🔍 Read the full analysis: Google’s AI Chip Strategy Adds Two More Chips To The Year’s Total 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 added two specialized systems to its eighth-generation TPU lineup on Nov. 9: TPU 8tx for confidential inference and TPU 8i Max for long-context workloads. Google says both are built from existing silicon and are expected in 2027; prices, rollout regions and independent performance tests have not been published.
Google announced two new AI chip systems on Nov. 9, adding the confidential-computing-focused TPU 8tx and long-context-oriented TPU 8i Max to its eighth-generation lineup. The additions bring Google to four TPU systems introduced in 2026, but none is available yet: the company’s product page lists them as coming soon, and Google executive Amin Vahdat told Bloomberg the two new versions are expected in 2027.
Google describes the TPU 8tx as a system for secure, confidential inference, aimed at regulated businesses and government customers. It uses the same silicon as the TPU 8t training system and is paired with fifth-generation AMD EPYC processors. Google says it can deliver up to twice the lower latency of TPU 8t with a near-zero performance penalty. Those figures are company claims; no independent tests have been published.
The TPU 8i Max combines two low-latency TPU 8i chips in one package. Google says the configuration doubles the TPU 8i’s compute, on-chip SRAM and high-bandwidth memory bandwidth while fitting within the existing rack footprint. The system is intended for workloads with very long context windows, such as processing a large software codebase or extended video. Google has not released absolute specifications for either new system.
Google’s Nov. 9 announcement describes the additions as new systems built around existing silicon, rather than entirely new chip designs. Vahdat told Bloomberg the versions were reconfigured for customer needs. Google has not announced prices or launch regions. Its TPU product page lists all four eighth-generation systems as coming soon, so buyers cannot yet compare availability or costs against alternatives.
Google adds two TPU systems for specialized workloads
TPU 8tx targets confidential inference, while TPU 8i Max is designed for very long context workloads. Both build on existing silicon and are expected in 2027.
Two systems, distinct workloads
The new configurations add options for customers with specialized computing needs.
TPU 8tx
Designed to protect sensitive data and model weights during active use, including for regulated businesses and government customers.
TPU 8i Max
Combines two TPU 8i chips for workloads such as processing a large software codebase or extended video.
Google’s stated system gains
The comparisons below come from Google; independent tests have not been published.
What buyers still need
The announcement describes product direction, but practical evaluation depends on details Google has not yet provided.
- Prices and specific cloud regions
- Absolute specifications for TPU 8tx and TPU 8i Max
- Customer availability and independent performance results
- Evidence on security protections and software migration
A wider route to TPU capacity
Google is also expanding software support and cloud access as it competes for AI workloads.
Google says the systems were reconfigured around customer needs rather than built as entirely new chip designs.
TorchTPU and vLLM and SGLang backends are part of Google’s software support efforts; their impact is not yet established.
Oracle plans to offer Google TPUs through its cloud beginning in calendar 2027, though capacity and terms are unknown.
A Second Route to AI Compute
The announcement adds options to Google’s effort to compete with Nvidia’s dominant AI accelerators and make its cloud infrastructure more attractive to customers seeking alternatives. Specialized systems may let buyers match hardware to a workload: confidential inference, long-context processing, training or routine model serving. That could matter for organizations whose needs are not well served by a single general-purpose configuration.
But a competing chip does not automatically give customers a practical substitute. Moving workloads from Nvidia GPUs to TPUs can require engineering work and software changes, and the benefits will depend on actual performance, availability and pricing. Google is promoting software support including TorchTPU and backends for the open-source vLLM and SGLang inference engines. Those efforts may reduce friction, but their effect for customers has not yet been established by independent comparisons of the new systems.
Supply is part of the commercial backdrop. Google says nearly 100 customers use its TPUs, while Alphabet has described a supply-constrained environment. Oracle plans to offer Google TPUs through its cloud beginning in calendar 2027, according to the source material. More routes to capacity could give some buyers additional negotiating leverage, although the scale and terms of that access remain unknown.
