Google Adds TPU 8tx, 8i Max for Agents

Google is attacking Nvidia from two practical angles: data isolation for sensitive workloads and memory density for long-context agents.

• –TPU 8tx pairs TPU 8t silicon with custom AMD CPU hosts and confidential computing; Google claims up to 2x lower latency than TPU 8t.
• –TPU 8i Max combines two TPU 8i chips, doubling compute, SRAM, and HBM bandwidth within the same rack footprint.
• –TorchTPU and native vLLM/SGLang support could reduce the software friction that has historically limited TPU adoption.
• –Pricing, regions, and general availability remain unspecified, making independent benchmarks and deployment economics the key open questions. TestingCatalog