The Cloud Exit Nobody Talks About: Where Workloads Actually Go
Archi4 min read
Contents
- The Migration Map Behind the Cloud Exit Headlines
- What the Announcements Leave Out
- The destination is usually another cloud
- Egress math is the real trigger
- Three Patterns That Show Up Repeatedly
- Lift, shift, and split
- The compliance detour
- The colocation middle ground
- What the Coverage Gets Wrong About Cost
- How to Read the Next Announcement
The Migration Map Behind the Cloud Exit Headlines#
When a company announces it is leaving the cloud, the coverage usually follows the same arc: a dramatic savings figure, a quote about control, and a story about workloads returning to a data center the company owns. The headlines frame it as a retreat from public cloud. The operational reality is messier. Most workloads that leave one cloud provider do not go back to a server closet. They move to a second provider, or to a smaller specialized host, or to a hybrid arrangement that the announcement never mentions.
This matters because the public conversation about repatriation shapes budgets, hiring, and contract negotiations at thousands of companies that will never issue a press release. If the only visible story is "we left the cloud," the less visible story, which is "we changed clouds," gets treated as a footnote. It should be the main event.
What the Announcements Leave Out#
The destination is usually another cloud#
A large enterprise that shifts a core billing system off one hyperscaler often has three reasons: a contract renewal that no longer pencils out, a latency problem tied to where the workload sits relative to users, or a data governance rule that names a jurisdiction. None of those reasons require owning hardware. The nearest fix is often a different provider with a region in the right place, or a smaller vendor willing to negotiate on egress and committed spend.
The published case studies tend to collapse this into "on-premises," because that framing tests better with boards and reads cleaner in a press release. In practice, a company might keep its analytics in one cloud, move its customer-facing API to another, and run a few regulated workloads in a colocation cage. That is not a cloud exit. It is a portfolio rebalance, and it has very different cost and staffing consequences.
Egress math is the real trigger#
Most migration decisions start with a number a finance team can see: the bill for moving data out. A workload that queries a large object store across regions, or serves media from a bucket to users on another network, can spend more on egress than on compute. When that line item crosses a threshold, the conversation changes from "can we optimize" to "where else could this run."
That is why the second cloud often wins. The workload does not need a forklift. It needs a provider whose egress pricing, region placement, or committed-use discount fits the traffic pattern. The migration is a routing change and a contract change, not a hardware purchase.
Three Patterns That Show Up Repeatedly#
Lift, shift, and split#
A common first move takes one service, not an entire estate, and places it somewhere cheaper for its specific traffic. A media transcoding pipeline, a batch analytics job, or a public API endpoint can move without dragging the rest of the application along. Companies that do this well treat the migration as a series of small, reversible steps. Companies that do it badly announce a full exit, then quietly keep half their estate where it started.
The compliance detour#
Data residency rules in several markets have pushed workloads toward providers with local regions or toward sovereign cloud offerings run by domestic operators. The workload did not come home. It moved to a provider that could show a regulator the right paperwork. Headlines about "leaving the cloud" often describe this, even though the company is still renting compute and still paying a monthly bill.
The colocation middle ground#
Some workloads do leave public cloud, but they land in rented racks in a colocation facility, not in a company-owned data center. The company controls the network and the hardware choice, and a colocation provider handles power, cooling, and physical security. This is a real change in operating model, and it requires staff who can manage hardware again. It is also a small fraction of the migrations that get reported as repatriation.
What the Coverage Gets Wrong About Cost#
The usual savings claim compares a public cloud bill to an estimate of owned infrastructure. That comparison is rarely honest. It tends to exclude the cost of the people who will run the new environment, the cost of the migration itself, and the cost of the duplicate systems that run during the transition. A migration that takes nine months includes nine months of paying for both sides.
The second cloud option has its own hidden costs: new IAM models, new networking primitives, new monitoring, and engineers who have to learn a second set of tools. A company that moves a workload to save on egress can easily spend the savings on retraining and duplicated observability for the first year. The headline number rarely includes this, and the internal postmortem almost always does.
How to Read the Next Announcement#
When a company says it is leaving the cloud, the useful questions are concrete. Where does the workload actually run after the migration? Who operates the hardware? What happens to the data that stays behind? How many systems are now in two places at once, and for how long?
The answers usually describe a smaller, more incremental change than the announcement suggests. That is not a criticism of the companies making the move. It is a correction to the story. Cloud infrastructure is not a binary choice between owning and renting. It is a set of routing decisions, and most of them point to another provider rather than back to the basement.
More articles