🔍 Read the full analysis: How A Second Source For Nvidia Chips Could Affect AI Costs And Choice 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 TPU 8tx for confidential inference and TPU 8i Max for workloads with very long inputs, expanding its eighth-generation lineup to four systems. The new variants are expected in 2027, but Google has not published prices, regional availability or independent performance tests. They could give cloud customers another option alongside Nvidia GPUs, though cost and practical benefits remain unverified.
Google announced two new eighth-generation TPU systems on November 9: the TPU 8tx, designed for confidential inference, and the TPU 8i Max, aimed at workloads involving very long inputs. The additions give cloud customers more options beyond Nvidia GPUs, but neither system is available yet; Google says they are due in 2027, and has not released prices or independent performance results.
The two systems extend a generation Google first introduced at its Cloud Next conference on April 22. That launch included the TPU 8t for model training and the TPU 8i for inference—the stage when a trained model generates responses. Google said it was the first time it had introduced four TPU systems in a single year. The company’s TPU webpage lists all four as “Coming soon.”
Google describes the 8tx and 8i Max as reconfigured systems using existing silicon, rather than entirely new chip designs. According to Google, the 8tx is based on the same silicon as the 8t and is optimized to protect data and model weights while they are being processed. Google claims up to two times lower latency than the 8t, with a near-zero performance penalty. The company has not published independent test results for that claim.
The 8i Max combines two low-latency 8i chips in one package. Google says it doubles the 8i’s compute, on-chip SRAM and high-bandwidth memory bandwidth, while fitting within the existing rack footprint. The company has not released absolute specifications for either new system. Google chief technologist for AI and infrastructure Amin Vahdat told Bloomberg the two variants would be available in 2027; Google has not announced prices or cloud regions.
More TPU options,costs still unproven
Google’s new systems could give cloud customers another option alongside Nvidia GPUs. Their prices, availability and real-world benefits remain unknown.
Two specialized systems
The additions target distinct workloads, using reconfigured systems based on existing silicon.
TPU 8tx
Designed to protect data and model weights during processing. Google claims lower latency with near-zero performance penalty.
TPU 8i Max
Combines two 8i chips in one package for workloads such as agents working across codebases or lengthy video.
What more choice could mean
A second supply path may help buyers seeking computing capacity, while specialized designs could fit particular workloads.
- Another provider alongside Nvidia GPUs
- Systems tailored to confidential or long-input workloads
- PyTorch and Hugging Face support through TorchTPU
- Prices and operating costs
- Cloud regions and available supply
- Migration effort and comparable performance
Claims need real-world checks
Google has not published complete specifications or independent performance tests for either new system.
Google says the systems are expected in 2027; dates, regions and prices have not been announced.
Independent tests and customer deployments can show how they perform on actual workloads.
Cloud pricing, availability, performance and engineering work will shape any cost impact.
More Options for AI Infrastructure Buyers
A second supply path could matter to companies trying to secure enough computing capacity as demand for AI processors remains high. Nvidia remains the dominant supplier, and Bloomberg reported that demand for its processors exceeds available supply. Google says nearly 100 customers use its TPUs and describes the market as supply-constrained. More capacity and a different provider could give some buyers negotiating leverage, although the announcement does not establish that Google’s systems will be cheaper.
The specialized designs also address different costs inside AI workloads. Google says the 8tx targets organizations that need confidential processing, including regulated companies and governments. The 8i Max is intended for models that must handle large amounts of context, such as an agent working across a codebase or lengthy video. If these designs perform as described, customers could select systems for particular tasks rather than rely on one type of accelerator. That potential remains a vendor claim until the products are available and independently measured.
Choice is not only about hardware. Moving a workload from Nvidia GPUs to TPUs can require engineering work, software changes and testing. Google points to support for PyTorch and Hugging Face models through TorchTPU, along with TPU backends for the open-source inference engines vLLM and SGLang. Those tools may reduce friction, but the announcement does not quantify migration effort or show how performance compares across equivalent workloads.
Products Worth Considering
Samsung Galaxy Tab S7 11-inch Android Tablet 128GB Wi‑Fi with S Pen Mystic Black
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
Samsung Galaxy Tab A9+ 11-Inch Android Tablet, 64GB, Graphite
As an affiliate, we earn on qualifying purchases.
Google’s Shift to Specialized TPUs
Google introduced the base TPU 8t and 8i systems in April, dividing its eighth generation between training and inference. Its published specifications show different priorities: the 8t is listed with 12.6 petaflops of peak FP4 compute, 216 GB of high-bandwidth memory and 6,528 GB/s of memory bandwidth. The 8i is listed with 10.1 petaflops, 288 GB of memory and 8,601 GB/s of bandwidth. These are Google-published figures, not independent benchmarks.
The distinction matters because inference with long inputs can be limited by moving data in and out of memory, not just by raw computing power. Google says the 8i Max is intended to address that type of workload by increasing memory and compute together. The company has not yet provided its absolute specifications, so the announced doubling cannot be translated into a complete comparison of performance or operating cost.
The new lineup also reflects a faster product cadence. Google’s chip division previously released a new AI processor about every two years, Bloomberg reported. Vahdat told the outlet the new variants are not completely new silicon designs, but configurations built around customer requests; Google says changing packaging, memory, networking and security can take months rather than years. Oracle is also expected to offer Google TPUs in its cloud from calendar year 2027, giving buyers another potential route to access them.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Prices, Tests and Delivery Remain Unknown
Google has not announced prices, specific cloud regions or launch dates beyond saying the two new systems are due in 2027. It has not published complete specifications for the 8tx or 8i Max, or independent benchmarks comparing them with the base TPUs or Nvidia hardware. The company’s latency and doubled-capacity figures are vendor claims; real-world results may depend on workloads, software and system configuration.
It is also unclear how much the systems will cost to operate, how much supply Google will have, or how readily customers can move existing workloads to them. The announcement does not establish that added TPU capacity will reduce AI service prices. Any effect on customer costs will depend on availability, performance, cloud pricing and the engineering required to use the systems.
Availability and Benchmarks in 2027
The next material details are expected as Google prepares the 8tx and 8i Max for availability in 2027. Buyers will need launch dates, regional access, pricing and full specifications to compare them with existing options. Oracle’s planned TPU offering in calendar year 2027 could provide an additional way to access Google hardware outside Google Cloud, though its terms and availability have not been announced here.
Independent tests and customer deployments will show whether the new configurations deliver the claimed performance and whether they make workloads less costly or easier to run. Until those details emerge, the announcement signals more choice in Google’s AI hardware lineup, not a confirmed reduction in AI computing costs.
Key Questions
What did Google announce?
Google announced the TPU 8tx, designed for confidential inference, and the TPU 8i Max, designed for workloads with very long inputs. They join the TPU 8t for training and TPU 8i for inference.
When will the new TPU systems be available?
Google’s Amin Vahdat told Bloomberg the two new systems are expected in 2027. Google has not provided exact launch dates or cloud regions.
Will the new TPUs make AI cheaper?
That has not been established. Google has not published prices or independent performance results for the new systems, so their cost compared with Nvidia GPUs or other TPUs remains unclear.
How are the 8tx and 8i Max different?
Google says the 8tx focuses on protecting data during processing, while the 8i Max combines two 8i chips in one package to increase compute and memory capacity. Both are described as systems built from existing silicon rather than entirely new chip designs.
Can customers access Google TPUs through other cloud providers?
Oracle plans to offer Google TPUs in its cloud starting in calendar year 2027, according to the supplied source material. The terms, locations and specific TPU systems available through Oracle have not been stated.
Source: ThorstenMeyerAI.com
