What the article is getting at, is that the service makes promises based on the assumption that later subscribers will show up to provide enough data to satisfy the SLA they give to the early subscribers, before those early subscribers attempt to "call them on it" (by i.e. actually attempting to use the less-common data sets in production.) If the later subscribers don't come, the service will eventually be found useless for the early subscribers, and so they've paid in for nothing.

Thus, like in a (complete-knowledge) Ponzi scheme, the early subscribers have an incentive to get later subscribers to sign up for the service, so that it can be more useful for them.

But, unlike in a Ponzi scheme, the later subscribers don't have to wait for subscribers even later than them, before they can benefit. The later you subscribe, the more immediate value the system has to you.