Google TPU 8tx and 8i Max announced, launch date undisclosed

Google announced TPU 8tx and 8i Max on November 9, 2026. The announcement was made at 13:00 UTC, or 22:00 in South Korea on the same day. The official description does not confirm who can currently use them or when access actually begins.

This article follows Google's official description and is current as of November 10, 2026.

8tx and 8i Max added to the eighth-generation TPU lineup

Google said it added two systems to its eighth-generation TPU lineup. TPU 8i Max combines two low-latency TPU 8i chips in a single package. Google Cloud described TPU 8tx as optimized for secure and confidential inference.

×

Google Cloud says TPU 8tx is optimized for secure and confidential inference.AI-generated image

Availability in South Korea and access start date not disclosed

The actual access start date and availability in South Korea or the Seoul region are unconfirmed. The official description also does not specify which customers can currently use the systems. The product announcement alone is not enough to conclude that they are immediately available in South Korea.

API identifiers for the new systems are unconfirmed. The published description also does not explain how to provision them by creating resources and having them allocated. This additional information is needed to connect them to development environments in South Korea.

Pricing and minimum contract and quota requirements not disclosed

Pricing and minimum contract and quota requirements for the two systems are unconfirmed. As a result, the published specifications alone are not enough to calculate actual adoption costs. Contract or cost reviews require official confirmation of the following conditions.

• Pricing and billing criteria for each system
• Minimum contract requirements
• Resource quotas and allocation conditions

8i Max's twofold figures come from the company

Google Cloud compared TPU 8i Max with TPU 8i in three areas. It stated that compute performance, SRAM, and HBM bandwidth are each twice those of TPU 8i. The figures below come from the company's description and were not presented as independent measurements.

These multiples are comparisons of individual specifications. They cannot be taken to mean that overall workloads run twice as fast. Independent test results from Artificial Analysis for the two systems could not be confirmed.

Memory capacity and supported context length also unconfirmed

The exact memory capacity and supported context length are unconfirmed. The relative SRAM multiple alone does not reveal its absolute capacity. The HBM bandwidth figure does not indicate memory capacity either.

• Memory capacity: Exact capacity figures unconfirmed
• Supported context length: Input length that can be processed unconfirmed
• Independent performance evaluation: Independent test results for the two new systems unconfirmed

Conditions to check before adoption in South Korea

When considering adoption in South Korea, first check the available regions and actual terms of access. Memory capacity and input length needed for the work must also be checked. The specification multiples provided by the company should be assessed separately from actual workload performance.