Google Unveils Two More AI Chips—Could They Change Your AI Options?

Read the full analysis: Google Unveils Two More AI Chips—Could They Change Your AI Options? on ThorstenMeyerAI.com

Google announced two new eighth-generation TPU systems on November 9: TPU 8tx, designed for confidential inference, and TPU 8i Max, aimed at processing very long inputs. The systems add more specialized options to Google’s AI computing lineup, but neither is available yet; Google says they are expected in 2027.

Google’s announcement expands a generation first introduced at its Cloud Next conference in April. The lineup now comprises TPU 8t for model training, TPU 8i for inference, and the new 8tx and 8i Max variants. Google says the new systems are based on existing TPU silicon with changes to packaging and other system components, rather than entirely new chip designs. Google chief technologist Amin Vahdat told Bloomberg the variants respond to customer requests.

Google describes TPU 8tx as a system for secure, confidential inference, targeting regulated businesses and governments. The company says it protects sensitive data and model weights while they are being processed. It claims up to two times lower latency than TPU 8t with a near-zero performance penalty. Google says the system uses the same silicon as TPU 8t and pairs it with fifth-generation AMD EPYC processors.

TPU 8i Max is intended for workloads with large context windows, such as AI agents that process an entire codebase or lengthy video. Google says it combines two low-latency TPU 8i chips in one package, doubling compute, on-chip SRAM and HBM memory bandwidth compared with TPU 8i, while fitting within the existing rack footprint. The company has not released absolute performance figures for either new system.

More Options Beyond Nvidia GPUs

The announcement matters to organizations seeking alternatives to Nvidia processors, which Bloomberg describes as dominant in the AI chip market. Google’s expanded lineup could give cloud customers another option for training and running AI models, including workloads where data confidentiality or long inputs are priorities. A second supplier may also give buyers more negotiating leverage as demand for AI computing capacity remains high.

That does not mean the new systems will immediately change what customers can buy. They are not available until 2027, and Google has not announced prices, deployment regions or independent test results. Moving software from one chip platform to another can require engineering work, so stated hardware capabilities alone do not establish whether a TPU system will be a practical or less expensive choice for a particular company.

Google is also working to broaden software support. The company points to TorchTPU, which it says lets PyTorch code and Hugging Face models run natively on TPUs, as well as TPU backends for the open-source inference engines vLLM and SGLang. Wider software compatibility could reduce adoption barriers, but the announcement does not provide independent evidence of how easily customers can migrate existing workloads.

Google’s Expanding TPU Lineup

Google introduced TPU 8t and TPU 8i on April 22, 2026. The first is intended for training large models, while the second is designed for inference, or generating outputs from trained models. Google says TPU 8t supports superpods with 9,600 chips. Its April announcement claimed up to a 2.7-times improvement in performance per dollar over the previous generation for large-scale training; for TPU 8i, Google claimed up to an 80% improvement for inference. Those are company-reported comparisons, not independent benchmark results.

Google’s published figures for the base systems show different priorities. TPU 8t is listed with 216 GB of high-bandwidth memory and 6,528 GB/s of memory bandwidth. TPU 8i has 288 GB of high-bandwidth memory and 8,601 GB/s of bandwidth, as well as more on-chip SRAM. That emphasis on memory is relevant to serving long inputs, where moving data can constrain performance. Google has not provided corresponding absolute specifications for 8i Max or 8tx.

The announcement follows a faster release pace: Google said it had introduced four TPU systems in one year. Bloomberg reported that the company previously released a new AI processor about every two years. Google says it can create task-specific systems by adapting existing silicon and changing elements such as packaging, memory, networking and security, rather than waiting for a new chip design.

“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 Unverified

Independent testing is not available for either new system, and Google has not published prices or specific cloud regions where they will be offered. The company’s latency and capacity comparisons remain vendor claims until outside testing or customer results are available. The source material also says Google’s TPU page lists all four eighth-generation systems as “Coming soon.”

It is also unclear how well the systems will perform on customers’ actual workloads, how much it will cost to switch from GPUs, and whether the claimed security protections will meet particular regulatory requirements. Google has said nearly 100 customers use TPUs and described a supply-constrained environment, but those statements do not establish how much capacity will be available for the new variants.

2027 Launch Details Awaited

Google has said TPU 8tx and TPU 8i Max are expected to become available in 2027. The company has not given launch dates, pricing or regional availability, so buyers will need to wait for those details and for performance data beyond Google’s own claims.

Oracle has said it will offer Google TPUs through its cloud from calendar year 2027, according to the source material. That could add another route for customers to access the chips, though the specific configurations and service terms have not been stated here. Benchmark results, software compatibility in practice and customer deployments will help show whether the new variants offer useful alternatives for their target workloads.

Key Questions

What are Google’s two new TPU systems?

TPU 8tx is designed for confidential inference, while TPU 8i Max targets workloads that process very long inputs. They join the TPU 8t training system and TPU 8i inference system.

When will TPU 8tx and TPU 8i Max be available?

Google’s Amin Vahdat told Bloomberg the two variants are expected in 2027. Google has not announced exact release dates or cloud regions.

Are the new systems entirely new chip designs?

No, according to Google. The company says the variants use existing silicon and adjust system elements such as packaging and memory; Vahdat described them to Bloomberg as not completely new silicon designs.

Have the performance claims been independently tested?

No independent benchmarks for TPU 8tx or TPU 8i Max are provided in the source material. Google’s latency and capacity figures should be treated as vendor claims until outside tests are available.

Could the new TPUs replace Nvidia GPUs for buyers?

The announcement does not establish that. The TPUs may offer another option for specific workloads, but prices, availability, independent performance results and the engineering costs of switching remain unknown.

Source: ThorstenMeyerAI.com