News

What Uptime Guarantees Should Mining Hosts Provide?

The one number that decides whether hosting is worth paying for – and the four words in the contract that quietly cancel it.

Uptime is the share of time your Bitcoin miner is powered on and hashing. It is also the only hosting metric that directly converts into coins: a machine that is switched off earns nothing while the hardware keeps depreciating. Yet it is the metric the industry is loosest about. There is no regulator auditing mining-host uptime claims, no standard definition of what counts as downtime, and no requirement to publish anything at all. A host can advertise 99.9% and mean something completely different from the host next to it advertising 99.9%.

So the useful question is not ‘what number should a host advertise?’ It is ‘what number should a host be contractually bound to, how is it measured, and what happens when they miss it?’ This guide answers all three. It covers the data-centre standard the industry borrows from, why that standard does not transfer cleanly to a mining facility, the three levels of uptime promise you will encounter, and the exclusions clause that quietly turns a strong-looking guarantee into a weak one.

Key takeaways

  • Ask for a guaranteed floor in the contract, not an advertised figure on a website. A written 95% beats an unwritten 99.9%.
  • Insist the measurement is per miner, not a facility average – a site average can look healthy while your specific machine sits dark.
  • Read the exclusions before the percentage. Planned maintenance, grid events and pool problems are commonly carved out.
  • Uptime Institute Tier III is 99.982% (about 1.6 hours of downtime a year) and Tier IV is 99.995% (about 26 minutes) – but those certify data-centre infrastructure, not your miner’s hash output.
  • Repair distance is the hidden variable: on-site spares turn a failure into days, shipping abroad turns the same failure into weeks.
  • OneMiners guarantees 95%+ across 20 sites and about 2,163 MW, and observes 98%+ in practice – we publish both numbers deliberately.

What uptime actually measures

The arithmetic is simple. If your Bitcoin miner runs 95 hours out of every 100, that is 95% uptime. Over a year, each percentage point is worth roughly 3.65 days. The difference between 95% and 99% is therefore about a fortnight of hashing per machine per year – and over a seven-year hardware life, closer to three months.

That framing matters because uptime is usually discussed as a badge rather than a quantity. Treat it as time. A host offering 99% instead of 95% is offering you roughly fourteen extra days of hashing a year, and you can price that against whatever the rate difference is. If the higher-uptime host charges meaningfully more per kilowatt-hour, the trade is not automatically in your favour.

There is a second subtlety. Uptime is not the same as full performance. A miner that is powered on but thermally throttled is ‘up’ by most definitions while producing materially less hashrate. This is why the temperature trend and the hashrate-versus-rated-speed comparison matter alongside the uptime figure, and why any host that reports only a single availability number is giving you an incomplete picture.

The industry yardstick, and why it does not transfer

When hosting providers quote availability, they are borrowing the language of the data-centre industry. The reference framework there is the Uptime Institute’s Tier classification, which is genuinely rigorous and worth understanding because it is the standard your host is implicitly comparing itself against.

  • Tier I – basic capacity, no redundancy. Roughly 99.671% availability, or about 28.8 hours of downtime a year.
  • Tier II – redundant capacity components. Roughly 99.741%, about 22 hours a year.
  • Tier III – concurrently maintainable, N+1 redundancy, multiple distribution paths with one active. 99.982%, about 1.6 hours a year.
  • Tier IV – fault tolerant, 2N+1, independent power and cooling paths. 99.995%, about 26.3 minutes a year.

Here is the catch, and it is the single most misunderstood point in hosting comparisons. Tier certification describes the resilience of the facility infrastructure – power paths, cooling paths, redundancy. It does not describe the availability of your individual machine. A Tier III building can deliver 99.982% infrastructure availability while your specific Bitcoin miner is offline for a week with a failed power supply, because the building did its job and the machine did not.

