Compass Mining's hosting SLA guarantees 95% uptime. Read the definition and it says something more specific: downtime "is calculated by ping, not hashrate. Hashrate, like cryptocurrency rewards, is not guaranteed and therefore cannot be reimbursed."
That single sentence is the whole problem with uptime in mining hosting. The host measures whether the building has power. You earn money on whether your machine is hashing. Those are different numbers, and the gap between them is where your revenue goes.
This guide covers what the SLAs actually promise, what realistic uptime looks like with figures, what a percentage point of downtime costs, and — most usefully — how to monitor your own machine so you find out before your host does.
Facility uptime versus machine uptime
Every hosted-mining SLA we have read measures the facility, not your hardware.
- Compass's status page states plainly: "Issues for individual ASIC machines are not shown on this page." If your unit is dead and the site is not under maintenance, you have to notice and raise a ticket.
- Compass advises waiting two hours before opening a ticket, and puts the burden of reporting on the customer.
- Underhashing and pool misconfiguration are excluded from credits entirely.
The nastiest version is thermal throttling. When intake air gets too warm, an ASIC does not shut down and does not raise an alarm — it quietly drops hashrate, often by 10–30%. A ping-based SLA scores that as 100% uptime while you lose a third of your revenue. Stock cooling is rated for roughly 25–35 °C ambient; above 40 °C it cannot keep up.
One provider measures the right thing: Sazmining offers a rig performance guarantee measured against nameplate hashrate rather than power availability. That is the model to look for.
What uptime hosted miners actually achieve
Published retail claims cluster at 95–98%. Compass commits to 95%. Simple Mining says quality providers commit to 95–98% in writing. We could not verify a published retail uptime percentage for Blockware, Luxor or Core Scientific.
Fleet-level data tells a harsher story. In June 2025, MARA's realised hashrate was 82.11% of its energised fleet, and Cipher ran at 62.95% utilisation. Those are sophisticated operators running their own sites.
For context on how far this sits from conventional hosting:
| Standard | Uptime | Downtime per year |
|---|---|---|
| Mining host floor (Compass) | 95% | 438 hours (18 days) |
| Uptime Institute Tier III | 99.982% | 1.6 hours |
| Uptime Institute Tier IV | 99.995% | 26 minutes |
| Enterprise colo (redundant power) | 99.999% | Under 5 minutes |
A 95% mining SLA permits roughly 275 times the downtime of a Tier III datacentre. Simple Mining's founder has said as much publicly: traditional datacentres "require five nines… you're not going to get that out of the standard bitcoin mining facility as of today."
What downtime actually costs
Take an Antminer S21 Pro at 234 TH and 3,510 W, at a hashprice of $31.89/PH/day (17 August 2026):
- Gross revenue: $7.46/day, about $2,724/year
- Power at $0.06/kWh: about $1,845/year
- Net profit: roughly $880/year
Now the downtime:
| Downtime | Hours/year | Gross revenue lost | As % of annual profit |
|---|---|---|---|
| 1% | 88 | $27 | 3% |
| 5% (the 95% SLA floor) | 438 | $136 | 15% |
| 18% (MARA's June 2025 gap) | 1,577 | $490 | 56% |
The headline revenue loss looks small. Measured against profit — which is what you actually keep — it is severe. And if your host bills a flat monthly rate rather than metered power, add the electricity you paid for and did not use.
This is why the billing model matters more than the SLA percentage. At our own rates from $0.014/kWh the power cost falls to roughly $430/year on the same machine, which widens the profit buffer considerably — see the profitability ranking for live figures.
What actually causes hosted downtime
Ranked by how often we see it:
1. Curtailment and demand response. Comfortably the most frequent cause. Tallying Compass's own public status log for August 2026: their Minnesota 2 site posted 8 curtailment notices totalling around 40 hours between 7 and 21 August, one window running 12:00–21:00. Texas 5 logged 6 events of roughly 3 hours each between 12 and 20 August. On 23 August 2026, 4 of 21 listed sites showed "Major Outage" and 2 more were under maintenance — about 29% not fully operational.
2. Facility outages and maintenance. The same log shows an unplanned North Dakota substation outage running from 17 to 21 August — over four days.
3. Heat. Both as triggered curtailment (Compass North Dakota, 10 August 2026: "due to elevated local temperatures, the facility will partially curtail") and as silent derating.
4. Hardware failure and the RMA queue. Compass credits RMA time only after the machine returns, and states "you will continue to be invoiced during the RMA/repair period." Hosts with in-house repair cut this from weeks to days.
5. Pool and configuration problems — excluded from credits at Compass and expressly disclaimed in institutional contracts.
6. Nobody noticing. The most preventable one, and the reason for the monitoring section below.
Curtailment: why your machine gets switched off
Miners are curtailed because curtailing pays better than mining. Riot Platforms earned $10.05 million in power curtailment credits in Q2 2026 and $31.08 million across H1 2026, credited against power invoices for pausing under ERCOT demand-response programmes.
The mechanism that drives Texas curtailment is 4CP: four 15-minute peaks across June to September set a site's transmission charges for the following twelve months. Exposure runs to roughly $50,000 per MW per year — a 25 MW site facing all four peaks at full load owes about $1.25 million. Against that, an hour of foregone mining is trivial.
