A data point on self-hosting being preferable, or using an alternate gpu cloud host who can run the model privately/semi-privately for you.