A specialized alternative may improve choice and negotiating leverage, but switching takes evaluation and engineering work.
- Match long-input or sensitive-data workloads to specialized systems.
- More compute options for companies facing constrained AI infrastructure supply.
- No independent benchmarks or absolute specifications for the new systems.
- Pricing, cloud regions and real-world costs remain unannounced.
Access and adoption
Google expects availability in 2027; Oracle also plans to offer Google TPUs through its cloud during calendar year 2027.
More Options for AI Compute Buyers
The announcement matters because AI infrastructure demand has outpaced available supply, leaving many companies reliant on Nvidia GPUs. A credible additional option could give cloud customers more flexibility and negotiating leverage, especially if they can match a workload to a specialized system rather than use one general-purpose accelerator for everything.
But a second product line does not automatically mean an easy switch. Companies may need to adapt software, retrain staff and test performance before moving workloads from Nvidia hardware. Google is promoting software support for PyTorch and open-source inference tools, including vLLM and SGLang, as part of its effort to reduce those barriers. How well that works for customers at scale remains to be seen.
The specialized designs also point to different constraints in AI computing. Google says TPU 8i emphasizes memory capacity and bandwidth for inference, while 8tx focuses on protecting data during use. Those features may matter to organizations with long-input applications or strict data requirements, but buyers will need real-world comparisons, costs and availability to decide whether the systems are useful alternatives.
Popular right now
De'Longhi Rivelia Automatic Espresso Machine with Grinder and Milk Frother, Beige
Breville Bambino Espresso Machine, Brushed Stainless Steel
Breville Barista Touch Impress Espresso Machine, Noir (BES881NRE)
As an affiliate, we earn on qualifying purchases.
From Two Base Chips to Four Systems
Google introduced TPU 8t and TPU 8i in April, marking a shift from its earlier pattern of releasing a new AI processor about every two years, according to Bloomberg. The company now describes a lineup of four eighth-generation systems in one year, with variants tailored through changes to packaging, memory, networking and security around existing silicon.
Google’s published figures for the two original systems help explain the split. TPU 8t has 216 GB of high-bandwidth memory and 12.6 petaflops of peak FP4 compute, according to Google. TPU 8i has 288 GB of HBM, more on-chip SRAM and greater memory bandwidth, while its listed peak compute is 10.1 petaflops. These are vendor specifications; they do not establish how the new systems will perform in independent comparisons.
Google has also highlighted software work intended to make TPUs easier to use, including TorchTPU and integrations with inference engines. Separately, Oracle plans to offer Google TPUs through its cloud from calendar year 2027, according to the source material. That could extend access beyond Google Cloud, though deployment details have not been provided.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Pricing, Testing and Deployment Details
Availability is not immediate: Google’s TPU page lists the eighth-generation systems as “Coming soon,” and Vahdat told Bloomberg the two new variants are expected in 2027. Google has not announced prices or the regions where customers will be able to use them.
There are also no independent benchmarks for TPU 8tx or TPU 8i Max. Google has published relative claims, but not absolute specifications for the two variants. The source material does not establish how their performance, reliability, security properties or total operating costs compare with Nvidia systems under customer workloads.
It remains unclear how many customers will adopt the new variants, how much capacity Google can offer, and whether the systems will be available only through Google Cloud or through additional providers. Google says nearly 100 customers use TPUs, while Alphabet has described a supply-constrained environment; those statements do not specify how much capacity will be allocated to the new systems.
2027 Launch and Customer Tests
The next milestone is commercial availability in 2027, the timing Vahdat gave Bloomberg. Google has not yet provided launch dates, regional availability or pricing. Oracle’s planned TPU offering in calendar year 2027 is another development to watch, though details on capacity and configuration have not been disclosed.
Once customers can access the systems, comparisons based on their own workloads should offer a clearer measure than vendor claims alone. Buyers will need to weigh performance, memory capacity, security requirements, software compatibility and cost against the effort of moving work from existing GPU infrastructure. Until those details emerge, Google’s announcement signals a broader product lineup rather than a demonstrated shift in market share.
Key Questions
What did Google announce?
Google announced TPU 8tx, intended for 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 TPU systems be available?
Google’s Amin Vahdat told Bloomberg the two new variants are expected in 2027. Google has not announced exact launch dates or cloud regions.
Has Google’s performance data been independently verified?
No independent benchmarks for TPU 8tx or TPU 8i Max are cited in the source material. Google’s latency and capacity figures are company claims, and absolute specifications for the two systems have not been published.
How could these systems affect Nvidia customers?
They could give some companies another AI-computing option and potentially improve their leverage in negotiations. Switching still may require engineering work, and the systems’ real-world cost and performance compared with Nvidia GPUs remain unknown.
Source: ThorstenMeyerAI.com
