Democratization of AI

The harness of AI is with big companies. This is because of the huge hardware requirements for running AI inference and training.

For real democratization of AI, the hardware requirements has to plummet. If we look at hardware requirements, we can classify them further into CPU/GPU and memory.

The GPU requirement is a myth, because we do not have tailor made ARM (RISC) for matrix mul and limited instruction set capability on large memory. We need a chip that can run of large memory set.

When the above happens, we will not need server farms. This would lead to true democratization of AI. With sub 20Watt devices with 16GB RAM able to run most AI tasks.

We need to rework AI from the bottom up again. Instead of fitting costly GPU’s into AI requirements, we need to design these hardware from scratch.