Cloud Exit Myths: Why One Company Stayed After the Egress Bill
Archi4 min read
Contents
- The Bill That Started the Exit Plan
- Myth: Egress Fees Are the Real Cost of Leaving
- The Part Nobody Modeled: Engineering Time
- Myth: You Can Reproduce Cloud Services Elsewhere for the Same Price
- A Note on Lock-In
- Myth: Negotiating the Bill Is a Sign of Weakness
- What Meridian Actually Changed
- The Decision That Mattered
The Bill That Started the Exit Plan#
In early 2025, the engineering team at a subscription analytics company (call it Meridian, 180 employees) opened its monthly cloud invoice and found a line item that had tripled in a year. Data transfer out: $74,000. The workload was a customer-facing dashboard that pulled aggregated event data from object storage and streamed it to browsers. Every refresh moved bytes across the provider's boundary, and every byte was billed.
Meridian's CTO, who had joined from a company that ran its own racks, did what any cost-conscious operator would do. She ordered a migration study. The board approved a budget of $400,000 for the project and set a deadline: nine months to move the analytics stack to a colocation facility in the same metro, plus a smaller secondary site for redundancy.
The plan did not survive contact with the details. What follows is a myth versus fact breakdown of the assumptions that shaped the decision, and why Meridian ultimately stayed put.
Myth: Egress Fees Are the Real Cost of Leaving#
Fact: Egress is the visible cost. The invisible costs decide the outcome.
Meridian's finance team modeled three years of spending on both paths. The cloud path came to roughly $2.9 million. The colocation path came to $2.2 million on paper, a savings of $700,000 that looked obvious.
Then the engineers added the line items that finance had left out. Rack space, power, cooling, and cross-connects at the colo: $11,000 a month, or $396,000 over three years. Replacement hardware for the storage tier, amortized over four years but paid up front: $280,000. A second site for disaster recovery: another $95,000 a year. Two additional site reliability engineers to cover the on-call rotation the provider had handled: $340,000 a year fully loaded. A one-time migration of 1.4 petabytes, including dual-running both systems for eight weeks: $210,000.
The revised colocation total came to $3.1 million. The cloud path was cheaper. Meridian's board killed the migration in month four.
The Part Nobody Modeled: Engineering Time#
The migration study itself consumed 1,800 engineering hours before it was cancelled. At a blended rate of $95 an hour, that is $171,000 spent to learn that the original invoice was not the problem. The CTO now says the study was worth it, but only because the company stopped before signing a five-year colo lease.
Myth: You Can Reproduce Cloud Services Elsewhere for the Same Price#
Fact: The services that are easy to move are the cheap ones.
The dashboard that generated the egress bill relied on a managed message queue, a managed search cluster, and a serverless function layer that triggered on upload events. Meridian's engineers estimated six months of work to replace those three services with open source equivalents, plus the ongoing cost of patching and upgrading them.
Object storage was easy to move. The queue was harder. The search cluster was the sticking point: the team's queries used features that the open source alternative did not support without a plugin the team would have had to maintain. In the end, the migration plan quietly kept the search workload on the provider and moved only the storage tier, which cut the egress bill by 38 percent and left the rest in place.
A Note on Lock-In#
Lock-in is not a single setting. It is a stack of dependencies, and the ones that cost the most to remove are usually the ones that saved the most time when they were adopted. That trade is not a trick. It is the deal the company signed, whether or not anyone read it that way at the time.
Myth: Negotiating the Bill Is a Sign of Weakness#
Fact: It is the normal next step, and it works more often than teams expect.
Meridian's procurement lead asked for a meeting with the provider's account team and brought three things: a usage forecast, the colocation quote, and a request for committed-use pricing on the storage tier. The provider offered a three-year commitment at a 22 percent discount on storage and a 40 percent reduction in egress for the first year, with a cap on future egress increases.
The savings came to roughly $310,000 over three years, more than the projected benefit of the migration, with none of the operational risk. The contract also included a clause Meridian had not asked for: a fixed egress allowance that reset annually, which the finance team now tracks monthly.
What Meridian Actually Changed#
The company did not exit the cloud. It did four smaller things instead. , Moved the dashboard's data pipeline to a regional endpoint so refreshes no longer crossed the billing boundary., Cached aggregated results at the edge, cutting outbound traffic by 61 percent., Added a monthly egress report to the engineering leadership review, with a threshold that triggers a design review when any single service exceeds $8,000., Rewrote the internal architecture guide to require an egress estimate for every new feature that serves data to browsers.
The egress bill for the most recent month was $19,400. The colocation lease was never signed. The migration study sits in a shared drive, labelled with the date it was cancelled.
The Decision That Mattered#
The decision was not cloud versus colo. It was whether to trust a single invoice line as a proxy for total cost. Meridian's CTO put it plainly in the post-mortem: the egress bill was a symptom, and the migration was a treatment for a disease the company did not have. The cheaper move was to fix the pipeline, negotiate the contract, and keep the operational burden where it already sat.
More articles