Every buyer opens the conversation the same way: what is the best machine right now. The honest answer is that the question has no content until four other things are known, and once they are known the machine usually picks itself.
1. What do you pay for power?
This is the whole decision compressed into one number, and it sorts the catalogue more sharply than anything a manufacturer prints.
Under about three cents a kilowatt-hour, efficiency stops being the binding constraint. An older machine at 23 J/TH still clears its energy cost, and the cheapest terahash you can buy wins — which is exactly why an S19k Pro at $1,150 keeps finding buyers in Central Asia. Capex per terahash is what you optimise.
Above about six cents, the ranking inverts. Energy dominates the lifetime cost, the efficient machine pays back its premium inside a year, and buying a cheap inefficient unit is buying a liability with a power bill attached. At $0.09 the S19k Pro barely breaks even; the S21 XP is still comfortably profitable.
Between three and six cents is where the argument actually happens, and where we spend most of our time with spreadsheets.
2. How much power do you have, and at what voltage?
Site capacity sets the shape of the order before it sets the model. Two megawatts is roughly 550 air-cooled S21 XP units, or 373 hydro units, and those are very different buildings.
Voltage matters more than buyers expect. The Bitmain air line wants 200–240 V; MicroBT’s M60 series tolerates 200–277 V, which sounds like a footnote until the supply sags at peak and one hall keeps hashing while the other throttles. If your grid wanders, the wider window is worth real money.
Hydro machines want 380–415 V three-phase and a loop. If neither exists, they are not a candidate regardless of how good the J/TH looks.
3. What is the design ambient?
Air-cooled units are specified to 40 °C and mean it. On a Gulf site with 45 °C summer afternoons you will lose six to nine per cent of rated hash rate for four months of the year, and you should build that into the model rather than discover it in July.
That derate is not a reason to avoid air. It is a reason to quote air honestly: a machine that gives you 91% of its rated output through summer and 100% the rest of the year is still, very often, the better buy than a hydro hall nobody on site can maintain.
4. How long do you intend to own it?
A machine bought for a two-year horizon and a machine bought as a ten-year asset are different purchases. Short horizon: buy the cheapest terahash that clears the tariff, and plan to sell into the next cycle. Long horizon: buy efficiency, because the machine that survives the next difficulty era is the efficient one, and resale tracks J/TH more tightly than it tracks brand.
Where that lands, in practice
| If your site is | The usual answer | Why |
|---|---|---|
| Cheap power, hard capex ceiling | Antminer T21, Avalon A1466 | Lowest cost per terahash; efficiency is not binding |
| Mid tariff, air only, long horizon | Antminer S21 XP, S21 Pro | Best efficiency available without a water loop |
| Unstable grid | Whatsminer M60S+ / M60S++ | Widest voltage window in the range |
| Loop and staff already in place | S21 Hydro, M66S+ | Density and quiet, once the loop is somebody’s job |
None of that is a ranking. It is four different buildings, and the machine follows the building rather than the other way round.
The thing worth insisting on, whatever you buy: get the derate assumption and the design ambient stated in writing on the quotation. A price against a rated hash rate with no operating conditions attached is a price for a laboratory, not for your site.