Dieser Artikel ist auf Englisch.
What the study compared
Hashrate Index set three kinds of machine against the same megawatt of power. The three were the most efficient Bitcoin ASICs, Bitmain's Antminer Z15 Pro mining Zcash, and a rented NVIDIA H100 GPU. It priced all three on a single day, 13 October 2026, at one power cost. Its own summary is that "the winner depends on the denominator": revenue per unit of energy gives one ranking, and revenue per dollar of hardware gives another.
One detail is easy to miss. A GPU's board power is lower than the power its data centre actually draws once cooling and overheads are counted. The study measured the H100 on facility power, because an ASIC's wall draw already includes its own fans and power supply. If you compare a GPU on board power with an ASIC on wall power, the GPU looks far stronger than it is.
Per megawatt-hour: the GPU leads while it is rented
On revenue per megawatt-hour, a fully rented H100 came first. That lead depends on the GPU staying busy. The study puts the crossover with the Z15 Pro at roughly 46 to 48 percent utilisation. Below that, the Zcash machine earns more from the same energy. Lower rental rates narrow the gap quickly, so moving from ASICs to GPUs is a bet that both the rental rate and the utilisation stay high.
The Zcash ASIC was ahead of Bitcoin hardware on this measure on that date. The source explains why: the ZEC price rose much faster than the Zcash network's hashrate over the previous year. Zcash miners share a fixed daily reward, so every new machine dilutes all the others. The source also says the whole Zcash network is about the size of 33,000 Z15 Pro units. That is a small network, and its economics can shift sharply in either direction.
For Bitcoin, revenue per unit of energy follows hashprice, which moves with BTC price, fees and difficulty. Today's values are below.
- Hashprice
- $0.0327 pro TH/s und Tag
- Netzwerk-Difficulty
- 148,4 T
Per dollar of capital: ASICs lead
Divide by hardware cost and the order flips. The study found that a megawatt of new H100s costs several times as much as a megawatt of Z15 Pros, even before the building. GPUs for rental also need a Tier-3 style data centre, which can cost more than the GPUs themselves. ASICs run in a much simpler hall with good airflow or water cooling, stable power and network access.
Bitcoin ASICs share that capital profile. They are compact, they draw power at a steady rate, and they fit into the kind of air-cooled and hydro-cooled halls used for hosting in the UAE. If you are weighing cooling options, see our guide to hydro, air and immersion cooling.
Why a one-day snapshot can move
Each machine's position rests on assumptions that change:
- GPU rental rates and utilisation decide whether an H100 leads or trails, and a rented GPU needs a steady customer.
- Zcash depends on one coin's price and a small network. In-stock Z15 Pros were scarce, and the source does not expect new Equihash silicon before the second quarter of 2027.
- Bitcoin has the largest mining network, several generations of hardware, many pools, an active resale market for used machines and an established hosting industry. Its revenue moves with hashprice, but the market around it is deep.
The study also notes that firmware can tune a machine for efficiency or for raw output. Only the most efficient profile raised revenue per megawatt-hour. Overclocking raised dollars per machine but lowered dollars per unit of energy. The same trade-off applies to Bitcoin ASICs.
What this means for a Bitcoin miner buyer in the UAE
If you own or plan to buy SHA-256 machines, the useful lesson is the method, not the ranking. Judge your own fleet per megawatt-hour, because power is the input you pay for every month. Efficiency in joules per terahash is the main lever, as we explain in checking your rig's efficiency. Uptime is the other lever: a stopped machine earns nothing, just as an idle GPU earns nothing. Ask any host what uptime it targets and how it measures it.
Watch hashprice over months, not days. Our piece on what a hashprice recovery means shows how to read it without chasing short swings.
Questions to ask before moving power to other compute
- Am I comparing revenue on the same power basis, facility power for every machine?
- What utilisation can I realistically expect, and who is the paying customer?
- How large and liquid is the network or market behind this hardware?
- Can the hardware be resold if conditions change?
- What building, cooling and staffing does it need, and what do they cost?
Fragen
- Why does a GPU earn more per megawatt-hour but cost more per megawatt?
- A rented H100 brings in a lot of revenue for each unit of energy, but the GPU and the data centre it needs cost far more than ASICs and a simple mining hall. Energy favours the GPU while it is rented. Capital favours the ASIC.
- Does overclocking a miner improve revenue per megawatt-hour?
- Usually not. In the study, only the most efficient firmware profile raised revenue per megawatt-hour. Higher overclock profiles raised output per machine, but power rose faster than hashrate, so revenue per unit of energy fell. Whether that trade suits you depends on your power cost and on whether your site is short of power.
- Should I switch my Bitcoin miners to Zcash or AI compute?
- The comparison is a single-day snapshot. Zcash depends on one coin and a small network, and GPU income depends on rental rates and utilisation. Bitcoin has the deepest mining network, an active resale market and established hosting. Judge your own machines on efficiency, uptime and revenue per megawatt-hour before acting on any headline ranking.
Quellen
- The Compute Competition · Hashrate Index (Luxor Technology), 2026-10-15