No, but, you should get excellent results. Once the model is loaded, it should be able to take full advantage of the GPU assuming you have a model that will fit within the VRAM of the GPU. Many higher end ML workstations that offer multiple GPU support on the motherboard, e.g. six and eight way GPUs at least, such as those from ASUS, use a PLX controller which is effectively splitting the bandwidth of one or two PCIe ports across multiple devices. And an eGPU dock is effectively a PLX for a single device talking to Thunderbolt. Device to device communication should still retain the full PCIe bandwidth capability, and you can see that reported in the screenshots of the article. The biggest issue is obviously going to, as always, be driver support. You could even consider an 8-way GPU "dock" such as the Cubix where you could plug your $1,500 laptop into 8x A100 80GB GPUs. 16 GPUs if you cascade them.