

Swan’s Agentic Software Factory: Cygnet Evolved
How Swan built an agentic software factory that finds production bugs, reaches consensus on the fix, opens the PR, and proves it works before a human merges.
In this article
It costs an average of about $75,000 to $80,000 to mine one Bitcoin in 2026. Public mining companies reported a weighted average cash cost of about $75,500 per BTC in the second quarter of 2026, according to CoinShares (report published September 15, 2026). On-chain analyst Checkonchain put the network’s average production cost at $78,254 in August 2026.
Your own cost depends mostly on two numbers: what you pay for electricity and how efficient your machine is. With a new machine and cheap industrial power ($0.05 per kWh), electricity alone runs about $25,000 to $43,000 per Bitcoin at October 2026 difficulty. At home electricity prices, with an older machine, the power bill alone can top the price of a Bitcoin, and that’s before paying for the machine.
Bitcoin price: About $82,600 (mempool.space, Oct 19, 2026)
Network hashrate: About 1,028 EH/s (Hashrate Index, Oct 19, 2026). New to the term? See what Bitcoin hashrate is.
Difficulty: About 140.6 trillion, up 5.92% on Oct 16, 2026 (mempool.space, Oct 19, 2026)
Hashprice (miner revenue per PH/s per day): About $36.80 (Hashrate Index, Oct 19, 2026)
Average cash cost, public miners (Q2 2026): About $75,500 per BTC (CoinShares)
Average production cost (Aug 2026): $78,254 per BTC (Checkonchain via Cointelegraph, Aug 12, 2026)
These figures are electricity only. They leave out the machine, hosting, cooling and staff. They assume difficulty of 140.58 trillion, the level set on October 16, 2026. At that difficulty one PH/s earns about 0.000447 BTC a day from the 3.125 BTC block subsidy, so mining 1 BTC takes about 2,236 PH-days of work. Transaction fees add less than 1% to miner revenue (see below), so we leave them out.
9.5 J/TH (Antminer S23 Hydro): $20,400 at $0.04/kWh · $25,500 at $0.05/kWh · $40,800 at $0.08/kWh · $61,200 at $0.12/kWh
13.5 J/TH (Antminer S21 XP, air-cooled): $29,000 at $0.04/kWh · $36,200 at $0.05/kWh · $58,000 at $0.08/kWh · $86,900 at $0.12/kWh
16 J/TH (older S21-class machine): $34,300 at $0.04/kWh · $42,900 at $0.05/kWh · $68,700 at $0.08/kWh · $103,000 at $0.12/kWh
To work it out yourself: a machine rated at X J/TH (joules per terahash) draws X kW for every PH/s of mining power, so it uses X × 24 kWh per PH-day. Multiply that by 2,236 PH-days per Bitcoin and by your price per kWh. For example, a 13.5 J/TH machine uses 324 kWh per PH-day, and 324 × 2,236 is about 724,500 kWh per Bitcoin. At $0.05 per kWh, that’s about $36,200.
The same math in one line: energy per BTC = (difficulty × 2^32 ÷ 10^12 TH per block) × J/TH ÷ 3.125 BTC. At 140.58T that’s about 193.2 billion TH per Bitcoin. At 13.5 J/TH that comes to about 2.61 trillion joules, or about 724,500 kWh.
Difficulty changes every 2,016 blocks, about every two weeks. The next change was estimated for about October 30, 2026, at roughly +0.7% (mempool.space, as of October 19, 2026). That would raise every number above by about 0.7%.
The breakeven power price is the most you can pay for electricity before the power bill alone uses up everything a machine earns. At October 2026 difficulty and a Bitcoin price of about $82,600, it works out to:
9.5 J/TH (Antminer S23 Hydro): about 16.2¢ per kWh
13.5 J/TH (Antminer S21 XP): about 11.4¢ per kWh
16 J/TH (older S21-class machine): about 9.6¢ per kWh
We get there two ways, and they agree. Dividing the Bitcoin price by each machine’s energy use per coin (about 510,000, 724,500 and 859,000 kWh) gives the same result as dividing the Hashrate Index hashprice of about $36.80 per PH/s per day by each machine’s daily energy use (228, 324 and 384 kWh per PH-day).
This is electricity only. The machine still has to pay for itself, along with hosting and cooling, so real profit needs power well under these numbers. At $0.05 per kWh, electricity takes about 31% of an S23 Hydro’s revenue, 44% of an S21 XP’s and 52% of an older 16 J/TH machine’s. The breakeven rises if the Bitcoin price does (see what analysts forecast for 2026 to 2040) and falls each time difficulty climbs.
The all-in cost includes machines (which lose value fast), buildings, staff, financing and hosting fees. CoinShares' Q2 2026 data shows a huge spread between companies, in all-in cost per BTC:
American Bitcoin (ABTC): $76,529
Bitdeer: $89,737
Riot Platforms: $113,499
HIVE: $116,739
CleanSpark: $135,691
IREN: $141,593
MARA: $163,866
Bitcoin ended June 2026 at $58,400, below the $75,500 average cash cost, so listed miners as a group were below cash breakeven in mid-2026. Hashprice hit a record daily low of $27.74 per PH/s per day on June 6, 2026. With Bitcoin near $82,600 in October 2026, the price is above the average cash cost, but still below the Q2 all-in cost of every company in the list above except American Bitcoin. Miners have been squeezed like this before: see the great miner squeeze of 2022.
