🔍 Read the full analysis: Two New AI Chips From Google: A Potential Alternative To Nvidia on ThorstenMeyerAI.com
TL;DR
Google announced two additions to its eighth-generation TPU lineup: TPU 8tx, aimed at confidential computing, and TPU 8i Max, designed for workloads with very long inputs. Google says both will be available in 2027, but has not published prices or independent performance results. The systems could give cloud customers another option alongside Nvidia GPUs, though availability and practical comparisons remain open questions.
Google announced two new eighth-generation TPU systems on November 9, expanding its AI chip lineup with one option for confidential inference and another for workloads involving very long inputs. The systems, called TPU 8tx and TPU 8i Max, are intended to give cloud customers more choices alongside Nvidia hardware, but Google says they will not be available until 2027.
TPU 8tx is aimed at governments and regulated businesses that need to process sensitive data. Google says it keeps data and model weights protected while they are being used, rather than only when stored or moving between systems. The company says the system uses the same silicon as TPU 8t and is paired with fifth-generation AMD EPYC processors. Google claims up to 2x lower latency than TPU 8t with a near-zero performance penalty; that figure has not been independently tested.
TPU 8i Max is designed for long-context work, such as AI agents processing a large codebase or extended video. Google describes it as two low-latency TPU 8i chips on one package, with double the compute, on-chip SRAM and high-bandwidth memory bandwidth of TPU 8i. The company has not published absolute performance or memory figures for either new system.
Google’s TPU product page lists all four eighth-generation systems as “Coming soon.” Google chip chief Amin Vahdat told Bloomberg the two new variants are expected in 2027. Google has not announced prices or cloud regions for them, and no independent benchmarks are available.
Google adds two TPUs for specialized AI workloads
TPU 8tx targets confidential computing, while TPU 8i Max is built for very long inputs. Both could give cloud customers another option alongside Nvidia, but availability and real-world comparisons remain open.
Two systems, distinct jobs
Google is adapting its eighth-generation lineup to different customer needs.
TPU 8tx
Designed to protect sensitive data and model weights while in use, for governments and regulated businesses.
TPU 8i Max
Designed for tasks such as analyzing a large codebase or extended video, combining two TPU 8i chips on one package.
What buyers still need
Key details are missing, so customers cannot yet judge cost, access or practical performance.
A possible route beyond Nvidia
More hardware options could improve access to compute and buyer negotiating leverage.
Google expects the systems in 2027, with no specific launch date or regions announced.
TorchTPU and support for PyTorch, Hugging Face, vLLM and SGLang may help adoption; production fit still needs checking.
Workload testing and full deployment costs will determine how the TPUs compare with Nvidia GPUs.
A Second Route to AI Compute
For companies trying to build or run AI services, additional hardware suppliers could improve access to computing capacity and give buyers more negotiating leverage. Nvidia remains a major supplier in a market where demand for AI processors has been strong. Google’s announcement signals an effort to make its own infrastructure more adaptable to different customer needs, rather than offering one system for every workload.
The practical effect will depend on whether customers can obtain the systems when needed, how their performance compares in real workloads, and what they cost. Switching software and infrastructure from Nvidia GPUs to TPUs can require engineering work. Google is promoting TorchTPU, which it says lets PyTorch code and Hugging Face models run natively on TPUs, as well as support for the vLLM and SGLang inference engines. Those tools may ease adoption, but they do not establish that migration will be simple for every customer.
The new variants also point to a strategy of adapting systems around existing chips. That could let Google respond to customer requirements faster than designing entirely new silicon, though the announcement does not by itself show how quickly the systems can be deployed at scale.
Popular right now
De'Longhi Magnifica Start Fully Automatic Espresso Machine with Milk Frother, Silver
De'Longhi Magnifica Duo Automatic Espresso Machine with Milk Frother – Silver
Breville Bambino Espresso Machine, Brushed Stainless Steel
As an affiliate, we earn on qualifying purchases.
Google’s Eighth-Generation TPU Lineup
Google introduced its eighth TPU generation at its Cloud Next conference on April 22, 2026, with two systems: TPU 8t for training large models and TPU 8i for inference, or generating answers from trained models. The November announcement added 8tx and 8i Max. Google said it had introduced four TPU systems in one year, a faster cadence than its prior pattern of releasing an AI processor about every two years, Bloomberg reported.
Google’s published specifications for the base systems show different priorities. TPU 8t has 216 gigabytes of high-bandwidth memory and 12.6 petaflops of peak FP4 compute, according to Google. TPU 8i has 288 gigabytes of high-bandwidth memory and 384 megabytes of on-chip SRAM, with 10.1 petaflops of peak FP4 compute. Google’s figures are vendor specifications, not independent test results.
Google also claims its base TPU 8t offers up to a 2.7x improvement in performance per dollar over the prior Ironwood generation for large-scale training, while TPU 8i offers up to an 80% improvement for inference. These are comparisons provided by Google; the company has not published equivalent absolute specifications for the two newly announced systems.
„This is the first time we have introduced four TPU systems in a single year.“
— Google, in its November 9 announcement
Prices, Tests and Deployment Details
Pricing, launch regions and deployment volumes have not been announced. Google has not provided independent benchmark results for TPU 8tx or TPU 8i Max, and it has not released absolute specifications for either system. The company’s latency and doubled-capacity descriptions are vendor claims, so their performance in customer workloads remains unverified.
It is also unclear how broadly the systems will be offered, or how much capacity Google will make available when they arrive. Google says nearly 100 customers use TPUs, while Alphabet has described a supply-constrained environment. Those statements point to interest and capacity pressure, but do not establish how much of that demand the new systems can meet.
The security properties of TPU 8tx will also need evaluation by prospective customers, especially organizations handling regulated data. Google has described the system’s intended protections, but the announcement does not provide an independent audit or customer test results.
The 2027 Availability Milestone
The next stated milestone is availability in 2027, as Vahdat told Bloomberg. Google has not provided a more specific launch date. Before customers can judge the systems, they will need further information on pricing, regions, capacity, technical specifications and performance testing.
Potential buyers will also be watching for evidence that software support works in their production environments and that the new systems can be integrated with existing AI services. Oracle has said it will offer Google TPUs in its cloud beginning in calendar year 2027, according to the source material, creating another possible route for customers beyond Google Cloud. The timing and terms of that offering have not been detailed here.
Until the systems are available and independently tested, Google’s announcement establishes intended use cases and vendor-stated capabilities, not a demonstrated alternative on cost or performance. Comparisons with Nvidia hardware will depend on the workloads tested, the systems’ availability and the full cost of deployment.
Key Questions
What are Google’s two new AI chip systems?
TPU 8tx targets confidential inference, while TPU 8i Max is designed for workloads with very long inputs, such as codebase analysis or extended video.
When will TPU 8tx and TPU 8i Max be available?
Google’s Amin Vahdat told Bloomberg the two systems are expected in 2027. Google has not announced a specific launch date or cloud regions.
Has Google published prices or independent benchmark results?
No. Google has not announced prices, and independent benchmarks are not available. Performance figures in the announcement are Google’s own claims.
Are the new systems entirely new chip designs?
According to Vahdat, they 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 on one package.
Do the new TPUs replace Nvidia GPUs?
The announcement does not establish that. Google’s systems could provide another option for some cloud customers, but comparisons will depend on availability, price, performance and migration costs for each workload.
Source: ThorstenMeyerAI.com
