Categories
Hosting, Mining Business

The planned end of a hosting agreement has more failure points than most operators expect. Notice windows, serial reconciliation, condition records, packing, freight liability and the final power true-up.

Most hosting advice is written for the beginning of the relationship. How to read the agreement, what to ask about power cost, which questions expose a site that is not what the brochure says. Far less is written about the end of it, even though every hosting agreement ends eventually and the orderly ending is by far the most common one.

This is about the ordinary case. The term is up, or you are consolidating fleets, or you have built your own site and you want your machines back. Nobody is in default and nobody is angry. The machines still have to come off the rack, get counted, get on a truck, and arrive somewhere else in the condition they left. That sequence has more ways to go wrong than most operators expect, and nearly all of them are avoidable with a few weeks of notice.

If your situation is not the ordinary one, and the provider has stopped answering the phone, that is a different problem with different rules. We have written about recovering hardware from a failing hosting provider separately. What follows assumes both sides want the same outcome.

Start with the notice window, not the truck

The first thing to find is not a freight quote. It is the clause in your agreement that says how much notice you owe and in what form. Notice periods in hosting agreements commonly run from thirty to ninety days, and they frequently specify a method: written notice to a named address, or through a portal, with email explicitly excluded. A verbal heads-up to the technician you have been dealing with for two years is not notice, however good that relationship is.

Two details inside that clause matter more than the number of days. The first is whether notice has to land before a renewal date, because many agreements roll over automatically and a missed window commits you to another full term. The second is whether the notice period runs from the date you send it or the date the provider acknowledges it. That difference can be a week, and a week is enough to push a removal past a month boundary and trigger another month of fees.

Count what you own before you schedule anything

Reconcile your records against the site before a removal date exists. You want three lists to agree: the serials you believe you own, the serials the provider believes are yours, and the serials physically in the racks. In a fleet that has been running for two years, with hashboard swaps, warranty replacements, and units pulled for repair, those three lists very often do not match.

This is the practical argument for keeping a serial-level asset register for a hosted fleet while the machines are running rather than reconstructing one under time pressure. A unit that was RMA’d eighteen months ago and came back with a different serial is easy to trace at the time and genuinely difficult to trace later. Units in a repair queue at the moment of removal need an explicit decision too: do they ship as they are, do they wait for the repair to finish, or do you write them off and leave them.

Ask for the provider’s list in writing and compare it to yours before you agree a date. Discrepancies found four weeks out are administrative. The same discrepancies found on a loading dock, with a truck waiting, become arguments.

Agree who does the work, and when

Deinstallation is labor, and labor is either included in your agreement or it is not. Read the clause before you assume. Some agreements include powering down and removal from the rack. Many include nothing beyond making the units available during business hours, which means you are sending people or paying the provider’s technicians at a stated rate.

Then agree the shape of the work. Pulling several hundred units is not a single afternoon, and a site cannot usually surrender an entire aisle at once without affecting the machines around it. A staged removal over several days is normal and easier on everyone. Fix in advance which hours the work can happen, who is allowed on site, what identification they need, and whether an escort is required. Sites with real access control will not improvise this on the day.

Record condition at the rack, not at the destination

The single most common dispute in an otherwise amicable exit is about condition. A unit arrives with a bent fan guard or a cracked plastic shroud, and there is no way to establish whether that happened in the rack, during the pull, in the crate, or on the truck.

The fix is unglamorous and takes minutes per rack. Photograph units as they come out, with the serial legible, before they are packed. Note anything already damaged, anything already missing, and anything that was not running. Both parties should hold the same record. A short written condition note signed by whoever supervised the pull is worth more at the other end than any amount of later argument, and it protects the provider as much as the client.

Packing is where value is actually lost

Miners are not fragile in the way servers are, but they are heavy, dense, and full of fans and heatsinks that do not enjoy being stacked. Units shipped loose in a trailer, or stacked without separation, arrive with bent fins and sheared fan mounts, and those are repairs that cost real money against hardware whose resale value is already a function of condition.

If you still have the original boxes and foam, this is what they were for. If you do not, buy proper cartons and use them. Palletize, strap, and stretch wrap. Do not exceed a pallet height that the destination can safely unload. Label every pallet with its contents and a sequence number so that a missing pallet is obvious on arrival rather than discovered during a count a week later.

Freight, and the question of who is liable where

Decide explicitly where custody transfers. The useful question is simple: at what exact moment do the machines stop being the provider’s responsibility and start being yours? The common answer is when they are loaded onto your carrier, which means a loss in transit is your problem and your insurance question, not the provider’s.

That makes transit insurance worth a direct conversation rather than an assumption. Standard carrier liability is typically calculated by weight and bears no relationship to what a pallet of modern miners is worth. If the hardware is worth insuring at replacement value, that has to be arranged deliberately, and it has to be arranged before the truck arrives.

The final invoice is not the last invoice you expected

Power billing in hosting is usually a pass-through with a lag, because the provider is themselves billed in arrears by the utility. That means a final invoice frequently arrives after the machines have left, covering the last partial period, and it is often the invoice people forget to budget for.

Ask in advance for the shape of the final reconciliation: which period it covers, when it will be issued, how any deposit is applied, and when the balance of a deposit is returned. If your agreement includes a true-up against metered consumption, ask what meter data will support it. The habit of checking a hosting invoice line by line matters most on the last one, because it is the one nobody is still watching.

Storage fees and the cost of being late

Once units are off the rack they occupy floor space, and floor space at a hosting site is the product being sold. Most agreements allow the provider to charge storage if hardware is not collected within a defined window after deinstallation, and those rates are frequently daily.

This is a genuinely avoidable cost and it catches people who sequence the work backwards. Book the freight before the deinstallation, not after. A pull that finishes on a Friday with no truck arranged until the following Thursday is six days of storage on hardware that is also not earning anything.

What a good provider does at this point

An exit is the part of the relationship where a hosting provider’s character is most visible, because there is no future revenue to protect. The providers worth signing with behave the same way on the last day as the first: they produce the serial list without being chased, they give you access on the dates agreed, they do not discover new fees, and they return the deposit when they said they would.

That consistency is also, in practice, the best diligence signal available to a prospective client. When you are evaluating a site, ask how the last client exit went and who it was. A provider with nothing to hide will tell you, and the answer is more informative than most of the questions on a standard colocation due diligence checklist.

A short sequence that avoids most of this

Read the notice clause and calendar the deadline the week you sign, not the month you leave. Reconcile serials with the provider in writing before proposing dates. Confirm who performs the deinstallation and at what rate. Photograph units at the rack with serials visible. Palletize properly and label by sequence. Fix the custody point and insure transit at replacement value. Ask for the final reconciliation schedule in writing. Book freight before the pull begins.

None of that is difficult. It is simply the part of hosting that gets no attention until the first time an exit goes badly, and by then the hardware is already on a truck.

Explore Rax Mining

Categories