What an Offline Hashboard Really Costs at 2026 Hashprice: A Formula, Not a Guess
When a hashboard goes down, most operators think in terms of lost revenue. That is half the picture. A machine that is offline also stops burning electricity, so the real loss is the difference between what the machine would have earned and what it would have cost to run. At the hashprice the market saw in the week ending November 2, 2026 ($32.86 per PH/s per day, per the Hashrate Index roundup), that difference is much smaller than most spreadsheets assume, and it varies enormously between models.
That sounds like good news for an operator waiting on a part. It is not, for two reasons we explain below. But it does change how you should think about downtime, spare parts and repair turnaround. This article gives you a formula, a worked table for four common models, and the reasoning for when holding stock is worth the capital.
Two numbers, not one
Every offline day has a gross cost and a net cost.
- Gross cost is the revenue the machine would have earned that day. It is what the pool would have paid you. This is what you lose in the top line.
- Net cost is gross revenue minus the electricity you did not use. It is what you actually lose in profit, if your power cost is genuinely variable.
The distinction matters because many sites do not have fully variable power costs. If you pay a fixed capacity charge, a flat hosting fee or a take-or-pay contract, your net cost of downtime moves closer to the gross figure. If you buy power by the kWh, the net figure is the right one.
The formula
- Revenue per day = hashrate in PH/s x hashprice in $ per PH/s per day
- Energy per day = power in kW x 24
- Gross cost per offline day = revenue per day
- Net cost per offline day = revenue per day minus (energy per day x your power price)
- Total cost of a repair delay = net cost per day x number of days offline
For a three-board machine with one dead hashboard, the machine usually keeps running on the other two, so only about a third of the hashrate is lost. Power falls by less than a third because the fans and control board keep drawing. If you want precision, measure the draw with one board out. For planning, scale the revenue by the share of hashrate you lose.
Worked example at $32.86 hashprice
The table applies the formula to a whole machine being offline for one day. Specifications are rated figures. These are our calculations, gross of pool fees and hosting margin.
| Model | Gross cost per offline day | Net cost at $0.05/kWh | Net cost at $0.07/kWh | Net cost at $0.09/kWh |
|---|---|---|---|---|
| Antminer S19j Pro (92 TH/s, 2,714 W) | $3.02 | -$0.23 | -$1.54 | -$2.84 |
| Antminer S21 (200 TH/s, 3,500 W) | $6.57 | $2.37 | $0.69 | -$0.99 |
| Antminer S21 XP (270 TH/s, 3,645 W) | $8.87 | $4.50 | $2.75 | $1.00 |
| Antminer S23 (318 TH/s, 3,498 W) | $10.45 | $6.25 | $4.57 | $2.89 |
A negative net cost means the machine would have lost money that day, so being offline saved you money. Read the table with that in mind. On an S19j Pro at 7 cents, a week of downtime does not cost you profit. It saves you about $11. On an S21 XP at the same price, a 30-day wait costs about $82 in profit and $266 in gross revenue. On an S23 the same wait costs about $137 in profit.
Why the low net figure is not a reason to relax
Three things make the small net numbers misleading if you stop there.
- Hashprice moves. The same S21 XP at 7 cents earns $2.75 per day today. Hashprice averaged $37.49 over the prior 30 days, and at that level the margin is higher. The margin you lose to downtime scales with hashprice, and you do not know in advance which way it will go. The futures-implied average for the next six months is about $33, which is a market price, not a promise.
- Your best machines carry the fleet. At these margins, the units that clear their power cost by the widest margin are the ones paying for everything else. Losing one of them for a month is a bigger share of your total profit than the table suggests.
- Delay compounds. A part that takes several weeks to arrive turns a small daily loss into a larger total, and it ties up a machine that could have been running when the market turns.
Turnaround time is the variable you control
The formula has one input you can change directly: the number of days. A repair that depends on shipping a board to a service centre and back adds transit and queue time. A repair done on your own bench with parts already in stock adds none. We are not going to quote a typical turnaround for third-party services here, because it varies widely by provider, and vendor figures are usually marketing. What you can do is substitute your own numbers: your actual turnaround today, and the turnaround with a stocked bench. The difference, multiplied by your daily net cost, is the value of holding stock for that model.
For the S21 XP at 7 cents and $2.75 per day, every week you remove from a repair is about $19 of profit and about $62 of gross revenue. If your standing stock for that model costs less than the weeks it saves across a year of failures, it pays for itself. If the stock for a low-margin model costs more than the profit it protects, it does not.
Which models justify standing stock
- High-margin models (S21 XP, S23-class): the net cost of downtime is meaningful and rises quickly if hashprice recovers. Hold fans, PSUs, connectors and the common repair components. See the cooling fans, power supplies and repair components collections.
- Mid-margin models (S21 around 5 to 7 cents): the margin is thin and sensitive. Stock the cheap items that fail often. Be selective about big-ticket parts.
- Negative-margin models (S19j Pro above about 5 cents): the right plan is usually curtailment or retirement, not stock. Spending on them extends losses.
Once you have your list, our 12-month standing stock budget template shows how to size quantities by fleet. For a PSU decision specific to your generation, see APW17 vs APW12.
Fleet view: a 100-machine example
Suppose a farm of 100 S21 XP machines at 7 cents, with 5 percent of machines each losing one of three hashboards for 14 days at a time. That is 5 machines losing a third of their hashrate. Using our table, the net cost per machine-day of full downtime is $2.75, so a third of that for 5 machines over 14 days is about $64 in profit. It is a small number, which is the honest point. At this margin, an individual failure costs little; what hurts is a broad failure wave or a long delay on many units at once, and a market where hashprice moves up while you are still waiting.
Run the same sums with your own failure rate, your own fleet and your own power price. The point of the exercise is to know your number, not to borrow ours.
FAQ
Does a dead hashboard stop all power use?
No. On a machine with several hashboards, the remaining boards, fans and control board keep drawing power. Measure your own draw if you need accuracy.
Should I use gross or net cost?
Net, if your power is billed by the kWh. Gross, if your cost is fixed regardless of whether the machine runs. If you are unsure, calculate both and look at the range.
Is it worth buying parts for a machine that is barely profitable?
Only for cheap, high-frequency parts such as fans and connectors, and only if you expect margins to hold. For anything expensive, work out the payback first.
Cut your repair time
Tell us your models, your fleet size and your usual failure pattern, and we will help you build a stock list sized to your margin, not to a generic checklist. Browse the Antminer repair parts collection or contact us directly.
Email: contact@lys-sz.com
Direct from our warehouses in Shenzhen. Worldwide shipping. DDP available for buyers in the US and the EU. Latam, SE Asia, Middle East, CIS, Africa quoted case by case.
Market data: Hashrate Index weekly roundup, week ending November 2, 2026. All dollar figures in this article are our own calculations from rated specifications and that hashprice, before pool fees and hosting margin.
