🔍 Read the full analysis: Could Google’s AI Chips Give Companies Another Path Beyond Nvidia? 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 announced two new systems in its eighth-generation TPU lineup on November 9: one aimed at confidential inference and another at workloads with very long inputs. The company says both will be available in 2027, but has not published prices or independent benchmark results. The announcement adds a potential alternative to Nvidia GPUs, though supply, software migration and performance comparisons remain open questions.
Google announced two new systems in its eighth-generation Tensor Processing Unit (TPU) lineup on November 9, extending its bid to give companies an alternative to Nvidia GPUs for AI computing. Google says the TPU 8tx, designed for confidential inference, and the TPU 8i Max, built for workloads with very long inputs, will be available in 2027; neither has a published price or independent performance test.
The new systems join two TPUs Google introduced at its Cloud Next conference in April: TPU 8t for training AI models and TPU 8i for inference, the process of running a trained model to generate results. Google described the November additions as specialized systems based on existing silicon, rather than entirely new chip designs. Its chief technologist for AI and infrastructure, Amin Vahdat, told Bloomberg that the variants were reconfigured in response to customer requests.
Google says TPU 8tx is intended for organizations handling sensitive workloads, including regulated businesses and governments. The company says it protects data and model weights while they are being processed, and claims up to two times lower latency than TPU 8t with a near-zero performance penalty. Google says the system uses the same silicon as TPU 8t and pairs it with fifth-generation AMD EPYC processors.
TPU 8i Max is aimed at applications that need to process large context windows, such as an AI agent working across a codebase or analyzing hours of video. Google says it combines two TPU 8i chips in one package, doubling compute, on-chip SRAM and high-bandwidth memory bandwidth compared with TPU 8i. Those figures are company claims; Google has not released absolute specifications for either new system.
Could Google’s AI Chips Give Companies Another Path Beyond Nvidia?
Google’s new TPU systems target confidential inference and very long inputs. They could expand companies’ AI computing options, but availability, pricing and performance remain unproven.
Two systems, distinct jobs
Google says the November additions adapt existing silicon for specialized workloads.
TPU 8tx
Designed to protect data and model weights during processing. Google claims up to two times lower latency than TPU 8t.
TPU 8i Max
Combines two TPU 8i chips in one package for tasks such as analyzing codebases or hours of video.
A broader TPU lineup
The new systems join April’s training and inference models.
| System | Intended workload |
|---|---|
| TPU 8t | Training AI models |
| TPU 8i | Inference |
| TPU 8tx | Confidential inference |
| TPU 8i Max | Very long inputs |
What could ease adoption
More access and familiar software could give buyers another route to AI compute.
Google points to PyTorch and Hugging Face support through TorchTPU, plus vLLM and SGLang backends.
Oracle says it plans to offer Google TPUs through its cloud in calendar year 2027.
Companies still need to adapt software, test performance and account for engineering effort.
What buyers still need to know
The announcement leaves key purchasing questions open.
- No prices or specific launch dates announced
- No cloud regions or capacity details published
- No independent benchmarks for the new systems
- Google’s latency and doubled-resource figures are company claims
A Second Route to AI Compute
For companies seeking large amounts of AI computing capacity, another cloud provider and chip platform could create more options in a market led by Nvidia. If Google can supply TPUs at scale and support the software companies use, buyers may have more leverage when negotiating access, cost and availability. Google says nearly 100 customers use TPUs, while Alphabet has described the market as supply-constrained.
Having another chip available does not automatically make switching easy. Companies may need to adapt software, test model performance and spend engineering time moving workloads from Nvidia systems. Google has highlighted support for PyTorch and Hugging Face models through TorchTPU, as well as TPU backends for the vLLM and SGLang inference engines. Those software steps may lower barriers, but the announcement alone does not show how much migration would cost or how the new systems perform in customer deployments.
The specialized designs also target two practical constraints: protecting data during processing and moving enough information through memory for long-input tasks. Those capabilities could matter to buyers whose workloads fit the systems’ intended uses. Their commercial value will depend on independent testing, availability, pricing and the effort required to run workloads on them.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight
Amazon Kindle Scribe (3rd Generation), 64GB, 11-inch Display, Graphite, with Premium Pen
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
From Two TPUs to Four Systems
Google introduced TPU 8t and TPU 8i in April 2026, marking a split between systems for training and inference. Google’s published specifications for those base models describe different trade-offs: TPU 8t has 216 GB of high-bandwidth memory and 12.6 petaflops of peak FP4 compute; TPU 8i has 288 GB of high-bandwidth memory and 10.1 petaflops. These are vendor specifications, not independent test results.
The November announcement is the first time Google has introduced four TPU systems in one year, according to the company. Bloomberg reported that Google’s chip unit had previously released a new AI processor about every two years. Google’s approach here is to adapt existing silicon with changes to packaging, memory, networking or security, which it says can be done faster than designing a new chip.
Google also points to software work intended to make its hardware more accessible. TorchTPU was open-sourced the month before the announcement, according to the source material. Separately, Oracle has said it will offer Google TPUs through its cloud starting in calendar year 2027, adding a distribution route beyond Google Cloud.
“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 Open
Google’s TPU product page lists the four eighth-generation systems as “Coming soon.” Vahdat told Bloomberg the two new variants would be available in 2027, but Google has not announced specific launch dates, cloud regions or prices. The company has also not published absolute performance specifications for TPU 8tx or TPU 8i Max.
Claims about lower latency, doubled resources and performance improvements come from Google; independent benchmarks are not available for the new systems. It is not yet clear how their results will compare with Nvidia GPUs or other cloud chips under equivalent workloads, what customers will pay, or how much capacity Google and Oracle will have when service begins. The source material also notes a conflict of interest: its original article was drafted with help from Anthropic’s Claude, and Alphabet has identified Anthropic as a TPU customer; neither Google nor Anthropic reviewed or sponsored that article.
Watch for 2027 Cloud Access
The next concrete milestone is the expected 2027 availability of TPU 8tx and TPU 8i Max. Buyers will need details on pricing, regions, capacity and supported configurations to judge whether the systems can meet their needs. Oracle’s planned TPU offering in calendar year 2027 will also test whether access extends beyond Google Cloud.
Independent benchmarks and customer deployments will help establish whether Google’s latency and memory claims hold across real workloads, and whether the software tools make migration manageable. Until those details emerge, the announcement signals a broader TPU portfolio—not proof that companies can readily replace Nvidia systems or secure equivalent capacity.
Key Questions
What did Google announce?
Google announced TPU 8tx for confidential inference and TPU 8i Max for workloads requiring very long inputs. They join TPU 8t for training and TPU 8i for inference.
When will the new TPUs be available?
Google’s Amin Vahdat told Bloomberg the two new systems are expected in 2027. Specific launch dates and cloud regions have not been announced.
Are Google’s performance claims independently verified?
No independent benchmarks for TPU 8tx or TPU 8i Max were included in the announcement. The latency and doubled-resource figures are Google’s claims.
Could the TPUs replace Nvidia GPUs?
The announcement gives companies another potential AI computing option, but it does not establish that the systems can replace Nvidia GPUs across workloads. Availability, pricing, independent performance data and migration costs remain unknown.
Source: ThorstenMeyerAI.com
