#### Resolution criteria

TPUs will be considered "on InferenceMAX" when Google TPU backends are integrated into the InferenceMAX benchmark platform and results are publicly available on the live dashboard at https://inferencemax.ai/. Resolution will be confirmed by checking the InferenceMAX dashboard for active TPU benchmark data and results.

Background

InferenceMAX is an open-source continuous inference benchmark that runs nightly, continually re-benchmarking popular inference frameworks and models to track real performance in real time. The initial v1 release benchmarked NVIDIA and AMD GPUs, with plans to expand to include Google TPU and AWS Trainium backends over the following two months. The project aims to achieve over 99% coverage of deployed FLOPS globally by adding Google TPUs and Amazon Trainium.

Considerations

InferenceMAX is designed to benchmark across multiple vendors including TPU, Trainium, and others, but integration of new hardware requires building and validating reproducible CI/CD pipelines and ensuring system reliability at scale. The timeline for TPU integration may shift based on technical challenges or resource allocation priorities.