System design for ML engineers: how different is the interview?

I'm an SRE, but I'm looking at roles that involve more ML infrastructure. I'm curious how system design interviews differ for ML engineers compared to traditional backend or infrastructure roles. Obviously, there's the added complexity of model serving, feature stores, data pipelines for training, and MLOps. How much traditional system design depth is still expected? Do they want deep dives into data partitioning and distributed databases, or is the focus primarily on the ML-specific components? I'm trying to understand the balance between demonstrating strong fundamentals and showcasing ML-specific architectural knowledge.