Using words like "remotely not the hard part" that is just not true man. There doesn't exist hardware that can do inference with even close to as low as the 100

Using words like "remotely not the hard part" that is just not true man. There doesn't exist hardware that can do inference with even close to as low as the 100 watts Tesla uses. Also it is not just to drive FSD, it is to provide data, to detect situations of from where to send data as they can't just stream everything to base (check the Autonomy day talk, it's quite interesting), which is needed for the development of the software which we agree is an extremely hard problem. To further demonstrate this, consider the fact that in the current AI landscape, hardware is actually the bottleneck. Everyone is trying to buy more GPU's NVIDIA simply can't supply fast enough. If Tesla still relied on NVIDIA for the in-car processing, the project would have to be shelved as they'd have to sell Teslas without the hardware, instead of just having it all cars even if they will never use it for running FSD directly on that car.