🔍 Read the full analysis: A Second Source To Nvidia: What Google’s AI Chip Launches Could Mean on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google announced two additions to its eighth-generation TPU lineup on November 9: TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads with very long inputs. The systems are based on existing silicon, are not yet available, and are expected in 2027; Google has not published prices or independent benchmark results.
Google announced two new systems for its eighth-generation tensor processing units on November 9, adding one aimed at confidential AI inference and another designed for models that handle very long inputs. The systems give Google Cloud customers more specialized options alongside Nvidia GPUs, but neither is available yet: Google lists its eighth-generation TPUs as coming soon, and says the two additions are due in 2027.
The new systems are TPU 8tx, which Google says is optimized for secure, confidential inference, and TPU 8i Max, which combines two TPU 8i chips in one package for workloads with large context windows. Google says the latter doubles the compute, on-chip SRAM and high-bandwidth memory bandwidth of a TPU 8i, while using the existing rack footprint. It cites examples such as AI agents analyzing a codebase or hours of video.
Google describes both as systems built from existing silicon, not wholly new chip designs. The company says TPU 8tx uses the same silicon as TPU 8t, paired with fifth-generation AMD EPYC processors. Google claims up to two times lower latency than TPU 8t with a near-zero performance penalty. Those performance statements are vendor claims; independent tests are not available.
The new systems join two base products introduced April 22: TPU 8t for training AI models and TPU 8i for inference, or generating responses from trained models. Google’s April specifications list 216 gigabytes of high-bandwidth memory for TPU 8t and 288 gigabytes for TPU 8i. For TPU 8i Max and TPU 8tx, however, Google has not published absolute specifications. Prices and deployment regions have also not been announced.
Google’s new TPUs and the Nvidia alternative
Two specialized systems expand Google’s eighth-generation TPU lineup. They could give cloud customers more options, but availability, pricing and independent performance evidence are still pending.
Two systems, distinct workloads
The additions tailor existing silicon to different AI computing needs.
TPU 8tx
Designed for secure inference. Google 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 for large context windows, such as analyzing a codebase or hours of video.
What customers need to weigh
- More choice alongside Nvidia GPUs
- Specialized options for training, inference and sensitive workloads
- Google’s TorchTPU and support for vLLM and SGLang may ease adoption
- No prices, regions or exact release date
- No independent benchmarks or absolute specifications for the new systems
- Moving software from Nvidia GPUs may require engineering work
The rollout path
Google introduced TPU 8tx and TPU 8i Max on November 9.
Pricing and cloud regions have not been announced.
Expected in 2027; buyers and independent testers can then assess real workloads.
A Second Cloud Chip Option
Google is trying to give organizations another place to obtain AI computing capacity as demand strains supply. A credible alternative could give cloud customers more choice and negotiating leverage, particularly if a company can use different processors for training, inference or sensitive workloads. But an announced product is not yet a practical substitute: buyers need access, transparent performance comparisons, and software that works with their models and infrastructure.
Switching from Nvidia GPUs to TPUs can require engineering work to adapt and test software, so chip specifications alone do not determine whether a system is useful or cheaper. Google points to TorchTPU, its open-source support for running PyTorch code and Hugging Face models on TPUs, and TPU backends for the open-source inference engines vLLM and SGLang. Those efforts may reduce adoption friction, but the announcement provides no independent evidence about how easily customers can move production workloads.
The security-focused TPU 8tx could matter to organizations handling regulated or sensitive data. Google says it protects data and model weights while they are in use. That is a description of the company’s intended security properties, not an independent assessment of the system or confirmation that it meets any particular customer’s regulatory requirements.
Products Worth Considering
Samsung Galaxy Tab A9+ 11-Inch Android Tablet, 64GB, Graphite
Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB‑C Charging, Up to 2X More Active Noise Cancelling, Transparency Mode, Adaptive
DJI Osmo Mobile 7P Gimbal Stabilizer with Mic Mini Transmitter, Infinity Black
As an affiliate, we earn on qualifying purchases.
Four Systems in One Year
Google introduced TPU 8t and TPU 8i at its Cloud Next conference on April 22, 2026. The November announcement brings the year’s eighth-generation lineup to four systems, a faster release cadence than the roughly two-year interval Bloomberg reported for earlier Google AI processor launches. Google says specialized systems can be assembled by changing packaging, memory, networking and security arrangements around existing silicon more quickly than designing a new chip.
Google’s existing performance figures also need their scope kept clear. The company claims TPU 8t provides up to a 2.7-times improvement in performance per dollar over its previous generation for large-scale training, and TPU 8i provides up to an 80% improvement for inference. These are Google’s comparisons, not independent benchmark results, and they concern the base systems rather than the two announced in November.
Google says nearly 100 customers use its TPUs, while Alphabet has described the company as operating in a supply-constrained environment. Oracle has said it will offer Google TPUs in its cloud beginning in calendar year 2027. Together, those developments point to a broader effort to make TPUs available beyond Google’s own cloud, although the new systems’ availability and commercial terms remain pending.
““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
The two systems have not been independently tested, and Google has not released absolute specifications for either one. Its latency and capacity figures are company claims, so it remains unclear how they will compare with competing hardware on real workloads, including when software, networking and operating costs are taken into account.
Google has not announced prices, specific cloud regions, or an exact release date. The company’s product page lists all four eighth-generation TPU systems as coming soon. Bloomberg reported that Vahdat put the new variants’ availability in 2027, but the source material does not specify when in that year customers can order them.
It is also unclear how much work customers will need to move applications from Nvidia GPUs, or whether the security features will satisfy particular regulators and enterprise requirements. Google’s customer count and its description of supply constraints do not establish how much capacity will be available to each customer.
The 2027 Cloud Rollout
The next milestones are Google’s release of regional availability and pricing, followed by access to systems that customers and independent testers can evaluate. Google has said the two new variants are expected in 2027; Oracle has also said it plans to offer Google TPUs through its cloud in that calendar year. Neither announcement, on its own, sets an exact launch date for these particular systems.
For prospective buyers, the practical test will be whether the TPUs are available at useful scale and deliver competitive performance and costs on their own workloads. Independent benchmarks, published specifications and customer experience will help clarify whether the specialized systems can serve as a workable alternative or complement to Nvidia GPUs.
Key Questions
What did Google announce?
Google announced TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads that need very long context windows. They extend the company’s eighth-generation TPU lineup.
When will the new TPUs be available?
Google’s AI and infrastructure chief told Bloomberg the two new systems are due in 2027. Google has not given an exact date or named availability regions.
Are Google’s performance claims independently verified?
No independent benchmark results for TPU 8tx or TPU 8i Max are available in the source material. Their stated latency and capacity gains are Google’s own claims.
Do the new systems replace Nvidia GPUs?
The announcement does not establish that. Google is offering additional AI chip options, but availability, workload performance, costs and the effort required to adapt software will affect whether customers use TPUs instead of, or alongside, Nvidia GPUs.
Source: ThorstenMeyerAI.com