On your own, a very long time. One Antminer S21 XP (270 TH/s, 3,645 W) earns about 0.00012 BTC, or about 12,000 sats, a day in a mining pool at October 2026 difficulty. That works out to roughly 23 years per Bitcoin. The machine uses about 87.5 kWh a day, which is about $4.40 of electricity at $0.05 per kWh, to earn about $10 of Bitcoin at an $82,600 price.
That’s why almost all mining happens in pools. You earn a small, steady share of the rewards instead of waiting years to win a whole block.
Bitcoin has no central bank or company that issues coins. Instead, miners add a new block of transactions to the blockchain about every 10 minutes, and the network pays the winner in newly created Bitcoin (3.125 BTC per block since the 2024 halving) plus transaction fees.
Winning takes real work: machines run trillions of hashes, and that costs electricity. The cost is what makes the ledger so expensive to attack, the idea at the heart of Satoshi’s white paper. This system is called proof of work, and it solves the Byzantine Generals Problem: how strangers can agree on one history without trusting each other.
The 2024 halving cut the block reward from 6.25 to 3.125 BTC, doubling the cost per coin overnight. The next halving, expected around April 2028, cuts it to 1.5625 BTC. Whether halvings still move the price is debated; see also three myths about the halving.
Fees almost disappeared. Transaction fees were 0.69% of miner revenue in August 2026, near a 10-year low, after falling to 0.52% in April. As the subsidy keeps halving, miners will depend more on fees (Lyn Alden on fee-based security).
The move to AI. Listed miners had announced more than $70 billion in AI and high-performance computing contracts by March 2026. CoinShares said they could earn as much as 70% of their revenue from AI by the end of 2026. Its Q2 2026 report puts disclosed backlogs at more than $100 billion. Core Scientific paid about $41.9 million to cancel a mining-chip contract with Block (July 2026) as it moves into AI.
More efficient machines. New machines use about 9.5 J/TH, compared with 13.5 to 16 J/TH for the previous generation.
For most people, buying is cheaper and simpler.
Mining can make sense if you have power under about $0.05 per kWh, a place to handle the noise and heat, or a way to use the waste heat (for example, heating a building). Lyn Alden explains why Bitcoin’s energy use is not a problem.
Buying makes sense for almost everyone else. At home electricity prices ($0.12 per kWh or more), electricity alone runs about $61,000 to $103,000 per Bitcoin at October 2026 difficulty. Add the cost of the machine, which loses value fast and has to make back its price within a few years, and buying is usually cheaper.
If you want to own Bitcoin, the simplest route is to buy a little at a time (here’s how that compares with investing a lump sum) and move it into a wallet you control.
Hold your IRA with the most trusted name in Bitcoin.
About $75,000 to $80,000 on average across the industry. Electricity alone for an efficient miner on cheap power ($0.05 per kWh) is about $25,000 to $43,000 at October 2026 difficulty.
About 9.6¢ to 16.2¢ per kWh for electricity alone, depending on the machine (16 to 9.5 J/TH), at October 2026 difficulty and a Bitcoin price of about $82,600.
With one top air-cooled machine (Antminer S21 XP) in a pool, about 23 years at October 2026 difficulty.
Only with cheap power and efficient machines. In Q2 2026, listed miners as a group were below cash breakeven, and all but one of the large public miners in CoinShares' table (American Bitcoin) had a Q2 all-in cost above the October 2026 Bitcoin price.
About 510,000 to 860,000 kWh at October 2026 difficulty, depending on the machine (9.5 to 16 J/TH).
The 2024 halving cut the reward to 3.125 BTC per block, and difficulty keeps rising (up 5.92% on October 16, 2026 alone). Fees, which could make up the difference, were under 1% of miner revenue in August 2026.
CoinShares Bitcoin Mining Report Q2 2026 (published Sep 15, 2026; accessed Oct 19, 2026)
CoinShares Q2 report coverage and June-end price (Let’s Data Science) (Sep 16, 2026)
Checkonchain production cost and fee share (Cointelegraph) (Aug 12, 2026)
mempool.space: current difficulty and hashrate (as of Oct 19, 2026)
mempool.space: next difficulty adjustment (as of Oct 19, 2026)
Hashrate Index: hashprice and hashrate (accessed Oct 19, 2026)
Luxor June 2026 lookback: record-low hashprice (Jul 10, 2026)
Core Scientific exits Block mining deal (Bitcoin.com News) (Jul 30, 2026)
Mickey Koss became a freelance writer in the Bitcoin space in an attempt to build a proof of work portfolio for when he left the Army. He graduated from West Point with a degree in Economics before serving in the Army for nearly a decade. He became orange pilled in graduate school and is now a regular contributor to Forbes, Bitcoin Magazine, and Bitcoin News. He’s been on popular podcasts such as BTC Sessions’ Why Are We Bullish, and is a regular on Café Bitcoin.
In this article
Thoughts on Bitcoin from the Swan team and friends.
How Swan built an agentic software factory that finds production bugs, reaches consensus on the fix, opens the PR, and proves it works before a human merges.
Total federal debt crossed $40 trillion last week. Jeremy Showalter maps how we got here, where four institutional projections take it by 2050, and why the honest total is already about $55 trillion.
From entropy and SHA-256 to proof of work and the 21 million cap: a first principles guide to how Bitcoin works, and what you can verify yourself.





