What mining actually costs.
The cost side of a deployment is knowable and I will set it out line by line. The production side is not, and this page deliberately does not contain a table pretending otherwise.
3 min readCosts are knowable. Output is not.
This is the distinction most cost breakdowns blur, usually on purpose. What you will pay for hardware, hosting and fees can be established with reasonable confidence before you commit. What the hardware will produce cannot, because it depends on two things that move on their own and one that varies.
So this page covers the first half properly and refuses the second half. If that is frustrating, it is worth asking why every other page on this subject is willing to give you a number to the nearest dollar for something nobody can know.
The cost lines.
Hardware
Current generation units run roughly $5,000 to $8,000 each depending on model, efficiency and order size. This is the number that moves most with market conditions, and quotes go stale quickly.
Hosting
Quoted in dollars per kilowatt hour against the machine’s power draw. I model $0.08, which is the market rate for hosted capacity. It is a bundled service fee covering rack space, power, cooling, network and security, not an electricity bill passed through at cost.
My fee
3% of pool output, taken as a split at the pool. It is never invoiced to you and it only exists when the machines produce.
Consumables and repairs
Fans, power supplies and hashboards on machines running continuously. Budget for this as a recurring cost rather than an exception.
Professional fees
Accountancy and the cost of forming or maintaining the entity, which is a real line item and is routinely left off comparisons.
The one nobody lists
The capital is illiquid. It is in equipment with a resale market that is thin and correlated with exactly the conditions that would make you want to sell.
Why there is no production table here.
To produce an annual BTC figure you have to assume a network difficulty path, a Bitcoin price, an uptime percentage and a pool outcome. Three of those four are unknowable and the fourth varies.
Difficulty rises as hashrate comes online and the same machine produces less each month as a result. A table that holds difficulty static for a year is not conservative, it is wrong, and it overstates production by a widening margin the further out it runs. A table that also runs to fifteen years and applies a favourable power rate can be made to show almost anything.
There is a second reason, which is that I do not publish production or return projections at all. A specific figure presented on a marketing page is a representation about future performance, and it is not one I am willing to make about an asset that behaves like this one.
The four inputs that actually decide it.
If you want to understand the economics rather than read a conclusion, these are the levers. Change any one and the answer moves.
Efficiency
Joules per terahash. This decides how much of your production is consumed by the power bill, and it decides how long the machine stays economic as difficulty rises.
The power rate
The largest recurring cost. Treat anyone quoting materially below the market rate as someone with something bundled or omitted.
Difficulty growth
I model 10% a year and state it. A trend fitted to recent history returns figures near 37%, which is plainly wrong given that period included difficulty falling.
Machine life
Three to five years, sooner at lower prices. I model five years because machines reach end of life around sixty months and longer horizons add dead years to flatter the chart.
The number worth having.
The useful output of this exercise is not a projected return. It is the Bitcoin price the deployment needs in order to work, given your hardware, your rate and your horizon. That is a single number, it falls out of the arithmetic, and you can then decide whether you believe it.
It is also the number that tells you when to stop. If the price sits below it for long enough, the rational response is to switch machines off rather than run them at a loss, and knowing that threshold in advance is worth considerably more than knowing what year three looks like in a good scenario. I would rather send you away with that figure than with a table.
Run it against your own inputs.
An hour on your capital, your entity and your timeline, with every assumption written down and open to challenge. The output includes the price the plan needs in order to work. You can also read the hardware and what can go wrong.
Book a strategy session Paid session. Assumptions stated. A genuine answer, including no.