🔍 Read the full analysis: The Case For A Second Nvidia Source Starts With Google’s New AI Chips 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 systems on November 9: TPU 8tx for confidential inference and TPU 8i Max for workloads with very long inputs. Google says both are due in 2027, but neither is available for testing or purchase, and independent performance data and pricing have not been published.
Google announced two additions to its eighth-generation Tensor Processing Unit (TPU) lineup on November 9: the TPU 8tx, designed for confidential inference, and the TPU 8i Max, designed to handle very long inputs. The systems widen Google Cloud’s AI-computing options as companies seek alternatives to Nvidia GPUs, but Google says they will not be available until 2027.
The two systems extend a product family Google first introduced on April 22, 2026. That launch included TPU 8t for training AI models and TPU 8i for inference, the process of using a trained model to generate answers. The November announcement brings the generation to four systems. Google said it was the first time it had introduced four TPU systems in one year.
Google describes TPU 8tx as a system for running AI models while keeping sensitive data and model weights isolated during processing. It is based on the same silicon as TPU 8t and paired with fifth-generation AMD EPYC processors. Google claims it can provide up to twice the lower latency of TPU 8t with a near-zero performance penalty; those figures are vendor claims, not results from independent testing.
TPU 8i Max combines two TPU 8i chips on one package. Google says the system doubles the base TPU 8i’s compute, on-chip SRAM and high-bandwidth memory (HBM) bandwidth, within the existing rack footprint. The company is targeting workloads involving large context windows, such as processing a codebase or extended video. Google has not released absolute performance figures for either new system.
Google expands its TPU lineup as Nvidia alternatives draw interest
Two specialized AI chips are expected in 2027. They offer another potential route to AI compute, but performance, pricing and availability remain unverified.
Two systems, distinct workloads
Google added specialized systems built around its existing eighth-generation TPU family.
TPU 8tx
Designed to isolate sensitive data and model weights during processing. Google claims up to twice the lower latency of TPU 8t with near-zero performance penalty.
TPU 8i Max
Combines two TPU 8i chips on one package for workloads such as processing a codebase or extended video.
What Google says changes
The new variants tailor existing silicon to different customer needs.
| System | Google’s stated design | Evidence status |
|---|---|---|
| TPU 8tx | Confidential inference; paired with fifth-generation AMD EPYC processors | Latency figures are vendor claims |
| TPU 8i Max | Doubles TPU 8i compute, SRAM and HBM bandwidth | No absolute performance figures published |
A second route, with switching costs
More accelerator options may help cloud customers plan capacity, but a move from Nvidia GPUs can take engineering work.
- Another option amid reported Nvidia supply constraints
- Google promotes TorchTPU, vLLM and SGLang support
- Will software and workflows fit the TPU environment?
- How will the systems compare on real workloads?
What buyers still need
Key details are not available yet.
- Pricing and exact availability dates
- Launch regions and customer access terms
- Independent benchmarks and complete specifications
- Measured migration effort and comparisons with Nvidia products
A Second Route to AI Compute
For cloud customers, another accelerator option could provide more choice in capacity and workload design at a time when demand for AI processors is high. Bloomberg has reported that Nvidia’s supply has struggled to meet demand. Google says nearly 100 customers use TPUs, while Alphabet has described its AI computing environment as supply-constrained.
More options do not automatically mean lower costs or an easy switch. Companies may need to adapt software, workflows and infrastructure when moving from Nvidia GPUs to TPUs. Google is promoting tools including TorchTPU, which it says allows PyTorch code and Hugging Face models to run natively on TPUs, as well as TPU support for the open-source inference engines vLLM and SGLang. Whether those efforts make migration practical for a given customer will depend on workload requirements and measured results.
The new systems also show a way for Google to differentiate its cloud offering without introducing entirely new chip designs. Specializing packaging, memory, networking and security around existing silicon may let the company respond to different customer needs faster. That is Google’s rationale, not independent evidence that the approach will deliver a particular cost or performance advantage.
Products Worth Considering
Apple MacBook Pro 16-inch (2021), M1 Pro, 10-core CPU, 16-core GPU, 16GB RAM, 512GB SSD, Silver
Apple MacBook Pro 14-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)
As an affiliate, we earn on qualifying purchases.
From Two TPUs to Four
Google’s April 2026 launch divided its eighth-generation lineup between training and inference. The company says TPU 8t can be assembled into superpods of up to 9,600 chips. For the base systems, Google lists TPU 8t with 216 GB of HBM and 12.6 petaflops of peak FP4 compute; TPU 8i has 288 GB of HBM and 10.1 petaflops. Those are published specifications, not independent benchmarks.
The memory emphasis in TPU 8i Max reflects a challenge in serving large AI models: moving information into and out of memory can constrain performance, particularly when models process long inputs. Google says the Max system doubles TPU 8i’s compute, SRAM and HBM bandwidth, but has not provided the absolute specifications needed for a direct outside comparison.
Google’s chief technologist for AI and infrastructure, Amin Vahdat, told Bloomberg that the additions are not entirely new silicon designs, but systems reconfigured for tasks customers requested. Google’s announcement also lands as Oracle plans to offer Google TPUs through its cloud starting in calendar year 2027, giving customers a potential route to the chips outside Google Cloud.
“This is the first time we have introduced four TPU systems in a single year.”
Pricing and Performance Remain Unknown
Neither TPU 8tx nor TPU 8i Max is available yet. Google’s TPU page lists the eighth-generation systems as coming soon, and Vahdat told Bloomberg that the two additions are expected in 2027. Google has not announced prices, launch regions or exact availability dates.
There are also no independent benchmarks for the new variants. Google’s latency and capacity comparisons are company claims, and it has not published complete specifications for either system. The source material also does not establish how the systems will perform against comparable Nvidia products on equivalent workloads, or how much engineering work a customer would need to migrate.
Google disclosed that Anthropic, which helped draft the source article, is a TPU customer, citing Alphabet’s October earnings call. The source article says neither Google nor Anthropic reviewed or sponsored it. That relationship is relevant when weighing the article’s discussion of TPU demand and competition; it does not independently verify Google’s product claims.
Availability and Customer Testing
The next concrete milestone is Google’s expected 2027 rollout of TPU 8tx and TPU 8i Max. Buyers will need details on pricing, cloud regions and access terms to judge whether the systems fit their needs. Oracle’s planned TPU cloud offering is also expected to begin in calendar year 2027, though service details have not been provided here.
Once systems become available, customer trials and independent benchmarks can test Google’s claims against real workloads, including confidential inference and long-context processing. Until then, the new variants show Google’s intended product direction, not demonstrated performance in general availability.
Key Questions
What did Google announce?
Google announced two additions to its eighth-generation TPU family: TPU 8tx for confidential inference and TPU 8i Max for workloads that need to process very long inputs.
When will the new TPUs be available?
Google’s Amin Vahdat told Bloomberg the two systems are expected in 2027. Google has not given exact launch dates or regions.
Has Google’s performance data been independently verified?
No independent benchmarks for the new variants are included in the source material. Google’s stated latency and capacity comparisons should be treated as vendor claims until tested independently.
Do the new TPUs replace Nvidia GPUs?
The announcement adds another option; it does not establish that customers can readily replace Nvidia GPUs. A switch may require software and engineering changes, and comparative performance and costs are not yet known.
Source: ThorstenMeyerAI.com
