Comment by altairprime

I think the other component of the moat is that their OS/GPU stack is power-optimized for this effect in hardware, which generic solutions for generic hardware will have trouble matching - even a lower fidelity replication’s power drain could be an order of magnitude higher as a result.