Atlassian Didn't Migrate Its Customers to the Cloud. It Removed Every Other Exit.

On October 16, 2020, Atlassian told the millions of teams running Jira and Confluence on their own servers that the ground beneath them had an expiration date.¹ First new license sales would stop, then support, then security patches, then bug fixes. The framing was warm and forward-looking: the future is cloud, and we're all going there together. Co-CEO Scott Farquhar pointed out that more than 90% of customers start with cloud and that more server customers were switching every day.⁸ It sounded like a company reading the room. It was a company emptying it.

The story you'll hear is that Atlassian led a cloud migration its customers wanted. Almost every word of that is doing quiet work it shouldn't. The 90% figure describes new customers — not the installed base being asked to leave. And the migration wasn't pulled by demand; it was pushed by a series of end-of-life mandates, each one closing a door the customer had paid to stand in.

Read the deadlines, not the press release

Strip away the language and look at the sequence of actions, because the sequence is the strategy. New Server licenses stopped selling on February 2, 2021. Server support — security updates, bug fixes, all of it — ceased entirely on February 15, 2024.¹ Customers who had bought into a self-managed product — one sold as an annual-maintenance license tied to a one-time deployment — were left with software that still ran but would no longer be defended against the next exploit. That is not a migration. It is a deadline with a friendly face.

The mechanism is plainest in the revenue. Server maintenance revenue fell year over year in widening steps — 29%, then 27%, 31%, 35%, 69% — and finally hit a clean 100% decline in the fourth quarter of fiscal 2024.⁴ A whole product line walked to exactly zero, on schedule. Meanwhile, the period since the October 2020 Server EOS announcement saw enterprise seats migrating to Cloud rise nearly sevenfold, on Atlassian's own disclosed count.³³ One line goes to zero by design; the other goes up by gravity. The gravity is the deadline.

The promise Atlassian made — and then took back

Here is the move that turns this from a migration into a thesis. When Server was killed, Data Center was the safe harbor — the self-managed product that big enterprises could keep. Atlassian even said publicly it had no plans to end-of-life Data Center — in a June 2022 official post, it stated explicitly: "To be clear, we have no plans to end of life or end of support for our Data Center offering."⁹ Then, in September 2025, it did exactly that, setting a hard Data Center EOL of March 28, 2029, with new sales ending in 2026 and the last purchase date in 2028.⁶⁷ The harbor was never a harbor. It was the next room in a hallway with one exit, and the exit is Cloud.

Notice what the second kill reveals about the first. A genuine cloud transition wouldn't need to keep removing self-managed options after the original deadline passed — customers who wanted cloud would already be there. The only reason to also eliminate Data Center is that a large, profitable cohort had chosen to stay self-managed, and staying was no longer permitted. Atlassian's own filings concede the friction in plain language: it offered discounts to certain enterprise Server customers to incentivize migration, and warned that if remaining Server or Data Center customers don't transition, revenue and profitability could suffer.⁵ That is not the language of pull. It's the language of a company that has to push, because the demand was never as universal as the slide deck said.

Isn't this just a vendor doing the obvious thing?

The honest counter is strong, so let's make it. Maintaining three deployment models — Server, Data Center, and Cloud — is genuinely expensive, and a smaller cloud-native codebase ships features faster and is more secure for most customers. Atlassian's own framing, that Data Center 'will no longer be able to deliver on our core promise,'⁶ is not pure spin; a self-managed product really does fall behind a continuously deployed cloud one. And the cloud penetration is real at some level — 99% of its 300,000-plus customers, the company says, already benefit from Cloud.⁶ So why call it a monetization reset rather than good engineering housekeeping?

Because of how that 99% is built and how the deadlines are timed. 'Benefit from Cloud' is undefined in Atlassian's disclosure; on any reasonable reading it likely counts any customer with any cloud product — free tiers, hybrid setups — not customers who have fully exited self-managed deployment.⁶ The figure flatters the very transition it's used to justify. And the tell is the broken promise: a company simplifying for the customer's sake doesn't reverse a public assurance about Data Center the moment the holdouts prove they'd rather stay. Engineering efficiency and forced monetization aren't mutually exclusive — but when the efficiency story requires removing every exit on a calendar the vendor controls, the customer is no longer the one being served. They're the one being scheduled.

Atlassian didn't move its customers to the cloud. It removed every place they could stand that wasn't the cloud, one product at a time, on a clock it wound itself — Server to zero by February 2024, Data Center to zero by March 2029.⁴⁷ The genius isn't the migration; anyone can ship a cloud product. The genius is converting an entire installed base from owners into subscribers without ever asking them to choose, by making every other choice expire. The future may well be cloud. But the customer never got to decide when their present ran out.