Comment by tangotaylor

I'm going to fight pessimism with cynicism here: the Department of Defense is not going to let everything move to the cloud because they need compute at the edge for AI-enabled weapons and R&D. For example, Anduril's products, Eric Schdmit's secretive Bumblebee project, or startups like Scout AI. Communications and GPS are just too easy to jam and their answer is giving weapons more last-mile autonomy to operate in radio silence.

War aside, I also bet there's going to be a huge demand for edge-compute for other kinds of robotics: self-driving cars, delivery robots, factory robots, or general-purpose humanoids (Tesla Optimus, Boston Dynamics Atlas, 1X NEO, etc). Moving that kind of compute to the cloud is too laggy and unreliable. I know researchers who've tried it, the results were mixed.

Also, the engineers working on these platforms aren't going to reinvent the wheel every time they need to connect hardware together and they're going to use interoperable standards, like PCIe for storage or GPUs, DIMM slots for memory, ATX for power, etc. So I don't see general-purpose computing dying.