Google Unveils TPU 8tx and TPU 8i Max to Expand Custom AI Chip Portfolio Against Nvidia

Google introduced two eighth-generation custom processors, unveiling the AMD-partnered TPU 8tx for confidential workloads and the TPU 8i Max for high-memory inference.

* Silicon independence accelerates: Following the rollout of TPU 8t and 8i earlier this year, Google has released four custom v8 processor variants within 12 months to lower reliance on external suppliers. * Hardware-level data protection: Developed alongside AMD, the TPU 8tx integrates secure enclaves and encryption directly into the silicon, allowing organizations in regulated sectors to train and run inference on proprietary datasets without exposure risk. * Expanded capacity in identical footprints: The TPU 8i Max combines two 8i dies to double high-bandwidth memory to 384 GB at 17.2 TB/s bandwidth, designed specifically to serve long-context AI agents and memory-heavy workloads efficiently.

As custom hyperscaler silicon matures from internal experiments into commercial cloud infrastructure, expect competitive cloud compute pricing and tailored hardware availability to shape your team's 2027 infrastructure roadmaps.