Most of the attention in a hosting deal goes to the rate. The rate is what gets compared across providers, modeled in a spreadsheet, and argued over. Almost nobody reads the part that governs what happens when the arrangement ends, which is unfortunate, because that is the part you will eventually need.

Every hosting agreement ends. Some end because the term ran out and both sides are content. Some end because you found cheaper power. Some end because the facility is in trouble. Those three exits look nothing alike, and the clause that covers all of them is usually a single sentence that nobody negotiated.

Our guide to reading a hosting agreement walks through the contract as a whole. This is about the end of it specifically.

The exit is a physical process, not a legal one

The mistake is thinking of termination as paperwork. It is a logistics job that happens inside somebody else’s building, using their labor, on their schedule, while their other customers are also running.

A machine that is leaving has to be powered down in a controlled way, unracked, disconnected, checked against a serial list, packed, palletized, staged somewhere, and loaded onto a truck that has to be booked. Every one of those steps costs someone time. The contract decides whose time, and whether you are billed for it.

It also decides how fast it happens. A facility running at capacity does not have a crew sitting idle waiting to de-rack your fleet. Your removal joins a queue behind whatever the site is already doing. If the agreement says nothing about how long that queue may be, it can be long.

The clauses that actually decide the outcome

Notice

How much warning you must give, and in what form. The thing to check is not just the length but what starts the clock. Notice given by email to a technician is not the same as notice given in the form the contract specifies, and a dispute about whether notice was properly served is a dispute you will lose while your machines sit powered down.

Also check whether notice can be given at any time or only within a window before renewal. An agreement that renews automatically unless cancelled in a narrow window is how operators end up committed to another full term they did not intend.

Who performs the removal

There are three models and they are not equivalent. The facility does the work and bills you. The facility does the work at no charge as part of the agreement. Or you send your own people, which sounds cheapest until you discover the site requires escorted access, insurance certificates, and a scheduled window, and your technicians spend most of two days waiting.

Whichever model applies, the question to settle in advance is what the de-rack rate is and whether it is per machine, per hour, or a flat mobilization. Ask before you sign, not when you are leaving.

Condition on return

Machines come back in the state the facility hands them over in. If units were repaired on site, parts may have been swapped between machines. If the facility holds a spares pool that your machines drew from, some of your hardware may contain components that started life in someone else’s unit, and some of your components may now be in theirs.

That is normal and usually sensible operationally. It becomes a problem only when nobody agreed what the rule was, and the serial numbers you shipped in do not match the serial numbers you get back. Establish whether tracking is at machine level or board level, and get the manifest.

Packing and freight

Machines that arrived in manufacturer boxes a year ago usually do not still have them. Boxes get flattened, stored badly, or thrown away. If the facility does not keep original packaging, somebody has to buy packing materials, and hardware shipped loose on a pallet arrives damaged. Our guide to shipping mining hardware covers what adequate packing looks like.

Freight is a separate cost from de-racking and is usually yours. The practical point is scheduling: a truck booked for a date the facility cannot meet becomes a detention charge, and a truck that arrives before the machines are packed becomes the same thing.

Money owed changes everything

The single biggest variable in an exit is whether your account is current.

If you owe nothing, removal is an operational task. If you owe money, the facility has leverage and will generally use it, because unpaid power is a cost they have already incurred and cannot recover elsewhere. Many agreements give the facility a right to withhold equipment until the balance is settled, and in some arrangements the facility may have rights over the hardware itself.

Two things follow. First, read what the agreement says about retention of equipment against unpaid amounts, because that clause is the difference between a billing dispute and a hardware recovery problem. Second, if you are heading toward a dispute, understand that machines sitting unpowered in a facility you are arguing with earn nothing while the argument runs. The cheapest outcome is usually to settle the balance, remove the fleet, and dispute afterwards from a position where you hold your own hardware.

The distressed exit is a different problem

Everything above assumes the facility is functioning. When the facility is the one in trouble, the questions change.

If a host fails, your machines are physical property sitting in a building you do not control, possibly leased from a landlord you have no relationship with, possibly alongside the host’s own equipment. Separating your property from the host’s property is straightforward only if the records are good enough to prove which is which. That is the real argument for maintaining your own serial manifest rather than relying on the host’s.

The practical protections are agreed at the start, not at the end. A clearly documented statement that the equipment is yours and is held on your behalf. A defined window in which you may retrieve it. A record of what you delivered, with serials, dated. None of that is exotic, and none of it can be created retroactively.

The same reasoning appears in our colocation due diligence checklist, which is worth working through before you sign rather than after.

If you are moving for a better rate, price the move

A voluntary exit is a different calculation from a forced one, and it is the one most often done badly. An operator sees a lower rate somewhere else, multiplies the difference by the fleet’s annual consumption, and treats that as the saving. It is not. It is the saving before the cost of getting there.

The move itself has four costs. De-racking at the old site. Freight. Racking and commissioning at the new one. And the downtime in between, which is the one people leave out, because it does not arrive as an invoice. Machines earn nothing while they are on a truck, and nothing while they sit in a queue waiting to be installed at the other end.

That downtime is not a day. Between powering down, packing, transit, receiving, racking, and bringing a fleet up in a controlled way rather than all at once, a move is realistically measured in weeks of partial or total outage. Whether the rate difference clears that depends on how big the gap is, how long you intend to stay, and what hashprice does while you are off.

The useful discipline is to express the switching cost as a number of months of the rate saving. If the move costs the equivalent of several months of the difference, and the new agreement is short, the arithmetic may not work even though the rate is genuinely lower. The lower rate is real. The saving is what is left after the move.

There is also a sequencing option worth considering: moving in batches rather than all at once. It extends the total project, and some facilities will not accommodate it, but it keeps part of the fleet earning throughout and reduces the risk of discovering a problem at the new site with every machine already committed to it.

Things that are easy to forget

A few items that routinely get missed in the handover and cost time afterwards.

• Pool and wallet configuration. If the host configured your pool settings, you need those details before the machines leave, not after they arrive somewhere else.
• Firmware state. Machines may be running custom firmware with settings applied by the facility. Know what is on them, because a fleet that was tuned for one site’s thermal envelope may behave differently at the next one.
• Monitoring access. Dashboard accounts, API keys and alerting integrations are usually the host’s. Export what you need while you still have access.
• Deposits. When is a security deposit returned, against what conditions, and is it applied automatically to final charges.
• The final invoice. Power is billed in arrears, so the last invoice usually arrives after the machines have gone. Know what is still coming.

What to ask before you sign

All of this is cheap to settle at the start and expensive to settle at the end, because at the start both sides are trying to make the deal work and at the end at least one of them is not.

Four questions cover most of it. How much notice, in what form, and does the agreement auto-renew. Who performs the de-rack and at what rate. What happens to equipment if there is a balance outstanding. And how quickly, in writing, can machines be released once notice is valid.

A provider who answers those plainly is telling you something useful about how they operate. So is one who does not.

If you are evaluating hosting and want straight answers to those four questions before there is a contract to argue about, get in touch. We would rather set the expectation at the start than discover a disagreement about it later.