🟠 Introduction:

Most mainstream coverage of Bitcoin mining economics boils down to a single variable: electricity. The narrative is clean, simple, and almost entirely incomplete. Power matters. Nobody disputes that. But if you’ve ever run a fleet of ASICs, you know that electricity is only the most visible cost eating into your margins. It’s rarely the only one that determines whether an operation survives a bear market or quietly shuts down.

The real variable cost structure of Bitcoin mining is layered, interdependent, and far less intuitive than “cheap power = profit.” Understanding how these costs interact is the difference between modeling mining on a spreadsheet and actually operating through a full halving cycle.

I’ve been mining Bitcoin since 2019, from home mining to larger scale. Every week I break down mining economics, fee markets, and what actually moves profitability. Subscribe for free to get new insights straight to your inbox.

🟠 Electricity Is the Floor, Not the Ceiling

Let’s start where everyone starts: power.

Electricity is the single largest variable cost in Bitcoin mining, typically representing 60–80% of total operating expenses depending on your rate, efficiency, and scale. At $0.05/kWh, running an Antminer S23 at 3,498 watts pulls roughly $4.20 per day in electricity alone. Scale that to a hundred units and you’re burning through $12,600 monthly before you’ve paid a single person or replaced a single hashboard.

But here’s what the “just find cheap power” crowd misses: your all-in electricity cost is never just the rate on your utility bill. It may also include demand charges, power factor penalties, transformer losses, and cooling overhead. A facility running at $0.04/kWh on paper might be paying an effective rate of $0.055/kWh once you factor in everything between the meter and the chip.

Demand charges alone can be brutal. Utilities often bill commercial customers not just for total consumption (kWh) but for peak demand (kW). If you’re running a facility at steady state and one section trips offline then comes back online simultaneously, that demand spike can cost you hundreds or thousands of dollars in a single billing cycle. Managing demand isn’t optional, it’s an active operational discipline that directly affects your variable cost per bitcoin mined.

Then there’s the increasingly relevant dimension of grid demand response programs. Some operators deliberately curtail mining during peak grid demand in exchange for credits or reduced rates. This sounds like free money until you calculate the hash rate you’re sacrificing during curtailment windows. If your utility offers a $0.02/kWh discount for agreeing to shed load 200 hours per year, you need to model that against 200 hours of lost mining revenue at current difficulty and price. Sometimes it’s a clear win. Sometimes it’s a wash dressed up as a savings program.

🟠 Hardware Depreciation

Here’s where some mining profitability calculators get it wrong. They treat hardware as a fixed, one-time capital expense and then calculate “daily profit” as if the machines will run forever. Unfortunatley, they won’t.

S19j Pro

ASIC miners are depreciating assets with finite productive lifespans. An S19j Pro that was once a leader in 2022 is now a marginal machine for many post-halving. The question isn’t whether your hardware loses value—it’s how fast, and whether you’re recovering that cost before efficiency improvements make your fleet uncompetitive.

Hardware amortization is functionally a variable cost because it scales directly with your mining capacity and its useful life depends on network conditions you don’t control. If you deployed 100 units at $2,500 each and plan to amortize them over 36 months, that’s roughly $6,944 per month in hardware cost—before power, before hosting, before anything else. If difficulty rises faster than expected or a new generation of ASICs drops efficiency by 20%, your amortization timeline compresses and your effective cost per terahash climbs.

The operators who survive multiple halvings aren’t the ones with the cheapest power. They’re the ones who correctly model hardware replacement cycles and time their fleet upgrades to network difficulty trends. Buying machines at cycle lows and deploying them before difficulty catches up is one of the few genuine edges left in mining economics.

There’s also a secondary market dimension that most new miners overlook entirely. Your old machines have resale value—but that value is a moving target tied to network difficulty, bitcoin price, and the release schedule of next generation hardware. Selling an S19j Pro six months before a new generation ships nets a very different recovery than selling it six months after. The timing of fleet rotation isn’t just about when you buy new machines. It’s about when you sell the old ones, and whether you’re strategic or reactive about that decision.

