🔍 Read the full analysis: Could A Second Nvidia Source Change How AI Gets Built? Google’s Chips May Help 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 new variants of its eighth-generation AI processors on November 9: TPU 8tx, designed for confidential computing, and TPU 8i Max, designed for workloads with very long inputs. Google says both will be available in 2027, but has not published prices, launch regions or independent performance tests.
Google announced two new eighth-generation Tensor Processing Unit systems on November 9, adding a confidential-computing option and a higher-memory system for very long inputs to its AI chip lineup. Google says TPU 8tx and TPU 8i Max are due in 2027, giving cloud customers another potential source of AI computing alongside Nvidia GPUs, though neither system is available for purchase or independent testing yet.
The new systems build on the TPU 8t, designed for training models, and TPU 8i, designed for inference—the process of running a trained model to produce answers. Google describes TPU 8tx as optimized for secure and confidential inference, including workloads for regulated businesses and governments. It says sensitive data and model weights remain protected while in active use, and claims up to two times lower latency than TPU 8t with a near-zero performance penalty. Those figures are Google’s claims and have not been independently verified.
Google says TPU 8i Max combines two low-latency TPU 8i chips on one package. The company claims it doubles the compute, on-chip SRAM and high-bandwidth memory bandwidth of TPU 8i while fitting within the existing rack footprint. The system is intended for long-context tasks, such as processing an entire codebase or hours of video. Google has not provided absolute performance figures for either new system.
Google’s chip page lists all four eighth-generation systems as “Coming soon.” Google chief technologist for AI and infrastructure Amin Vahdat told Bloomberg the two additions would be available in 2027. Google has not announced prices or which cloud regions will offer them. The November announcement follows the introduction of TPU 8t and TPU 8i at Google Cloud Next on April 22.
Google expands its AI chips A possible second source
Two new TPU systems could give cloud customers another option alongside Nvidia GPUs. Their availability, performance and cost remain unproven.
Two new systems
Each targets a different AI workload.
TPU 8tx
Designed for secure inference, including sensitive workloads in regulated businesses and government.
TPU 8i Max
Combines two TPU 8i chips for tasks such as processing an entire codebase or hours of video.
A broader TPU lineup
Google’s eighth-generation family now covers training, inference and specialized workloads.
| System | Designed for |
|---|---|
| TPU 8t | Model training |
| TPU 8i | Inference |
| TPU 8tx | Confidential inference |
| TPU 8i Max | Very long inputs |
What customers need to weigh
More hardware choice only helps if it fits real workloads.
- Another source of AI computing capacity
- Specialized options for sensitive data and long inputs
- Support for PyTorch, Hugging Face models, vLLM and SGLang
- Prices and cloud regions are not announced
- No independent tests or absolute performance figures
- Switching from Nvidia may require engineering work
What happens next
Customers will need evidence from availability and real-world use.
Pricing, regions, release timing and technical specifications are still pending.
Google says the new TPUs are expected in 2027; Oracle is also expected to offer TPUs through its cloud that year.
Independent benchmarks and customer deployments can show how performance, compatibility and costs compare.
More Options for AI Computing
A second source of specialized AI hardware could give cloud customers more choice when securing computing capacity, particularly as demand has strained supply. The source article reports that Alphabet has described a supply-constrained environment and that Google says nearly 100 customers use its TPUs. Oracle is also expected to offer Google TPUs through its own cloud starting in calendar year 2027, according to the supplied reporting.
The practical effect will depend on more than chip specifications. Organizations that use Nvidia GPUs may need to adapt software, workflows and infrastructure before moving workloads to TPUs. Google points to support for PyTorch, Hugging Face models, vLLM and SGLang as ways to make TPU use more accessible, but buyers will still need to test compatibility, performance and costs for their own applications. Until systems are available and independently evaluated, the announcement signals a possible alternative—not a demonstrated replacement for Nvidia hardware.
Products Worth Considering
Apple MacBook Pro 2026 14.2 Laptop with M5 Pro chip, 15‑core CPU, 16‑core GPU, 24 GB Unified Memory, 1 TB SSD, Wi‑Fi 7 – Silver
Apple MacBook Pro 16.2-inch with M5 Pro, 18-core CPU, 20-core GPU, Nano-Texture Display, 48GB Memory, 1TB SSD, Space Black
Apple MacBook Pro 16-inch (2021), M1 Pro, 10-core CPU, 16-core GPU, 16GB RAM, 512GB SSD, Silver
As an affiliate, we earn on qualifying purchases.
From Two TPU Systems to Four
Google introduced its eighth-generation TPU family in April 2026, dividing it into a chip for training and one for inference. At the time, Google claimed the TPU 8t offered up to 2.7 times better performance per dollar than its previous generation for large-scale training, and that TPU 8i improved inference performance per dollar by up to 80 percent. These are vendor comparisons, not independent test results.
Google’s April specifications list TPU 8t at 216 GB of high-bandwidth memory and 12.6 petaflops of peak FP4 compute. TPU 8i is listed at 288 GB of high-bandwidth memory and 10.1 petaflops of peak FP4 compute, with more on-chip SRAM and memory bandwidth than TPU 8t. The distinction reflects different workload needs: inference involving lengthy inputs can be constrained by moving data through memory, not just by raw computing power.
The November additions are reconfigurations using existing silicon rather than entirely new chip designs, Vahdat told Bloomberg. Google says changes to packaging, memory, networking and security can be made faster than designing a new processor. The source article also discloses that it was drafted with help from Anthropic’s Claude and says Anthropic is a Google TPU customer; it says neither company reviewed or sponsored the article.
““This is the first time we have introduced four TPU systems in a single year.””
— Google, in its announcement of the expanded TPU lineup
Prices, Tests and Launch Regions
Availability details remain incomplete. Google has not published prices, specific cloud regions or firm release dates for TPU 8tx and TPU 8i Max. The 2027 timing comes from Vahdat’s comments to Bloomberg, while Google’s own TPU page describes all four eighth-generation systems as coming soon.
There are also no independent benchmarks for the two new systems. Google has announced relative performance claims for them but not absolute specifications, making it difficult to compare their speed, capacity or cost with competing hardware. The announcement does not establish how much capacity Google will have, how quickly customers can access it, or whether the chips will meet their needs better than existing options.
Availability and Independent Testing
The next milestones are further details from Google on 2027 availability, cloud regions, pricing and technical specifications. Once customers can access the systems, independent benchmarks and real-world deployments should help test Google’s claims about latency, memory capacity and performance per dollar.
For prospective users, the key questions will be whether their models and software run well on the TPU platform and whether access is sufficient for their workloads. Google’s software support and Oracle’s planned TPU cloud offering may broaden access, but the scale and timing of that access remain to be seen.
Key Questions
What did Google announce?
Google announced TPU 8tx, aimed at confidential inference, and TPU 8i Max, aimed at workloads with very long inputs. They join the TPU 8t training system and TPU 8i inference system announced in April.
When will the new TPUs be available?
Google’s Amin Vahdat told Bloomberg the two new systems are expected in 2027. Google has not announced exact release dates or cloud regions.
Have Google’s performance claims been independently tested?
No independent tests of TPU 8tx or TPU 8i Max are included in the supplied reporting. The announced performance figures are Google’s claims, and the company has not published absolute specifications for either new system.
Could the TPUs provide an alternative to Nvidia GPUs?
They could give some cloud customers another hardware option, but the announcement does not show that TPUs can replace Nvidia GPUs across workloads. Switching platforms can require engineering work, and comparisons will depend on availability, software compatibility, performance and cost.
Source: ThorstenMeyerAI.com