So when a mining host quotes a data-centre-style figure, ask the follow-up: is that the facility’s power availability, or my miner’s operational availability? Those are different products. The first is about the building. The second is about the building plus the hardware plus the people who fix it – and only the second one earns you anything.

The three levels of uptime promise

In practice, every uptime claim you encounter falls into one of three tiers of seriousness. Only the third is worth paying for.

Level one is a marketing number. The website says ‘99.9% uptime’ and nothing else exists – no contractual commitment, no published measurement method, no consequence for missing it. This is a design element, not a guarantee. It costs the host nothing to print and nothing to break.

Level two is a facility average. The host genuinely measures something, but it is aggregate availability across the whole site. This hides exactly the failure you care about. If a host runs ten thousand machines and yours has been dark for two weeks, the site average barely moves. You will read a healthy number in a monthly report while earning nothing.

Level three is a per-miner figure you can see yourself. The host measures your specific machine, the number is written into the agreement, and you can check it independently in an app rather than waiting for a report. This is the only level where the promise and the thing you are buying are the same object. Ask for it explicitly, because hosts that offer it will say so immediately and hosts that do not will change the subject.

The exclusions clause: where guarantees quietly die

The percentage gets the attention. The exclusions decide what it means. Before comparing two headline figures, find the clause that defines what does not count as downtime, because a 99.9% guarantee with wide exclusions is weaker than a 95% guarantee with narrow ones.

  • Planned maintenance – almost always excluded. Ask how much notice you get and whether there is an annual cap on maintenance hours.
  • Grid events and utility interruptions – frequently excluded. This is the big one, because grid interruption is a leading cause of real downtime.
  • Force majeure and severe weather – reasonable to exclude, but ask how broadly ‘severe’ is defined.
  • Pool-side or network problems – usually excluded, and usually fairly, since the host does not control your pool.
  • Customer-supplied hardware faults – often excluded or treated under different terms than hardware bought through the host.
  • Curtailment for grid demand-response – increasingly common at large sites, and worth asking about explicitly.

Once planned maintenance, grid events, weather and pool issues are all carved out, a 99.9% promise can describe a machine that was actually hashing considerably less than 99.9% of the year. None of these exclusions are dishonest in isolation – most are industry-normal and some are entirely reasonable. The problem is comparing two headline numbers as though they mean the same thing when the definitions underneath them differ.

There is also the remedy question. A guarantee without a consequence is a preference. Ask what specifically happens if the host misses the floor: a service credit, a rebate against hosting fees, a defined escalation? ‘We would make it right’ is not a service level.

Repair distance: the variable nobody advertises

Two hosts can publish an identical guarantee and deliver wildly different outcomes, because the percentage says nothing about how quickly a broken machine returns to service. Consider two failures of the same component. At the first host, a spare power supply is on the shelf and a technician swaps it in two days. At the second, the miner is crated and shipped to a regional service centre and comes back five weeks later. Same promise, same failure, thirty-three extra days of nothing.

The failure list in mining is short and repetitive: power supply units, fans or coolant pumps, hash boards, control boards, and firmware faults. Most are replaceable parts rather than write-offs. That is precisely why local spares inventory is the single highest-leverage thing a host can invest in, and why it is worth asking which parts are physically held at the site your hardware will live in.

Automation matters just as much at the front of that chain. A large share of stoppages clear on a restart. If detection and restart are automatic, those events resolve in minutes without anyone raising a ticket. If they depend on a human noticing, an overnight stop becomes an overnight loss. Ask whether monitoring restarts a stopped machine on its own, or whether that is your job.

What we guarantee, and what we actually observe

We publish two numbers deliberately, and the gap between them is the honest part. Across the OneMiners network – 20 sites and roughly 2,163 MW of contracted capacity – the contractual floor is 95%+ and the observed operating figure is 98%+. Every site carries a 7-year hardware warranty, 0% management fees, all-inclusive electricity pricing averaging $0.0480/kWh, and remote monitoring and control through the iOS and Android app.