🟠 Cooling: The Variable Cost Hiding in Your Infrastructure

Cooling is one of those costs that looks fixed until you actually examine it. The fans inside your ASICs, the ventilation systems in your facility, immersion fluid circulation pumps—all of these consume power proportional to your heat output, which scales directly with your hash rate.

Cipher Mining Inc. data center in Odessa, Texas

In air-cooled facilities, cooling can add 15–30% to your base power consumption depending on ambient temperature and facility design. A shipping container operation in Quebec running through January has fundamentally different cooling economics than an identical setup in West Texas during August. The same machines, the same hash rate, wildly different variable costs.

The S23 series makes this tradeoff especially concrete. The air-cooled S23 runs at 11 J/TH, while the S23 Hydro pushes efficiency to 9.5 J/TH—the first ASIC to break below 10 J/TH—at the cost of significantly higher infrastructure investment. You trade ventilation and fan power for fluid circulation and heat exchange systems. The upfront infrastructure cost is higher, but the variable cost per terahash can drop significantly, especially in hot climates where air cooling becomes a losing battle above 35°C ambient.

What’s often overlooked is that cooling failures don’t just increase costs, they destroy revenue. An overheating ASIC throttles itself, reducing hash rate and therefore revenue while still consuming near-full power. Worse, sustained thermal stress accelerates hashboard degradation, compressing your hardware amortization timeline. Cooling isn’t overhead. It’s revenue protection.

🟠 Firmware and Tuning: Margins at the Software Layer

This is where the gap between casual miners and serious operators becomes obvious. The difference between running stock firmware and running optimized third-party firmware like Braiins or Vnish can shift your efficiency by 10–15% depending on the machine and your tuning targets.

firmware screen shot

Autotuning firmware dynamically adjusts chip frequency and voltage to optimize for either maximum hash rate or maximum efficiency. On the S23, the stock efficiency baseline is already an industry leading 11 J/TH, which compresses the marginal gains available from third-party firmware compared to older generations. Firmware optimization still matters for underclocking to chase efficiency in low-price environments or overclocking to maximize hash rate when margins are strong, but the delta is narrower on modern hardware than it was on S19-era machines.​

Firmware isn’t free in the variable cost sense. Most third-party firmware providers take their cut through devfee—a percentage of your hash rate (typically 2–2.5%) that mines to their pool for a portion of each hour. That devfee is a direct variable cost proportional to your total hash rate. Whether it’s worth paying depends on whether the efficiency gains outweigh the hash rate you’re surrendering.​

The math usually favors optimized firmware on older machines fighting to stay above the profitability line. Firmware optimization can be the difference between running and unplugging. On a current generation machine like the S23 that already ships with strong efficiency, you need to run the numbers honestly before paying a devfee for diminishing marginal gains.

🟠 Pool Fees and Payout Structures

Mining pool fees are the most straightforward variable cost—typically 0–4% for competitive pools, though FPPS pools commonly charge 2.5–4% to cover the variance risk they absorb—but the payout structure matters more than most operators realize.

miningpoolstats.stream

FPPS (Full Pay Per Share) pools pay you for every valid share regardless of whether the pool finds a block. You get predictable, steady payouts, but the pool charges a higher fee to compensate for the variance risk they’re absorbing. PPS+ and PPLNS structures have different fee-risk profiles that can meaningfully affect your realized revenue over time.

For small to mid-scale operators, the consistency of FPPS is usually worth the higher fee. Variance kills small operations. One unlucky week with a PPLNS pool can wreck your monthly cash flow, while the same week on FPPS is invisible. The “cost” of the higher fee is really insurance against variance—and insurance is a legitimate operating expense.

