Please tell us why, statistically or psychologically, one would need more than 999 versions. Once you have about 10 to 30 versions, the statistical benefits and also the psychological benefits of concept rotation, are extremely minimal for including even more versions. More than 100 versions is probably overkill, though it isn't harmful. The statistical and psychological benefits between having 500 versions and 5000 versions is probably not even measurable unless hundreds of thousands of respondents are involved.
So, my guess is that the person who is asking for more than 999 versions is not well informed about the issues involved. That person somehow believes that respondents must have unique versions of the questionnaire. Maybe there is a real reason in your case that unique versions must be seen by each respondent, but I just don't get it. Once you have enough versions to achieve statistical precision and enough rotation of the attribute combinations (and that usually occurs well before 999 versions), then you are butting up against a hard ceiling.
If you really must have more than 999 versions, then generate two SSI Web questionnaires. Specificy 999 versions in each. Use a different design seed in the CBC Design tab, so that you get different designs in each study. Then, randomly assign respondents to get one questionnaire or the other.
Then, on the back end, export the data to a .CHO file (or .CSV file). Combine the two .CHO files using a text editor (they are just text files). Submit the combined data file to CBC/HB for utility estimation.