🔍 Read the full analysis: Google’s New Chips Bring AI Hardware Diversity Into Focus on ThorstenMeyerAI.com
Get hardware and tech essentials 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 TPU 8 variants on November 9: TPU 8tx, aimed at confidential computing, and TPU 8i Max, designed for workloads with very long inputs. Google says they will be available in 2027, but has not released prices, regions or independent performance results.
Google announced two new systems for its eighth-generation AI chips on November 9: the TPU 8tx, designed for confidential computing, and the TPU 8i Max, intended for AI workloads that handle very long inputs. The company says both are due in 2027, adding specialized options to a lineup that could give cloud customers an alternative to Nvidia GPUs, though neither system is available for purchase yet.
The new systems extend the TPU 8 generation Google introduced at its Cloud Next conference on April 22. That initial lineup had two products: TPU 8t for training AI models and TPU 8i for inference, or running models to generate outputs. Google said the November announcement was the first time it had introduced four TPU systems in one year.
Google describes TPU 8tx as optimized for secure, confidential inference, aimed at regulated businesses and government customers. It uses the same silicon as TPU 8t, paired with fifth-generation AMD EPYC processors. Google says the system can keep sensitive data and model weights isolated while they are in use, and claims up to two times lower latency than TPU 8t with a near-zero performance penalty. Those figures and security descriptions are Google’s claims, not independently verified results.
TPU 8i Max combines two TPU 8i chips on one package. Google says it doubles the base TPU 8i’s compute, on-chip SRAM and high-bandwidth memory bandwidth while fitting twice the silicon in the existing rack footprint. The system is intended for large context windows, such as AI agents processing an entire codebase or long video. Google has not published absolute specifications for either new system.
Google’s new chips bring AI hardware diversity into focus
Google added two specialized systems to its eighth-generation TPU lineup. Their 2027 availability could give cloud customers more options, but pricing and independent performance results are still unknown.
Two systems, distinct needs
The new variants adapt existing silicon to address security and memory demands.
TPU 8tx
Designed to isolate sensitive data and model weights during inference. Google claims up to two times lower latency than TPU 8t with near-zero performance penalty.
TPU 8i Max
Combines two TPU 8i chips on one package for workloads such as processing an entire codebase or long video.
From two systems to four
Google’s April lineup separated training and inference; November’s additions target specialized workloads.
More choice, with migration work
TPUs could offer cloud customers another source of AI computing alongside Nvidia GPUs.
- Specialized options for secure inference and long inputs
- Oracle plans to offer Google TPUs beginning in calendar year 2027
- TorchTPU, PyTorch, Hugging Face, vLLM and SGLang could reduce software friction
- Pricing, regions, capacity and launch timing are unpublished
- No independent benchmarks are available
- Migration may require software, infrastructure and workflow changes
More Options Beyond Nvidia GPUs
The announcement matters because AI computing capacity remains in high demand, and buyers often depend on Nvidia GPUs. Google is positioning its TPUs as another source of specialized computing, potentially giving cloud customers more choice in how they run training and inference workloads. The company says nearly 100 customers use TPUs; Alphabet has also described a supply-constrained environment. Those statements indicate demand, but they do not show how much capacity Google can provide or how many customers will adopt the new systems.
The two designs target distinct bottlenecks. Confidential computing is intended to address concerns about protecting sensitive information during processing, rather than only when data is stored or moving between systems. TPU 8i Max targets the memory demands of long inputs: Google’s existing TPU 8i has more memory and bandwidth than TPU 8t despite lower peak compute, reflecting the importance of moving data as well as performing calculations.
A second hardware source may improve buyer choice, but it does not make switching automatic. Companies may need to adapt software, infrastructure and staff workflows to another processor platform. Google is promoting TorchTPU support for PyTorch and Hugging Face models, along with integrations for the open-source inference engines vLLM and SGLang. These efforts could reduce friction, but the announcement does not establish how easily particular customer workloads will migrate or what their costs will be.
