Not hype, but thermophysics: where air cooling ends

Every six months, air cooling is buried in mining huts. And every six months, someone installs a new batch of air S21s and quietly makes money on them. The truth, as usual, is not in the slogan, but in the numbers — and in where exactly the line is drawn between "air is still worth it" and "from now on, it's just water."

This article is about that boundary. Without funerals and without advertising: where does air cooling have a physical ceiling, why hydro breaks through it, and in which cases does it make no sense to pay extra for water.

Why does air have a ceiling at all?

The task of cooling is always the same — to remove heat from the chip faster than the chip generates it. Air can handle this as long as the heat flow is moderate. But air has a low heat capacity, so the only way to remove more heat is to move more air. That is, to spin the fans faster.

And here comes the physics that marketing can't get around: noise grows faster than heat dissipation. The air-cooled S21 at 200 TH/s is already howling at 70–80 dB — that's a vacuum cleaner in your ear, twenty-four hours a day. In order for the fans to remove 5+ kW of heat from the chip through the air, they would have to reach speeds at which the machine turns into a turbine: unacceptable due to noise, bearing wear, and the energy consumed by the fans themselves.

Water can remove the same 5+ kW from one chassis without the acoustic tax — it has a heat capacity several times higher. And that's the whole trick. Not the "hydro fad", but the different physical limits of the two media.

Numbers: where exactly is the border?

The key metric in mining isn’t hashrate, but efficiency: how many watts are used per terahash (J/TH). This determines how much of your electricity bill is converted into bitcoin and how much is wasted as heat. Less is better.

This is what the current line looks like Bitmain according to effectiveness:

Note the limit. The best air has to offer today is the S21 XP at 13,5 J/TH . The best hydro is the S23 at 9,5 J/TH . The difference is about 30% in efficiency . And it's not that "air is bad": 13,5 J/TH is an excellent result that was unattainable a few years ago. Air simply doesn't physically pull anymore, while water does.

It is characteristic that air and hydro versions are often built on the exact same chip. S21 XP and S21 XP Hydro are one generation of silicon. The only difference is that the liquid circuit keeps the chip in a stable temperature regime and allows it to extract more from it — 473 TH/s versus 270 for the air sibling, and this with better efficiency. Same chip, different ceiling — and this is thermophysics in its purest form.

Density: where the air loses not by a little, but by several times

Efficiency is the electricity bill. But hydro has a second trump card, which is often more important in practice: density.

The flagship S23 Hyd 3U delivers 1,16 TH/s in a single 3U chassis . To put that into perspective, one such block replaces roughly 40 S19 generation machines , or about 2,5 air-cooled S21 XP Hydros , or two separate S23 Hydros at 580 TH/s. Forty chassis fit into a single rack slot.

And it’s not just about space. Forty machines mean forty Ethernet connections, forty power lines, forty places where something can go wrong, and forty endpoints that need to be monitored. One S23 Hyd 3U — one connection, one power line, one endpoint on the dashboard. The real failures in farms arise precisely at the level of cables and connectors — and there are ten times fewer of them. Consolidation saves not so much watts as nerves and man-hours of maintenance.

Where the air always wins — and honestly

Now the half that commercials usually cut out. Hydro doesn't always win. There are scenarios where switching to water is a waste of money.

At home and on a small scale (1-5 machines). Here, air is a practical choice, period. Building a liquid circuit with a pump, radiator and coolant maintenance for five pieces is economically nonsense. Air S21 will do its job with normal ventilation.

Standard electrical network. Air machines operate from a standard single-phase network of 220–277 V. Hydro flagships require a three-phase industrial power supply of 380–415 V. If you do not have such a supply, it is a separate construction even before the first hash.

Initial investment and startup speed. The air-powered S21 XP costs approximately ~$3,800 versus ~$5,500 for the S23 Hydro 580 (and several times more for the 3U flagship). The air turns on right away: power, internet — and the machine is running. Hydro requires a cooling circuit, which sometimes costs more than the miner itself.

Short horizon. If the goal is to enter quickly, make a profit and eventually exit, the air will return faster due to the low entry threshold.

The honest conclusion of the industry is this: for most home miners with 1-5 machines, air cooling with proper ventilation remains the right choice. No thermophysics prevents this.

Where the game takes its water: the economy after the halving

And then the territory of hydro begins — and it is not determined by fashion, but by the mathematics of the electricity bill.

Consider the efficiency difference between the air cap (13,5 J/TH) and the flagship hydro (9,5 J/TH) — that’s 4 J/TH, or 4 kW per petahash. That’s 96 kWh per PH per day. At an electricity price of $0,05/kWh, that’s roughly $4,8 per day per petahash that the air farm is simply donating to the energy industry. Over the course of a year, that’s ~$1,750 per PH; for a 10 PH operation, that’s on the order of $17,500 per year just in efficiency difference , for the same bitcoin mined.

After the halving, the block reward is cut, and that changes everything. The less efficient pool is the first to lose money when the difficulty increases, but the reward does not. A 17 J/TH machine that was profitable last year may find itself beyond the break-even point if the difficulty increases again. A 9,5 J/TH machine will survive the increase with a margin. Efficiency stops being a line in the specification and becomes a question of the pool's survival.

In addition — stable temperatures without temperature cycling (less wear, longer life), silence at ~50 dB instead of 70–80 and the possibility of dense installation. For serious long-term operation, water wins overall.

What hydro actually requires — without embellishment

To avoid turning this article into the same old advertising text, let's get the price of things straight. Hydro is not a "set and forget" product:

• CDU (Coolant Distribution Unit) — a central hub with a pump, tank, filtration, and monitoring that supplies coolant to dozens of machines. You have to buy it, install it, and maintain it.
• Three-phase power supply 380–415 V and the correct connectors (on the flagship LP34).
• Circuit maintenance: Check the refrigerant level monthly, quality test quarterly, complete replacement every 2-3 years, filters every 6-12 months. The mixture is usually water + propylene glycol.
• Protection against leaks and loss of flow: flow, temperature, leak sensors; firmware must throttle or shut down the machine when cooling is lost.
• System incompatibility: circuits Bitmain (Antminer Hydro) and MicroBT (Whatsminer Hydro) are not interchangeable — different dimensions, connectors, fittings, and CDU.

This is why for many operators, deciding on a cooling system is more difficult than choosing the miner itself.

So where is the border?

Briefly, without passwords:

• 1–5 machines, at home, regular network, short horizon → air. The S21 XP at 13,5 J/TH is an excellent machine and you don't need water.
• Farm, industrial power supply, long horizon, expensive electricity, density and silence important → hydro. A 30% difference in efficiency and a tenfold consolidation will pay for the infrastructure.

“Air is dead” is a lie. The truth is that air has a clear physical ceiling around 13,5 J/TH, and everything above that — in both density and efficiency — begins where air ends and water begins.

How is this handled in practice?

The main barrier to hydro is not the miner itself, but the infrastructure: three-phase, CDU, circuit, coolant maintenance. Hosting is taking care of that. On OneMiners platforms, flagship hydro machines like the S23 Hyd 3U run in a ready-made data center environment — with three-phase power, industrial cooling, and real-time monitoring — for electricity prices starting at $0,04/kWh. So you get 9,5 J/TH and a petahash from one chassis without having to build a gas station or three-phase power supply.

Simply put: you don't need to start building where the air ends. Just build the machine where the water is already flowing.