"edge deployment" by which you mean client-side application-specific inference. That's not bad, but it's also table-stakes when ONNX/Tensorflow exists and obvio

"edge deployment" by which you mean client-side application-specific inference. That's not bad, but it's also table-stakes when ONNX/Tensorflow exists and obviously isn't analogous to the server products Nvidia is shipping today. You can "deploy" CoreML the same way Android developers deployed Tensorflow for edge compute 5 years ago, but that wasn't even interesting when people demoed it on Raspberry Pi. Apple is behind.