You are misunderstanding MTBF.
What MTBF means is that if you have a large number n of drives running at the same time, you can expect to see a failure once every MTBF / n time units.
Basically, it measures how frequently relatively young drives fail; it does not say anything about the long-term survival of particular drives. These would be equivalent only if failures are linear over time, but obviously they are not. The failure rate generally increases over time.
Consider humans. Suppose that around age 20, 0.1% of males die each year. That would produce an MTBF of about 1000 years. This MTBF can be used, for example, by the military to predict the number of male non-combat deaths per year. But clearly it doesn't imply that people live for 1020 years! Mortality increases with age, so that inference would be spurious.
For more information, see https://en.wikipedia.org/wiki/Survival_analysis.