This calculated column makes everything slow

Note that Greg's measure will have similar performance issues (compared with the calculated column) but using it instead of a calculated column will keep you from needing to recalculate it each time you edit your model.

Using the new OFFSET function, here's a version that's about 30x faster (~6 ms vs ~180 ms) according to my testing against the pbix attached to the post I linked. It should be much better for large models as well as more dynamically adaptable to various filters (plus it only needs to calculate over all selected data rather than the entire fact table).

MTBF (Hours) AO =
VAR _Summary_ =
    CALCULATETABLE (
        SUMMARIZE (
            Repairs,
            Repairs[MachineName],
            Repairs[RepairStarted],
            Repairs[RepairCompleted]
        ),
        KEEPFILTERS ( Repairs[RepairType] <> "PM" )
    )
VAR _AddNext_ =
    ADDCOLUMNS (
        _Summary_,
        "@NextRepair",
            SELECTCOLUMNS (
                OFFSET (
                    1,
                    _Summary_,
                    ORDERBY ( Repairs[RepairStarted] ),
                    PARTITIONBY ( Repairs[MachineName] )
                ),
                "RepairStarted", [RepairStarted]
            )
    )
VAR _Uptime_ =
    ADDCOLUMNS (
        _AddNext_,
        "@Uptime",
            DATEDIFF (
                [RepairCompleted],
                COALESCE ( [@NextRepair], NOW () ),
                SECOND
            )
    )
RETURN
    AVERAGEX ( _Uptime_, [@Uptime] ) / 3600