A “51% attack cost” estimate can be wrong before it reaches a hardware price. If an attacker adds capacity A while honest miners keep capacity H, the relevant share is A / (H + A). Adding 51% of H gives only 33.77% of the enlarged total. Reaching 51% requires about 1.0408 times H in additional capacity.
We publish 42 declared hardware-capacity scenarios to make that distinction inspectable. They are not a quote for an attack, proof that machines are available, or a prediction of reorganization success. The old edition’s precise rentable-capacity and feasibility claims lacked preserved evidence and have been removed.
Two different capacity assumptions
| Situation | Equation | What is being assumed |
|---|---|---|
| Capacity redirected from an existing total | Share = controlled work / existing work | The total is not enlarged by that redirection |
| Entirely new capacity joins unchanged honest miners | q = A / (H + A); A = qH / (1 − q) | Honest capacity remains online and unchanged |
| Example: A = 0.51H | q = 0.51 / 1.51 = 33.77% | Not a majority of the enlarged total |
| Example: target q = 0.51 | A = 1.0408163265H | Additional capacity exceeds H |
These are alternative models, not two costs to add together. Real fleets, pools, rentals and miner responses can mix assumptions. A reported pool share also does not prove that a single organisation owns or can redirect all the underlying machines.
Machine equivalents under one declared baseline
The table fixes honest capacity at an illustrative 3 PH/s and target share at 51%. It uses manufacturer specifications and displayed prices frozen on 30 September 2026. It rounds machine counts upward. Some listings were sold out or offered future delivery. A unit price cannot be multiplied into evidence that a manufacturer can deliver an entire fleet at that price.
| Configuration | Machine equivalents | Gross listing multiplication, $m | Machine power, MW | Energy $/h at $0.10/kWh |
|---|---|---|---|---|
| Antminer L7 9050M | 345,022 | 249.80 | 1,124.08 | 112,408 |
| Antminer L9 16G | 195,154 | 655.72 | 655.72 | 65,572 |
| Antminer L11 20G | 156,123 | 1,464.43 | 574.53 | 57,453 |
| Antminer L11 HU2 35G | 89,213 | 1,464.43 | 515.21 | 51,521 |
| ElphaPex DG2+ | 152,315 | 1,530.00 | 594.03 | 59,403 |
| Goldshell MINI DOGE III PLUS · default | 3,854,876 | 1,827.21 | 1,927.44 | 192,744 |
| Goldshell MINI DOGE III PLUS · low power | 5,036,209 | 2,387.16 | 1,510.86 | 151,086 |
The complete CSV varies the honest baseline across 1.5, 3 and 4.5 PH/s, and the target across 51% and 60%, for seven configurations. MINI DOGE’s two modes are alternatives for the same hardware, not two separate products to buy. Manufacturer tolerances and actual wall measurements are outside this model.
Why the capital column is not an attack budget
A delivered installation adds freight, duties, electrical infrastructure, cooling, site work, failures, maintenance and financing. Acquisition lead time and available inventory can dominate the result. Running costs also have to account for legitimate mining revenue and opportunity costs; electricity expenditure alone is not the net cost of malicious activity.
Conversely, an actor already controlling machines does not need to buy the same fleet again. A hardware acquisition table answers a new-capacity hypothetical, not the opportunity cost for every possible adversary. We provide no verified rental-market census and no statement that an attack is commercially feasible or impossible.
What a majority of work does not override
Full nodes still validate signatures and permitted issuance. A majority of proof-of-work does not give somebody the private key to your wallet. Reorganizations can nevertheless undermine recent settlement assumptions. The exact outcome depends on the threat model and recipient acceptance policy; there is no universal safe confirmation count.
The developer account of the March and April 2026 MWEB incidents describes a validation exploit and a later disruption with an invalid-chain reorganization. It should not be relabelled as evidence of a conventional majority-hashrate attack. Absence of a documented attack in an article is also not a mathematical security guarantee.
Practical use of the model
Use the table to interrogate a numerical claim: which denominator, whose hardware, what date and what costs? For actual mining economics, use Is Litecoin mining profitable now? The costs of seven ASIC setups and the mining calculator. For payment decisions, the transaction lab explains what changes when a block is reorganized.
Reproduce the arithmetic
Declared inputs and standard-library Python code are public. Save both with input-sha256.json into one folder and run python analyze.py.txt .. Compare with results.json.
These files reproduce declared arithmetic. They do not verify hardware availability, execute purchases or audit a company’s holdings.
Sources and review scope
Reviewed against the linked materials on 2 October 2026. Calculations and drafting were AI-assisted; no independent external review or personal attendance is claimed. Original publication date retained. See our corrections policy.
