Google Targets NVIDIA with Dual TPU Strategy at Cloud Next 2026

On April 22, 2026, at its Cloud Next event, Google Cloud unveiled its eighth-generation Tensor Processing Units (TPUs), signaling a stronger push into the AI hardware race. According to TechCrunch, the new lineup introduces two specialized chips: TPU 8t for large-scale model training and TPU 8i for inference, reflecting a growing industry shift toward workload-specific silicon.

Google’s official announcement highlights major improvements in performance and scalability. The TPU 8t is designed to scale across massive clusters with shared memory, significantly reducing training times, while TPU 8i focuses on low-latency, high-efficiency inference, critical for deploying AI agents and real-time applications.

This dual-chip approach underscores a broader transition in AI infrastructure: separating training and inference hardware to optimize cost and efficiency. It also reinforces Google’s strategy to vertically integrate its AI stack and compete more directly with NVIDIA’s dominant GPU ecosystem.

References: Google Cloud, Tech Crunch