​The less obvious pool-related cost is transaction fee revenue. Pools that don’t share transaction fees with miners (or share them opaquely) are effectively taking an additional cut beyond their stated fee percentage. Transaction fee revenue is highly volatile: it dropped below 1% of block rewards during quiet periods in 2025, spiked to 6.7% post-halving in May 2024, and has sat around ~15% of miner revenue during active on-chain periods in 2026. Leaving that on the table because you didn’t read the fine print on your pool’s payout structure is an expensive oversight.

Pool selection also interacts with your geographic and regulatory risk. Operators concentrated in a single pool are exposed to that pool’s uptime, fee changes, and policy decisions. Diversifying across two or three pools adds operational complexity but reduces single-point-of-failure risk. The “cost” here isn’t a line item—it’s the expected value of avoiding a catastrophic payout disruption if your sole pool goes down for 48 hours or changes its fee structure without meaningful notice.

🟠 Maintenance and Downtime: The Costs That Don’t Show Up on Calculators

Every hour a machine is offline is revenue permanently lost. Unlike most businesses, mining has zero ability to “make up” lost production—the blocks you missed are gone. This makes maintenance and uptime management a genuine variable cost, even though it’s almost never modeled as one.

American Bitcoin Corp mine

Hashboard failures, PSU replacements, fan swaps, thermal paste reapplication, firmware crashes—these are not edge cases. They’re the normal operational rhythm of running ASIC hardware 24/7 in thermally aggressive environments. The question is whether you’re budgeting for them or pretending they won’t happen.

A reasonable maintenance budget for a fleet of current generation ASICs is 3–5% of hardware cost annually, plus the opportunity cost of downtime during repairs. Operators who stock spare PSUs and hashboards on-site can turn around repairs in hours. Those who don’t are waiting days or weeks for parts, bleeding revenue the entire time.

This is also where labor becomes a variable cost for large operations. A solo miner maintaining 10 units can absorb repair time into their day. An operation running 500+ units needs dedicated technicians, and their labor cost scales directly with fleet size and failure rates.

Environmental factors compound maintenance costs in ways that don’t show up until you’ve been operating for a full year. Dust accumulation in air-cooled facilities, humidity-driven corrosion on hashboards, rodent damage to wiring in rural deployments. These aren’t hypotheticals, they’re the Tuesday afternoon realities of running industrial hardware in non-cleanroom environments. Operators who budget only for “normal” component failure may learn expensive lessons about the difference between a data center and a converted warehouse.

🟠 Putting It All Together

When you stack these costs honestly, the picture of mining economics becomes much more complex than “revenue minus electricity equals profit.” A realistic variable cost breakdown for a mid-scale operation looks something like:

 Electricity (including demand charges and cooling overhead): 60–69% of variable costs
 Hardware amortization: 15–19%
 Pool fees: ~0–4%
 Firmware devfees: ~2–2.8%
 Maintenance and downtime: ~3–5%

The operators who survive aren’t the ones who just found the cheapest electricity. They’re the ones who understand every layer of their cost structure, optimize across all of them simultaneously, and model their business against multiple difficulty and price scenarios. Cheap power with poor fleet management loses to moderate power with disciplined operations every single cycle.

The next time someone tells you Bitcoin mining profitability is “just about the electricity,” ask them how they’re modeling their hardware replacement cycle. Ask them what their effective rate looks like after demand charges. Ask them whether their firmware devfee is worth the efficiency gains on their specific generation of machines.

The silence that follows will tell you everything about how many halvings their operation will survive. Mining isn’t a spreadsheet exercise. It’s an operational discipline, and the operators who treat their full variable cost stack with the same rigor they apply to finding cheap power are the ones still running when the next difficulty epoch closes.

✅ Subscribe for Weekly Bitcoin Mining Insights

I publish weekly operator breakdowns on mining economics, fee markets, and Bitcoin fundamentals. No hype, no price calls—just hands-on analysis from an operator mining since 2019. Subscribe now to receive early access straight to your inbox!

✅ Socials

Follow along on my socials:

🔸 🐦 X: @OrngeHorizonBTC
🔸 🦋 Bluesky: @orngehorizonbtc.bsky.social