🔍 Read the full analysis: Four Google AI Chips In One Year: A Look At The Nvidia Alternative on ThorstenMeyerAI.com
Get the latest gadgets 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 added two specialized systems to its eighth-generation TPU lineup on Nov. 9, bringing its total to four TPU systems introduced in 2026. The company says TPU 8tx targets confidential computing and TPU 8i Max supports very long inputs, but both are listed as coming soon and are expected in 2027. Pricing and independent performance tests have not been published.
Google announced two new AI chip systems on Nov. 9, adding confidential-computing TPU 8tx and long-context TPU 8i Max to its eighth-generation lineup. The additions bring Google to four TPU systems introduced in 2026, giving cloud customers more alternatives to Nvidia GPUs, though Google says the new variants will not be available until 2027.
Google’s first two eighth-generation systems, TPU 8t for training and TPU 8i for inference, were introduced at its Cloud Next conference on April 22. The Nov. 9 announcement adds variants aimed at narrower workloads. Google’s TPU product page lists all four systems as “Coming soon”; the company has not published prices or availability by region.
Google says TPU 8tx is designed for confidential inference in settings such as regulated businesses and government. It uses the same silicon as TPU 8t, paired with fifth-generation AMD EPYC processors. Google says it protects data and model weights while they are being used, and claims up to two times lower latency than TPU 8t with a near-zero performance penalty. Those figures are company claims, not independent test results.
TPU 8i Max combines two TPU 8i chips in one package. Google says the system offers double the compute, on-chip SRAM and HBM bandwidth of TPU 8i, within the existing rack footprint. It is intended for workloads with very long context windows, such as processing a codebase or extended video. Google has not released absolute performance or memory specifications for either new variant.
Four Google AI Chips In One Year A Look At The Nvidia Alternative
Google introduced two specialized systems for confidential inference and very long inputs. They expand customer options, but availability and independent comparisons are still ahead.
Two new systems, specific workloads
Both variants build on existing TPU silicon and target distinct customer needs.
TPU 8tx
Designed to protect data and model weights while in use. Google claims lower latency with near-zero performance penalty.
TPU 8i Max
Combines two TPU 8i chips in one package for workloads such as processing a codebase or extended video.
How the lineup expanded
Google added two specialized variants to its training and inference systems.
What buyers still need
The announcement describes potential capabilities, but leaves key purchasing questions open.
- Pricing and regional availability
- Independent performance comparisons
- Launch capacity and supply
- Evidence of software migration effort
A Second Route to AI Compute
The announcement gives cloud buyers another potential source of AI computing at a time when demand for accelerators is high. Google says nearly 100 customers use its TPUs, while Alphabet has described a supply-constrained environment. Bloomberg has reported that demand for Nvidia processors exceeds available supply. Google’s expansion could give some customers more choice and negotiating leverage if the new systems meet their needs and become available at scale.
That does not mean TPUs are an easy substitute for Nvidia GPUs. Moving workloads can require engineering work, software changes and testing, and buyers need dependable comparisons of cost and performance before committing. Google points to support for PyTorch and open-source inference software, including vLLM and SGLang, but the usefulness of that support will depend on how well specific applications run. The new systems have not been independently benchmarked.
The two variants also show Google pursuing specialization through system design rather than a wholly new chip each time. The company says the additions use existing silicon with changes to packaging, memory, networking or security. That approach may let Google respond to distinct customer requirements faster than designing a new processor, but the announcement alone does not establish delivery volumes, reliability or total cost.
Popular right now
De'Longhi Magnifica Start Fully Automatic Espresso Machine with Milk Frother, Silver
Dyson Airsmooth Styling Brush
Apple AirPods Pro (1st Generation) with MagSafe Charging Case
As an affiliate, we earn on qualifying purchases.
From Two Systems to Four
Google introduced TPU 8t and TPU 8i on April 22, dividing the generation between large-scale model training and inference, or running a trained model to produce responses. Google says TPU 8t supports superpods of 9,600 chips. Its published specifications list 216 GB of high-bandwidth memory and 12.6 petaflops of peak FP4 compute. TPU 8i is designed around inference and memory capacity, with 288 GB of high-bandwidth memory and 10.1 petaflops of peak FP4 compute.
The two systems reflect different workload demands: training emphasizes processing power at scale, while inference can be constrained by moving data quickly, particularly for long inputs. Google claims the base TPU 8t delivers up to 2.7 times better performance per dollar than the previous generation for large-scale training, and says TPU 8i improves that measure by up to 80% for inference. These are Google’s comparisons; independent tests of the new generation are not included in the source material.
The November announcement is a change in release pace as well as product range. Google said it had introduced four TPU systems in one year. Bloomberg reported that Google had previously released a new AI processor about every two years. Google’s Amin Vahdat, its chief technologist for AI and infrastructure, told Bloomberg the two latest systems are not entirely new silicon designs but configurations made in response to customer requests.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its November 9 announcement
Prices, Tests and Supply Remain Unknown
Neither new system is available yet, and Google has not provided pricing, regional availability or expected capacity. Vahdat told Bloomberg the variants would be available in 2027, but the company’s product page gives no more specific release schedule. Google has also not published absolute specifications for TPU 8i Max or TPU 8tx.
Performance claims remain unverified outside Google. No independent benchmark results are provided for the new variants, so buyers cannot yet compare their speed, cost or power use against Nvidia products on a common basis. Google’s claims about confidential processing and latency also have not been independently tested in the source material. It is unclear how many systems will be available at launch or whether demand will exceed supply.
There is a further practical unknown: how easily customers can move existing software and workloads onto the TPUs. Google has announced software support and points to open-source integrations, but the announcement does not quantify migration effort or show results across a broad range of applications. The source article also disclosed that it was drafted with Anthropic’s Claude and that Anthropic is a Google TPU customer; it said neither company reviewed or sponsored the article.
Availability Moves to 2027
The next confirmed milestone is expected availability in 2027 for TPU 8tx and TPU 8i Max, according to Vahdat’s comments to Bloomberg. Google has not announced exact launch dates, prices or cloud regions. Potential buyers will need those details, along with independent measurements, to assess whether either system fits their workloads.
Google TPUs are also expected to become available through Oracle Cloud Infrastructure in calendar year 2027, according to the source material. That would give customers a route beyond Google’s own cloud, although the timing, product configurations and commercial terms are not specified here. Until launch details and testing emerge, the new systems remain announced alternatives rather than options customers can order.
Key Questions
What did Google announce on November 9?
Google announced TPU 8tx, aimed at confidential inference, and TPU 8i Max, designed for workloads with very long inputs. They join the TPU 8t training system and TPU 8i inference system introduced in April.
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 dates or cloud regions, and its TPU page lists them as coming soon.
Has Google published prices or independent benchmark results?
No. Google has not published prices or independent performance tests for the new variants. Its performance comparisons are vendor claims, and absolute specifications for TPU 8tx and TPU 8i Max have not been released.
How do the four TPU 8 systems differ?
TPU 8t is for training and TPU 8i for inference. Google positions TPU 8tx for confidential inference and TPU 8i Max for long-context workloads that benefit from more compute and memory bandwidth.
Will Google TPUs be available outside Google Cloud?
The source material says Oracle plans to offer Google TPUs in its cloud starting in calendar year 2027. Product details and commercial terms for that offering have not been specified.
Source: ThorstenMeyerAI.com
