🔍 Read the full analysis: Two More Google AI Chips: What They Could Mean For AI Buyers on ThorstenMeyerAI.com
Get the latest gadgets 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 additions to its eighth-generation TPU lineup on Nov. 9, 2026: TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads with very long inputs. Google says both will be available in 2027, but has not disclosed prices, launch regions or independent test results.
Google announced two new eighth-generation AI chip systems on Nov. 9: the TPU 8tx for confidential inference and the TPU 8i Max for workloads that need very long context windows, as part of a broader push to expand beyond Nvidia. The additions give cloud AI buyers more specialized alternatives to Nvidia GPUs, but neither system is available yet; Google says they are due in 2027.
The new systems extend a TPU 8 generation Google introduced in April with two products: TPU 8t, designed for model training, and TPU 8i, designed for inference—part of the infrastructure investments discussed in AI spending and memory demand. Google’s Nov. 9 announcement describes the new products as configurations built around existing silicon, rather than entirely new chip designs. Google AI and infrastructure chief technologist Amin Vahdat told Bloomberg the variants respond to customer requests and are not completely new silicon designs.
Google says TPU 8tx is optimized for secure, confidential inference, including use by regulated businesses and government agencies. The company says it protects sensitive data and model weights while they are being processed, and claims up to two times lower latency than TPU 8t with a near-zero performance penalty. 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 while fitting within the existing rack footprint.
Those performance figures are Google’s claims, not independently verified results. The company has not released absolute specifications for either new variant. Its TPU product page lists all four eighth-generation systems as “Coming soon.” Vahdat told Bloomberg the new systems would be available in 2027; Google has not announced prices or regions.
Two More Google AI Chips What They Mean for Buyers
Google added two specialized systems to its TPU 8 lineup. They could give cloud AI buyers more options, but pricing and real-world performance remain unknown.
Two systems, distinct workloads
Google describes both as configurations built around existing silicon.
TPU 8tx
Designed to protect sensitive data and model weights during processing, including for regulated businesses and government agencies.
TPU 8i Max
Combines two TPU 8i chips in one package for workloads such as agents handling large codebases or systems analyzing long video.
Google’s stated advantages
These claims have not been independently verified.
- Up to two times lower latency than TPU 8t, Google says.
- Near-zero performance penalty, according to Google.
- Double TPU 8i’s compute, on-chip SRAM and HBM bandwidth, Google says.
- Fits within the existing rack footprint.
What buyers still need
The announcement does not establish fit, cost or performance for a buyer’s own models.
Google has not published absolute specifications, prices or launch regions.
Migration may involve adapting software, testing workloads and training teams.
Independent benchmarks and workload-specific cost comparisons are not yet available.
More access may follow
Oracle has said it plans to offer Google TPUs through its cloud starting in calendar year 2027.
More Options for AI Compute Buyers
For buyers, the announcement points to a cloud market where hardware can be matched more closely to a workload: training, routine inference, confidential processing or requests that involve unusually large amounts of input. TPU 8i Max, for example, targets tasks such as agents handling large codebases or systems analyzing long video, where moving information to and from memory can matter as much as raw computing power.
A credible alternative to Nvidia hardware could also give customers more negotiating leverage and another source of capacity. Alphabet has described a supply-constrained environment, while Google says nearly 100 customers use its TPUs. Oracle has said it plans to offer Google TPUs through its cloud starting in calendar year 2027, potentially making the hardware available beyond Google Cloud.
But a second hardware option is not automatically an easy replacement. Buyers may need to adapt software, retrain teams and test performance and cost on their own models. The new products’ fit and value remain unproven outside Google, and the announcement provides no pricing or independent comparisons with Nvidia systems.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
Samsung Galaxy Tab A9+ 11-Inch Android Tablet, 64GB, Graphite
As an affiliate, we earn on qualifying purchases.
From Two TPU Systems to Four
Google introduced TPU 8t and TPU 8i at its Cloud Next conference on April 22, 2026. It described TPU 8t as a training system built for large superpods and TPU 8i as an inference system designed to respond quickly. Google claimed the base chips improved performance per dollar over the previous Ironwood generation: up to 2.7 times for large-scale training on TPU 8t and up to 80% for inference on TPU 8i. Those are vendor-reported comparisons, not independent benchmark results.
Google’s published specifications for the base systems show different priorities. TPU 8t has 216 GB of high-bandwidth memory and 128 MB of on-chip SRAM, while TPU 8i has 288 GB of high-bandwidth memory and 384 MB of SRAM. Google also lists peak FP4 compute of 12.6 petaflops for TPU 8t and 10.1 petaflops for TPU 8i. The larger memory resources in TPU 8i reflect its focus on inference, where handling data efficiently can be important for answering requests at scale.
Software support is part of the buying decision, too. Google points to TorchTPU, which it says lets PyTorch code and Hugging Face models run natively on TPUs, as well as TPU support in the open-source inference engines vLLM and SGLang. These tools could reduce migration work, but their practical effect for a specific customer will depend on its models and infrastructure.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its Nov. 9 announcement
Prices, Tests and Rollout Remain Unknown
Google has not announced prices, launch regions or specific release dates for TPU 8tx or TPU 8i Max. It has also not published their absolute specifications, including the capacity and performance figures buyers would need for a detailed comparison. The 2027 availability window comes from Vahdat’s comments to Bloomberg; the systems are not currently available for purchase.
There are no independent benchmarks for either new variant. Google’s latency and doubled-resource claims have not been verified by outside testers, and the announcement does not establish how the systems will perform across different models, software stacks or customer workloads. It is also unclear how readily existing GPU-based services could move to TPUs, or what costs and engineering work that would involve.
Watch for 2027 Launch Details
The next concrete milestones are Google’s release of pricing, regional availability and full specifications, followed by customer access and independent performance testing. Those details will help buyers judge whether the new systems offer a practical cost or performance advantage for their workloads.
Oracle’s planned 2027 offering is another date to watch: access through a second cloud provider could broaden the choice for organizations already using Oracle Cloud. Until the systems are available and tested, buyers can treat the announcement as an expansion of Google’s planned TPU lineup, rather than evidence that the new variants outperform competing hardware.
Key Questions
What are Google’s two new AI chip systems?
TPU 8tx is designed for confidential inference, while TPU 8i Max targets workloads with very long inputs and context windows. Google describes them as configurations using existing silicon.
When will TPU 8tx and TPU 8i Max be available?
Google executive Amin Vahdat told Bloomberg the two systems are due in 2027. Google has not announced exact launch dates or regions.
Has Google published independent performance tests?
No independent benchmarks for the new variants were included in the source material. Google’s claims, including lower latency for TPU 8tx and doubled resources for TPU 8i Max, have not been independently verified.
Could the new TPUs replace Nvidia GPUs for buyers?
They could provide another option for some cloud AI workloads, but the announcement does not show that they can replace Nvidia systems generally. Buyers will need pricing, workload-specific tests and information about software migration before making that comparison.
Source: ThorstenMeyerAI.com
