Sorry, just saw your request. I'm taking that number from the Robert Koch Intitute daily status reports, which is basically the central official report regarding COVID-19 in Germany.

Here's the one from today (in English): https://www.rki.de/DE/Content/InfAZ/N/Neuartiges_Coronavirus...

The interesting table is on page 6, where there's the percentage of ventilated patients on ICU, which is 73%, so we know most ICU patients are ventilated, which means we can treat someone being on ICU as roughly equal to someone being ventilated. Right below is the number of discharges from ICU, and the percentage of deaths, which is 30% of all discharges. Since almost all ICU patients end up being ventilated, we can conclude that these about 30% of deaths mostly happened on ventilation, and most of the 70% recoveries also happened on ventilation.

I know that my calculation doesn't exactly result in the real percentage of deaths on ventilation, but the error range of my estimation does not allow the actual percentage of deaths on ventilation to even come close to 90%.

>so we know most ICU patients are ventilated

This is where your logic fails. Most patients in the ICU at this point in time are ventilated. That doesn't (necessarily) mean that most patients that went to the ICU were ventilated. Those that are ventilated will stay in the ICU for weeks. Those that aren't might only be there for a few days.

That uncertainty is why I already considered my calculation to be a rough estimate with a relatively large error window. But it is not large enough to allow the true percentage to even come close to 90%. For that, huge numbers of patients would have to be in ICU for a single day only, which is a ridiculous assumption.