An AI project has a lot of servers. Who pays to keep them running?

...

Photos of the equipment and an impressive amount of computing power show what the project is capable of offering. They don’t show how much customers are willing to pay for it.

Imagine infrastructure for AI applications in the $NEAR ecosystem, $ETH or another network. The equipment requires upfront investment, followed by bills for electricity, maintenance, and replacements. Low paid usage can make large-scale infrastructure expensive to maintain.

It’s useful to distinguish three metrics: available capacity, work actually performed, and payments from external customers. Free trials and activity driven by rewards are not the same as sustainable commercial demand.

Even a functioning business raises another question: how is its economic performance connected to the token? Using a service doesn’t necessarily generate income for every asset holder.

When evaluating an AI project, what interests you first: the size of its infrastructure or recurring customer payments for using it?