A regulator asks how long a stressed exit from your primary cloud would take, meaning the provider is not helping and the contract ends in 30 days. The estate is 180 TB in object storage, 40 TB of databases, about 600 container services, one managed Kubernetes cluster, and the provider's identity, queueing and secrets services. Work the number and say what actually binds.

The assumptions, stated

220 TB to move. A sustained 10 Gbps of egress with enough parallel transfers, which is optimistic but achievable; 2 Gbps if the path is shared with production traffic. Egress list pricing in the region of five to nine US cents per GB at major providers as of 2025. A team of ten engineers who also keep the lights on.

The arithmetic

10 Gbps is about 1.25 GB/s, so 220,000 GB takes roughly 176,000 seconds, a little over two days. At 2 Gbps it is ten days. The bill at 220 TB is roughly 11,000 to 20,000 dollars, which is a rounding error against the programme. Note that the EU Data Act removes switching charges for in-scope switching from 12 January 2027, which changes this line further.

So the bytes move inside the window and the money is trivial. Anyone whose exit plan leads with the egress bill has not costed the exit.

What actually binds

• Provider-specific control-plane dependencies. Identity is the worst: every service authenticates through it, workload identity is stitched into deployment manifests, and replacing it touches all 600 services. Queueing semantics and the secrets store are the same class of problem, different size.
• Other people's change control. DNS is minutes, but partner IP allowlists, bank and network registrations, certificate pinning in shipped mobile apps and integrations that hold provider-specific endpoints all need a third party to make a change on their schedule. Weeks to months, and none of it is yours to accelerate.
• The analytics and feature estate, which is usually larger than the transactional one and is nobody's named responsibility in the plan.

The number, with its range

A genuine full exit is 6 to 18 months. Thirty days buys you one critical business service, degraded, on a footprint you had already built, with a written list of what is switched off. That is the answer to give the regulator, with the list attached, because a plan that claims 30 days for the whole estate is the one that fails examination.

Which assumption dominates the error

Not the data volume. It is the count of provider-specific managed services on the critical path, because each one is an unbounded piece of work with an external dependency. Count them and publish the count; it is the single most honest number in an exit plan.

When this is the wrong answer

Not every workload earns this. For a non-critical system the proportionate answer is portable data formats, an exportable backup held outside the provider, and no rehearsal at all. Reserve the rehearsal budget for the services the regulator would name, and rehearse one of them for real each year: a measured partial exit beats a complete document, because the document's estimates are always the ones that were never tested.