The effect is visible on-chain. 2025's 4CP peaks fell between 16:00 and 18:00; during likely-peak hours in June 2025, block times ran about 51 seconds slower, and the month produced two negative difficulty adjustments (−0.45% and −7.48%). Across 2022–25, June-to-September difficulty adjustments averaged +0.53% against +2.34% for the rest of the year.
The retail catch: the host collects the demand-response payment. You absorb the lost hashrate. Ask explicitly whether curtailment revenue is shared — and see our guide to hosting contracts for how these clauses are written.
One honest limitation: no public dataset reports miner curtailment in hours per year. ERCOT, Riot and MARA disclose dollars, megawatts and credits — not hours. Anyone quoting an hours figure is estimating.
What an SLA actually pays you
Where credits exist, they are narrow:
- Compass credits hosting fees only — "customers will not receive compensation/reimbursement for lost mining rewards."
- Credits are non-transferable, cannot be converted to cash, and arrive one to two months after month end.
At institutional scale there may be no uptime commitment at all. The Stronghold–Bitfarms hosting agreement filed with the SEC lets the host curtail for maintenance, hazards, force majeure "or if Stronghold determines that it is… more economical for all Parties to sell power to the grid," and disclaims liability for "ANY LOST REVENUE… DURING OUTAGES, CURTAILMENT, EQUIPMENT FAILURES" including pool operator failures.
Compare a conventional colocation SLA: 99.999% redundant power, temperature defined as 18–27 °C and guaranteed at 99.99%, credits of 1/30th of the monthly charge per event — and, crucially, three credit events in 90 days allows immediate termination without early-termination fees. Mining hosting contracts rarely contain an equivalent escape hatch.
Monitor your own machine — pool-side
Your mining pool sees your hashrate directly. That view cannot be shaped by the host, and it is the only monitoring that catches thermal derating.
- Braiins Pool sends email or mobile alerts when a worker's effective hashrate drops below a limit you set — automatically from past performance, or manually.
- ViaBTC lets you filter for offline or inactive workers. Occasional dips self-correct; a drop persisting 15–30 minutes signals a real failure.
- Set the alert on a hashrate threshold, not just offline status. An offline alert never fires on a machine that is running at 70%.
- Subscribe to your host's status page so you can distinguish announced curtailment from an unnoticed dead machine.
- Log everything with timestamps. Credit claims are usually time-barred — the enterprise colo benchmark requires a written request within 14 days. Mining hosts vary, but the burden is always on you.
What recourse you realistically have
Honestly assessed, from best to worst:
- Suspension of billing. The most valuable clause, and it exists — the Stronghold agreement provides that during suspended service the customer "shall not be liable or responsible for any obligations under this Agreement." Negotiate for this rather than credits.
- Metered billing. Paying only for power actually consumed while hashing removes the argument entirely. Simple Mining and Sazmining both work this way.
- Fee credits. Real but narrow — hosting fees only, delayed, never lost coins.
- Litigation. Not a realistic retail remedy. Consequential-damages waivers are near-universal, and the 2022 suit alleging Blockware displayed "100% uptime" while a facility sat curtailed for around 50 days remains unresolved years later.
- Leaving. In practice this is the remedy. Month-to-month terms and the ability to physically retrieve your machine matter more than any percentage in a PDF.
Seven questions to ask before you sign
- Give me monthly uptime by facility for the last 12 months — measured by ping or by hashrate?
- Is the SLA on the facility or my machine? Does underhashing count? Does thermal derating count?
- Is the hosting fee waived or suspended during downtime, or do I pay and claim a credit later — and how much later?
- Who monitors, at what interval, and what is your response time to a dead machine?
- Do you share curtailment revenue, and how many curtailment hours did this site log last summer?
- What is the RMA turnaround, is repair in-house, and am I billed while the machine is out?
- Is there a chronic-outage exit — does repeated breach let me leave without penalty and collect my hardware?
A provider that answers all seven in writing is worth more than one quoting a lower rate and a higher percentage.
How we handle it
For transparency: our facilities publish per-location uptime on the locations table, currently 98.4–99.8% depending on site. Power is billed per kWh actually consumed and deducted from mining earnings, not invoiced as a flat monthly fee — so a machine that is not hashing is not accruing power charges. Your miner points at your own pool account, which means you can monitor hashrate independently of anything we tell you.
We do not guarantee returns, and no host can guarantee hashrate. What we can do is not bill you for power you did not use.
Metered power, your pool, your coins
Buy a Antminer S21 or S23 Hyd and host it from $0.014/kWh. You mine to your own wallet and see your own hashrate.
Read next: what hosting contracts actually say, why ASIC prices fell in 2026, or the hosting cost breakdown. SLA terms cited from Compass Mining support documentation and the SEC-filed Stronghold–Bitfarms hosting agreement; curtailment figures from Riot Platforms' Q2 2026 results and Luxor's Hashrate Index; fleet utilisation from TheMinerMag. Dollar figures computed at a hashprice of $31.89/PH/day (17 August 2026) and move with hashprice.