I think the trick here is going to be using at a minimum only automotive rated parts or better, FRAM storage if possible, tantalum and ceramic caps, things like

I think the trick here is going to be using at a minimum only automotive rated parts or better, FRAM storage if possible, tantalum and ceramic caps, things like that. Speculatively, with AI moving along as it is, a “computer” might be very much like a typewriter, primarily a device for creating documents, getting them to peripherals, etc for the human user and their AI ghost/API. With thermodynamic neural nets, especially if we can get them working at room temperature, we could easily see a situation where it would be more cost and power efficient to have local generative AI simulate standard computer architecture than to actually build a Von Neumann computer from discrete components. If we can get the thermodynamic wells down to the size of flash cells, that could mean running 1024b models locally on chips the size of an SD card, peaking at around 20 watts at full utilisation. I could easily see using MCU scale compute to run a stripped down system to provide the wireframe from which the GAI could hang on the pixels and pizaz, helping the user to stay on the rails of a strictly deterministic system while decorating it with GAI. A “50 year computer” might be useful as a stand alone, but basically be an interface device when combined with generative AI running on devices that would probably be much more needful to keep current.