Video and 3D generation devour massive compute

How to verify

Will training and inference for video and 3D generation models account for at least 30% of global AI compute consumption?