How to Read a Hosting Provider’s SLA: 6 Clauses Where Miners Lose Money
Uptime is the number everyone compares. The exclusions, credit formula and exit terms underneath it are what actually decide your cost base.
The difference between 95% and 98% uptime looks like three percentage points. In practice, 95% allows 36 hours of downtime per month; 98% allows 14.4 hours. Those 21.6 hours nobody scrutinises before signing are 3% of the month your machines spend earning nothing while the rest of your cost base carries on.
With electricity typically consuming 75–85% of a mining operation’s operating costs and hosting rates clustered around $0.07–$0.08/kWh, there is very little margin to absorb idle time. Yet the real problem is not the uptime figure itself. The real problem is what sits behind that figure in the contract.
Why Your SLA Deserves as Much Attention as Your Energy Rate
A common scenario: an investor compares three hosting offers, all priced within a cent of each other. They pick the cheapest, sign, and skim the SLA. Hours go into comparing hashrate and energy efficiency; a few minutes go to the one document that determines how risk and cost are split between them and the provider.
The asymmetry is easy to demonstrate. A $0.01/kWh difference on a 3 kW machine running continuously works out to roughly $260 per miner per year (3 kW × 8,760 h × $0.01). That is the number everyone negotiates. Meanwhile, a single clause on relocation notice or auto-renewal can cost several times that — and it is usually the clause nobody reads.
Below are six clauses worth checking in any hosting contract. Each includes a typical formulation, its hidden meaning, and the financial consequence.
If you have not signed yet, read with a calculator open. If you already have, pull up your agreement now.
Clause 1. Uptime Definition and Exclusions
“99% uptime” has no fixed meaning across the industry until you see how it is measured. The raw allowances, per year:
- 99% — 87.6 hours of permitted downtime
- 99.9% — 8.76 hours
- 99.99% — 52.6 minutes
- 99.999% — 5.26 minutes
Each decimal is not abstract precision. Moving from 99% to 99.9% removes roughly 79 hours of permitted downtime a year — over three full days.
The headline number, however, is secondary. What matters is the exclusions list. Providers typically carve the following out of their uptime calculation:
- Scheduled maintenance
- Force majeure events
- Failures caused by the client’s own equipment
- Third-party carrier outages
- Security actions initiated by the client
“Planned maintenance” is the most dangerous loophole. If the wording lets the provider declare any outage “planned” with no cap on frequency or duration, the uptime guarantee becomes decoration. Routine maintenance should not automatically qualify as force majeure either; an overly broad clause lets the host shed nearly all accountability.
For context on what is at stake operationally: in the Uptime Institute’s annual outage analysis, more than half of surveyed operators put the cost of their most recent significant outage above $100,000, and one in five above $1 million. Those are enterprise data-centre figures rather than mining ones, but the underlying point holds — outage frequency has been falling while the cost of each individual incident rises.
What to check: locate the “Uptime Guarantee” section and highlight every exclusion. If you see “excluding planned maintenance” with no concrete limits, request written clarification: how many maintenance hours per year are permitted, how many days’ notice is required, and what the maximum window per event is.
Clause 2. Credit Formula and the True Cost of Downtime
SLA credits almost never cover actual losses. They function as a penalty for the provider, not insurance for the client.
The mechanics are straightforward. A service credit is a financial offset against future invoices, not a payout for damages. Amazon’s EC2 SLA is a useful reference because the terms are public: at an instance-level uptime percentage below 99.5% but at or above 99.0%, the credit is 10% of that instance’s monthly bill. On an instance billed at $3 per month, up to roughly 7 hours 18 minutes of unavailability yields 30 cents. The standard formula across the industry is the same — monthly fee for the affected service multiplied by a credit percentage.
Apply that logic to mining, and note one structural detail first. If your contract prices power purely per kWh, downtime means you do not pay for the power you did not draw; the loss is revenue only. If you are on a fixed monthly rate or a minimum-consumption commitment, you pay for the idle hours too, and the loss is revenue plus fees.
To size it for your own fleet, use:
Lost revenue = fleet hashrate (TH/s) × current revenue per TH/day × days offline
Take the revenue-per-TH figure from a live hashprice index on the day you run the calculation rather than from any published estimate — it moves with network difficulty, transaction fees and the BTC price, and a number written into an article is out of date within weeks.
Then set that against the maximum credit: monthly hosting fee × credit percentage. In most hosting agreements the two are not the same order of magnitude, and the credit is capped besides — sometimes per incident, sometimes as an annual aggregate. Find the cap; it is usually a separate sentence from the credit table.
What to do: run the calculation above for your own fleet and compare it with the maximum recoverable credit. If the gap is 5× or greater, negotiate a higher credit percentage, a shorter credit-trigger threshold, or a performance-based fee structure that rewards the provider for exceeding guaranteed uptime rather than merely penalising failure.
Clause 3. Claim Deadlines and Filing Procedures
A miner notices an outage, logs the incident, and decides to batch all claims at the end of the quarter. Sixty days later, they discover the right to a credit has expired.
Get WOW Innovation GmbH’s stories in your inbox
Join Medium for free to get updates from this writer.
