Google Unveils TPU 8tx and TPU 8i Max: Focused on Confidential Computing and Doubled Memory Bandwidth

Google is expanding its eighth-generation TPU lineup with two new hardware systems: TPU 8tx for confidential computing and TPU 8i Max designed specifically for long-context inference. TPU 8tx uses the same silicon as the TPU 8t training chip paired with a custom 5th Gen AMD EPYC Turin host to protect data in use within a hardware-isolated environment, while adopting a Boardfly topology to reduce multi-step agentic inference latency.