Google’s TPU 8tx protects sensitive AI workloads, while the dual-chip TPU 8i Max targets memory-hungry AI agents.