Popular right now
Breville Barista Express Espresso Machine with Built-In Grinder, Black Sesame
Ninja Foodi Air Fryer 10 Qt DualZone XL 6-in-1 Grey (DZ401)
Dyson Airsmooth Styling Brush
As an affiliate, we earn on qualifying purchases.
From Two TPU Systems to Four
Google introduced the first two eighth-generation systems at its Cloud Next conference on April 22, 2026. The TPU 8t is designed for large-scale model training, while the TPU 8i targets inference, the process of running a trained model to generate answers. The Nov. 9 announcement extended that split with systems tailored to confidential workloads and unusually large inputs.
Google’s published specifications for the base systems show different priorities. The TPU 8t has 216 GB of high-bandwidth memory and 12.6 petaflops of peak FP4 compute, according to the company. The TPU 8i has 288 GB of high-bandwidth memory, more on-chip SRAM and higher memory bandwidth, while listing 10.1 petaflops of peak FP4 compute. These are vendor figures, not independent benchmark results. The additional memory capacity and bandwidth in the 8i line are relevant to workloads that must keep large amounts of model or input data readily accessible.
The faster release cadence is a shift for Google. Bloomberg reported that the company’s chip unit had previously released a new AI processor about every two years. Google says this is the first year it has introduced four TPU systems. The company’s stated approach is to adapt packaging, memory, networking and security around existing silicon more quickly than it can develop a new chip design.
“This is the first time we have introduced four TPU systems in a single year.”
— Google, in its Nov. 9 announcement
Performance and Launch Details Pending
Neither TPU 8tx nor TPU 8i Max has been independently tested in the material available for this announcement. Google has not published prices, specific cloud regions, absolute specifications for the two systems or detailed customer availability. The stated performance comparisons are Google’s own claims, and the company has not provided third-party results that would allow buyers to evaluate them against Nvidia GPUs or earlier TPUs on a common basis.
The 2027 availability estimate comes from Vahdat’s comments to Bloomberg; Google’s product page continues to label all four eighth-generation systems as coming soon. The announcement also does not spell out how the 8tx’s security protections will be assessed, or what workloads and operating conditions underlie its latency comparison. The extent to which software compatibility will make migration straightforward is also not yet clear.
Availability Details Expected in 2027
Google’s next practical milestones are to provide availability dates, cloud regions and pricing for the new systems, then make them accessible to customers for evaluation. The company has indicated a 2027 launch for TPU 8tx and TPU 8i Max. Buyers will need real-world results, including independent or customer testing, to judge the advertised latency and memory gains for their own workloads.
Oracle’s planned Google TPU offering in calendar 2027 could also broaden access beyond Google Cloud, though the source material does not provide launch terms or capacity details. Until availability and performance data emerge, the announcement establishes Google’s intended lineup and product direction, not how much demand the systems will meet or how competitive they will be in practice.
Key Questions
What did Google announce?
Google announced TPU 8tx, designed for confidential inference, and TPU 8i Max, designed for workloads with very long context windows. They join the TPU 8t and TPU 8i in Google’s eighth-generation lineup.
When will the new TPU systems be available?
Google executive Amin Vahdat told Bloomberg the two new systems are expected in 2027. Google has not provided specific release dates or regions, and its product page lists the eighth-generation systems as coming soon.
Are these entirely new chip designs?
No. Google says the new systems use existing silicon with changes to the system configuration. Vahdat told Bloomberg they are not completely new silicon designs.
Has Google’s performance data been independently verified?
No independent test results are included in the source material. Google’s claims about lower latency and doubled TPU 8i Max resources should be treated as vendor figures until outside or customer testing is available.
Why might customers consider Google TPUs?
They may offer another source of AI computing capacity and hardware tailored to particular workloads. Whether they are a practical alternative will depend on availability, price, performance and the engineering effort required to move software from other platforms.
Source: ThorstenMeyerAI.com
