Yes it is more efficient in $/tok to run at scale than to run just for yourself. Everyone selling Deepseek V4 inference is selling an undifferentiated good. Th

Yes it is more efficient in $/tok to run at scale than to run just for yourself. Everyone selling Deepseek V4 inference is selling an undifferentiated good. They have run the numbers on how much it costs and are competing against a dozen other outfits also selling undifferentiated open weights tokens. Whatever the dollar cost they face to rent those GPUs will be what they are able to charge in the competitive market. That is great for you and me because we can buy tokens at pretty much exactly what it costs to produce them.