Setting the guarantee below the observed figure is intentional, not conservative marketing. A contractual floor has to be a level the operator can hold in a bad month – through a heatwave, a grid event, a maintenance window – not the number from a good quarter. Any host guaranteeing a figure identical to its best observed performance is either very new or not planning to honour it.

The operational side is what makes those numbers hold: AI-driven monitoring that detects a stopped miner and restarts it automatically, repair centres at the facilities themselves plus continental hubs in the USA, EU and Asia, dead-on-arrival protection for hardware that fails on delivery, and a liability fund behind the commitments. Uptime is not a promise you write – it is a supply chain you build.

How to audit any host’s uptime claim in ten minutes

You do not need technical expertise to test whether an uptime claim is real. You need six questions and a willingness to notice which ones get vague answers.

  • Is the figure in the contract, or only on the website? Ask them to point at the clause.
  • Is it measured per miner or per facility? If per facility, the number is not about you.
  • Can I see my own machine’s uptime myself, whenever I want, without asking anyone?
  • What is excluded from the definition of downtime? Ask for the list in writing.
  • What is the remedy if you miss the floor, and is it automatic or do I have to claim it?
  • What is the typical time from fault to running again, and are spare parts held on site?

Question three is the fastest filter. A host that can show you live per-miner status has already built the measurement infrastructure that makes the other answers meaningful. A host that cannot is asking you to trust a number that nobody outside the company can verify.

Our verdict

The right answer to ‘what uptime should a mining host provide?’ is not a percentage. It is a structure: a floor of at least 95% written into the agreement, measured per miner, with the exclusions listed explicitly, a named remedy when it is missed, and live visibility so you can check the claim yourself. A host that provides all five can guarantee 95% and be more reliable in practice than one advertising 99.9% with none of them.

If you are comparing providers right now, rank them on verifiability before you rank them on the headline figure. The percentage is the cheapest thing on the page to change. The measurement infrastructure, the local spares and the repair centres behind it are what actually keep your Bitcoin miner hashing – and those are considerably harder to fake than a number in a hero banner.

Frequently asked questions

What uptime guarantee should a Bitcoin mining host provide?

At minimum a written 95% floor, measured per miner, with the exclusions listed and a named remedy if it is missed. A contractual 95% with live per-machine visibility is worth more than an advertised 99.9% with no measurement behind it. OneMiners guarantees 95%+ and observes 98%+ across 20 sites.

Is 95% uptime good for mining hosting?

As a contractual floor, yes. 95% is roughly 18 days a year, which sounds high until you understand that a floor is set to be survivable in the worst month, not the average one. What matters is the observed figure alongside it – ours is 98%+ – and whether you can verify it yourself.

What is the difference between guaranteed and observed uptime?

The guaranteed figure is the contractual floor with consequences attached, so operators set it conservatively. The observed figure is the measured result over time and is normally higher. A host quoting only one of the two is leaving out half the picture – ask which one you are being shown.

Do Uptime Institute tiers apply to Bitcoin mining hosts?

Only partly. Tier III (99.982%) and Tier IV (99.995%) certify the resilience of facility infrastructure – power and cooling paths – not the availability of your individual Bitcoin miner. A Tier III site can hit its infrastructure target while your machine sits offline awaiting a part.

What is usually excluded from a hosting uptime guarantee?

Commonly planned maintenance, grid and utility interruptions, severe weather, pool-side problems, and sometimes curtailment for grid demand-response. Most exclusions are industry-normal, but they change what the headline percentage means – always read the list before comparing two numbers.

Compare a guarantee you can actually verify: 95%+ contractual uptime, 98%+ observed, per-miner visibility in the app, and repair centres on site across 20 facilities.

Informational only, not financial advice. Electricity rates are costs per kilowatt-hour, not earnings, and rates, capacity and service terms change – verify current figures on the linked pages before committing.