This makes running larger machine learning models in the browser feasible - see e.g. https://github.com/webonnx/wonnx (I believe Microsoft's ONNXRuntime.js will also soon gain a WebGPU back-end).