>>107163249 (OP)
Yes and no. Realistically any existing network of computers will always have to be heterogeneous with lots and lots of specialization and centralization. It'll always be a multilevel system because it's vastly more efficient that way and at serious grid scale percentage points on prices are a big deal, let alone orders of magnitude.
On the other hand he's entirely right about software and hardware being largely unreliable fragile garbage that, compared to any other form of engineering, is less robust than a child's tinker toy construction.
Probably the ideal for practical implementation is just a mesh network of socs, not anything as low level as a cellular automata fabric.
Though there's legitimate cases for using that sort of extreme level of redundancy for certain specific applications like high performance radiation hardening or very long term autonomous drones. So basically spacetravel, or possibly ocean exploration. Anywhere that you can't just order a replacement computer and have it shipped to you in a few days.
