Google ships four TPU variants in one year as it ramps up competition with Nvidia

Why it matters

Google has added two new systems to its eighth-generation AI accelerator lineup, the TPU 8tx and the TPU 8i Max, marking the first time it has shipped four TPU systems in a single year. Both variants are built on existing processors rather than a new chip generation, reflecting a modular design approach Google says lets it release hardware variants in months instead of years. The TPU 8tx keeps data and model weights encrypted even during active processing, aimed at enterprises and government agencies handling confidential workloads; Nvidia already offers a comparable confidential computing feature for its GPUs. The TPU 8i Max pairs two chips on a single package to address memory bottlenecks, doubling compute, memory, and memory bandwidth within the same rack. Google also open-sourced TorchTPU so PyTorch code runs on its chips with minimal changes. The moves reflect Google's shift from internal tool to external platform, competing with Nvidia in the growing inference market, where Amazon, Microsoft, and OpenAI are also building their own accelerators.