Most SLAs require claims to be filed within 30–60 days after the end of the billing month. Missing that window forfeits the right to compensation, even when the downtime is fully documented.
Timing is not the only trap. Some agreements mandate a specific filing method: a designated email address, a particular online portal, with date, time and a monitoring-system reference attached. Others impose dual windows — notify the provider within five days, then submit a formal credit request within 30. Two separate deadlines, two separate actions. Miss either and the claim is void.
What to implement:
- Log every incident on the day it occurs — timestamp, dashboard screenshot, support ticket with the provider
- At month-end, reconcile all incidents and submit claims 3–5 days before the filing window closes
- Run third-party uptime monitoring (UptimeRobot, Pingdom, or equivalent) — this is the only evidence the provider cannot dispute
Clause 4. Auto-Renewal, Minimum Terms, and Exit Penalties
One formulation recurs in colocation contracts and deserves particular attention: automatic renewal for a full term unless written notice is provided 90–180 days before expiry.
The arithmetic: a 12-month contract signed on 1 January. To exit on 31 December, notice must reach the provider by early October (90 days) or early July (180 days). Miss the deadline and the contract extends another 12 months — often at a rate the provider sets unilaterally.
Typical minimum commitments run 12–24 months, and longer in some markets. Early termination is expensive. The standard ETF formula is the current monthly fee multiplied by the remaining months. Cancelling in month five of a 12-month contract at $1,000 per month leaves seven months outstanding — a $7,000 penalty.
What to do:
- Enter the renewal-notice date in your calendar with an alert two weeks ahead of the deadline
- When signing a new contract, negotiate a 60-day notice period instead of 180
- Add an option to extend in 3-month blocks at a pre-agreed rate, so you are not locked in for another full year
Clause 5. Hidden Charges Beyond the Base Price
The base hosting rate is the tip of the iceberg. Charges commonly excluded from it:
- Bandwidth overage — exceeding your contractual traffic limit. A simultaneous firmware update across 100 miners can trigger a spike that costs more than the firmware work itself
- Cross-connect fees — a charge for each connection to another provider or rack
- Remote hands — on-demand technical support, typically billed hourly
- Equipment maintenance — if the provider manages your switches or PDUs
- Electricity above threshold — consumption beyond the contractual cap, often at a materially higher rate
- One-time setup and installation — racking, commissioning, network provisioning
To see how these stack up, build the comparison yourself. Take each provider’s quoted monthly figure, then add: PDU rental, committed bandwidth plus expected overage, cross-connects, an estimated monthly allowance for remote hands, and the amortised setup fee. In practice the difference between the headline rate and the loaded rate is where offers that looked identical stop being identical.
What to request: an itemised pricing sheet with full breakdowns — bandwidth tiers, cross-connect rates, remote-hands hourly rate, PDU rental, setup fees. Compare three offers not by base kWh price but by total cost including all fees, on the same assumed load and support usage.
Clause 6. Equipment Relocation and Decommissioning Terms
A formulation worth searching your contract for: the provider may require relocation of client equipment on a minimum of 180 days’ notice, or immediately in case of emergency. If, on month 18 of a 24-month project, the provider decides to reorganise the facility, you absorb the migration window — with the risk of equipment damage and a period of zero output.
Decommissioning is a separate concern, and the trap is not the physical work but the contractual runway. Unracking miners is fast; arranging logistics, customs paperwork where a border is involved, a receiving site with available power, and transport for the containers is not. If your agreement gives you a short window to vacate and you continue paying facility fees until the equipment is gone, every week of delay is a direct loss on hardware that is producing nothing.
What to write into the contract:
- Non-emergency relocation — a minimum of 12 months’ advance notice instead of 180 days
- The right to extend the decommissioning period in 3-month blocks at a pre-agreed rate
- Clear allocation of migration costs: logistics, installation, downtime — who pays for what
- A defined maximum duration and frequency for provider-initiated relocations over the contract term
If the provider refuses to commit these terms in writing, treat it as a signal: the facility may be unstable, or a restructuring may already be planned.
Return to those 21.6 hours separating 95% from 98% uptime. You now know that exclusions in the SLA can erase them, that compensation can reduce them to a rounding error on your invoice, and that a missed 30-day filing window can void any claim entirely. Relocation terms can consume additional months, unlisted fees can move the real rate well above the quoted one, and auto-renewal can lock you in for years.
A hosting contract is not a formality. It is a financial instrument that distributes risk between you and the provider.
Your first step: open your current contract — or the draft of a new one — and walk through the six clauses above. Two hours with a calculator are worth more than a month spent comparing energy tariffs.
If you need help evaluating a hosting offer, reach out to us at WOW Innovation GmbH — we have been reviewing data-centre and hosting contracts since 2016.
This article is for informational purposes only. It does not constitute legal, tax, financial or investment advice, and it does not offer or imply any guarantee of mining returns, profitability or uptime. Mining outcomes depend on network difficulty, energy prices, hardware performance and market conditions, all of which vary. Consult qualified legal and financial advisers before entering into a hosting agreement.
