I think the vast majority of gpu’s serving open models are in those data centers