SQL Server update same column using different conversion without overwriting

I have a table DATES that looks like this:

The column TimeStamp is of type nvarchar and TimeStamp_UTC of type datetime.

I want to convert the data from TimeStamp into the column TimeStamp_UTC by using just one query.

That implies using two CONVERT functions, one for the first two dates and a slightly different one for the last two.

The SQL statements to convert both types are shown here:

For the first two:

UPDATE DATES
SET [TimeStamp_UTC_JM] = (SELECT CONVERT(datetime, [TimeStamp], 20) 
WHERE LEN([TimeStamp]) IN (18, 19))

For the other two dates:

UPDATE DATES 
SET [TimeStamp_UTC_JM] = (SELECT CONVERT(datetime, [TimeStamp], 21) 
WHERE LEN([TimeStamp]) BETWEEN 23 AND 26)

Individually both updates work, but when running the second query the converted values of the first update disappear, so I would like to perform the update in just one step, without overwriting.

Just run these two UPDATE statements after each other - check to ensure you're not overwriting any existing values in Timestamp_UTC in your WHERE clause:

-- update the first style of date/time formats
UPDATE dbo.Dates
SET [TimeStamp_UTC] = CONVERT(DATETIME2(3), [TimeStamp], 120) 
WHERE LEN([TimeStamp]) IN (18, 19)
  AND TimeStamp_UTC IS NULL;

-- update the second style of date/time formats
UPDATE dbo.Dates
SET [TimeStamp_UTC] = CONVERT(DATETIME2(3), [TimeStamp], 101) 
WHERE LEN([TimeStamp]) BETWEEN 23 AND 26
  AND TimeStamp_UTC IS NULL;

I had to also change the styles used for conversion - since you have 4-digit years, those have to be styles in the 100er range - and the second style you had doesn't exist in 4-digit years - but style = 101 seems to work just fine.

I also used DATETIME2(3) as datatype, since this is recommended since the days of SQL Server 2008 - it has a better precision, uses less memory to store its values, and has a larger range of supported dates - basically nothing but benefits over using the old DATETIME datatype. I'd strongly recommend defining any new columns to store date and time with the DATETIME2(n) datatype and phase out DATETIME