Products Worth Considering
Apple MacBook Pro 16.2-inch with M5 Pro, 18-core CPU, 20-core GPU, 48GB Memory, 2TB SSD, Space Black
Apple MacBook Pro 14-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 24GB Memory, 1TB SSD, Midnight
As an affiliate, we earn on qualifying purchases.
From Two TPUs to Four Systems
Google announced TPU 8t and TPU 8i in April 2026, separating the generation into a training-focused system and an inference-focused one. In its April technical material, Google listed TPU 8t at 216 GB of high-bandwidth memory and 6,528 GB/s of memory bandwidth. It listed TPU 8i at 288 GB and 8,601 GB/s, respectively, alongside different peak compute figures. The company also claimed performance-per-dollar gains over its previous Ironwood generation: up to 2.7 times for TPU 8t training and up to 80% for TPU 8i inference. These are vendor comparisons, not independent benchmarks.
The November additions are described as reconfigurations of existing silicon, not wholly new chip designs. Google AI and infrastructure chief technologist Amin Vahdat told Bloomberg that the systems were built around tasks customers requested. The company says changes to packaging, memory, networking and security can be made faster than designing a new chip from scratch. Bloomberg has reported that Nvidia remains dominant in the AI processor market and that demand for its processors exceeds supply.
Google is also expanding access beyond its own cloud: Oracle plans to offer Google TPUs in its cloud beginning in calendar year 2027, according to the source material. That could add another route for customers to use the hardware. It does not, on its own, establish availability, pricing or performance for either newly announced system.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Performance and Availability Still Pending
Neither new system is available yet. Google’s TPU page lists all four eighth-generation systems as “Coming soon,” and Vahdat told Bloomberg the two new variants would be available in 2027. Google has not announced prices, specific availability regions or a more precise launch schedule.
There are also no independent benchmarks for TPU 8tx or TPU 8i Max. Google has provided relative performance claims for the new systems, but not complete absolute specifications. The announcement does not establish how the systems will perform on customer workloads, how their costs will compare with Nvidia-based alternatives, or whether their security features meet the requirements of particular regulated customers.
The source article was drafted with help from Anthropic’s Claude and disclosed that Anthropic is a Google TPU customer, as Alphabet said on its October earnings call. It also said Google and Anthropic did not review or sponsor the article. That relationship is relevant when weighing the article’s account of the market; product claims above are attributed to Google.
Watch for 2027 Launch Details
The next concrete milestone is Google’s planned 2027 availability for TPU 8tx and TPU 8i Max. Buyers and developers will need further details on rollout timing, cloud regions, pricing, capacity and the systems’ complete specifications. Independent testing and customer experience will be needed to evaluate Google’s performance and security claims against real workloads.
Oracle’s planned TPU offering is also due to begin in calendar year 2027, according to the source material. Together, those launches could clarify whether Google’s expanded lineup becomes a practical alternative for more organizations or remains limited by availability, software changes and the engineering work required to move workloads. Until more information is published, claims about relative cost, speed and ease of migration remain unproven.
Key Questions
What did Google announce?
Google announced TPU 8tx for confidential inference and TPU 8i Max for workloads with very long inputs. They add to the TPU 8t training system and TPU 8i inference system announced in April.
When will the new TPU systems be available?
Google has said the two new variants are expected in 2027. It has not published a specific launch date, regional availability or prices.
Has Google’s performance data been independently verified?
No independent benchmark results for TPU 8tx or TPU 8i Max are included in the source material. The performance figures released so far are Google’s own claims.
Are TPU 8tx and TPU 8i Max entirely new chip designs?
No. Google says the new systems use existing silicon with changes to system design and configuration. Vahdat described them to Bloomberg as not completely new silicon designs.
Do the new TPUs mean companies can easily replace Nvidia GPUs?
Not necessarily. Google is adding another hardware option and promoting software support, but moving workloads can require engineering work. The announcement does not provide independent comparisons of cost, performance or migration effort.
Source: ThorstenMeyerAI.com
