We recently planned a carve-out: one Atlassian Cloud estate, 4,500 users, that had to become two separate environments.
It is not a migration. A migration has a source and a destination. A carve-out has one source that must become two, and the hard part is never the data.
Three things do not divide cleanly.
1. People who belong on both sides
The organisation that verifies an email domain manages the accounts on that domain. One domain, one managing organisation. That is fine until the split produces people who genuinely need to work in both entities.
At that point you are not toggling a setting. You are deciding which entity a person belongs to, and which one gets to enforce policy on their account. Shared services, group functions, anyone seconded across the line: every one of them is a decision.
That conversation belongs to HR and legal, not to Jira admins. It also has to happen early, because access design, licence counts and migration sequencing all depend on the answer.
2. App licensing gets more expensive, not less
Marketplace apps are licensed per site and priced by user tier. Split 4,500 users across two sites and you do not pay half twice. You pay two tiers, and the total lands above what the single estate costs today.
That surprises finance, and it surprises them late, usually after the separation budget is already fixed.
There is a second problem underneath it. Several apps have no supported way to move part of their data to another site. For those the options are rebuilding configuration by hand on the new side, or leaving the app behind entirely. Both are decisions with a cost, and both need the app vendor involved rather than assumed.
3. Content both sides need
There is no share in a carve-out. Every Confluence space and every Jira project ends up on one side of the line.
Where both sides genuinely need something, you have two options. Duplicate it, and accept that the two copies start drifting the day after cutover. Or pick a side, and let the other one operate without it.
Neither is a technical decision, though it will be handed to you as one.
The pattern underneath
All three have the same shape. A carve-out surfaces every decision a company has been quietly avoiding. Who owns this. Who pays for that. Which entity does this person actually work for.
None of those questions are new. The split just removes the option of not answering them.
The Atlassian work is the easy half.
If this is coming
If a split, merger or divestment is on the horizon and your Atlassian estate sits in the middle of it, start the planning before the legal completion date is fixed. Once that date exists, your options narrow quickly and the expensive workarounds start looking like the only ones.