Hi @super richmann - Do you mean ML.NET AutoML (C#) or Azure AutoML.
The only case where ML.NET AutoML (C#) can use GPU is when training Image related models (such as ImageClassification), I believe.
Most ML.NET algorithms are classical algorithms and would not get much improvement if using GPU (Probably LightGBM, but I don't think GPU is supported for classical algorithms in ML.NET).
The second (Azure AutoML) is Python based and if using Azure ML compute you can use GPU for DNN operations such as text featurization using BERT under the covers. For Azure AutoML we're also releasing CLI and REST API to trigger training jobs in the Azure cloud.
I work for the Azure AutoML team, so if you are interested on enrolling the CLI/REST-API private preview, send me an email to cesardl at microsoft.com, ok?
Azure AutoML Python SDK is GA for more than 1 year, though (without CLI and REST API).