AI-generated research, checked but not peer-reviewed. It may contain errors.

E838　 KILL　 UnderpoweredTested · October 2026

Rented Bitcoin hashrate is priced at about what it earns

Renting Bitcoin hashrate on days when mining revenue per hash exceeded the rental price by more than 8% gave only 6 signal days after 2022 against 30 required, while the all-days average lost 6.85% of rental cost.

hashrate-rental bitcoin-mining fair-value compute-markets daily

Readable summary Full research record JSON recordFeedback on E838

Mining revenue per unit of hashrate can be computed from the difficulty and the block rewards. A common belief says the rental price lags that figure, so renting hashrate on days when it is cheap and mining with it should pay. We compared the daily price on the largest hashrate rental marketplace with the computed revenue per hash from 2020 to 2026 across 2,475 days. Revenue per hash averaged 0.994 of the rental price in both periods (before fees), and renting on an average day lost 6.85% of rental cost. Only 19 days passed the 8% signal, 6 of them after 2022 against 30 required.

Revenue over price0.994mean ratio of mining revenue per hash to the rental price, before fees; discovery and holdout alike

All days, net-6.85%of rental cost per day, 2,475 days, fees and stale allowance included

Holdout signal days6against the 30 required, too few to say anything about them

Discovery signal days13mean -1.08% of rental cost

01 · The idea being tested

Hashrate is rented at a price that lags mining revenue per hash, so renting when it is cheap and mining with it earns a profit.

Why people believe it

Mining revenue per hash follows from public numbers: the network difficulty and the block reward, which includes fees. People argue that rental prices adjust more slowly than difficulty and fees, and that pools paying on a trailing average of fees (full pay per share, FPPS) hand the renter the gap.

Why it would persist The argument, as stated before testing

Common belief, not tested here: renting needs an account, a pool and a seller with enough depth, so the gap may be left open by participants who do not rent.

02 · How it was tested

Every day we computed what one unit of hashrate would earn from mining at the day's difficulty, using the previous day's average block reward because such a pool pays on the trailing fee average. We compared it with the price of renting that hashrate. Whenever the computed revenue was more than 8% above the previous day's rental price, we assumed we rented for the day at the day's actual average price and mined. The return is revenue after the pool fee and a 1% allowance for work that arrives too late to count, minus the rental price and the marketplace buyer fee, as a share of the rental cost. The rule was fixed before the results were looked at, and tested on a holdout starting in 2022.

Pass required

At least 30 holdout signal days, holdout mean net at least +2% of rental cost, week-block bootstrap interval with lower bound above 0, median above 0, and both holdout halves above 0.

Killed if

Holdout mean at or below 0, fewer than 30 holdout signal days, or a positive result only before 2022.

Costs

Net e = V x (1 - 2.5% pool fee - 1% stale allowance) - P x (1 + 3% marketplace buyer fee); return = e / (P x 1.03). Stress: pool fee 4%.

Controls

The unconditional mean over all days; the subset of signal days after downward difficulty adjustments.

03 · What happened

Why it was rejected

The pre-registered minimum of 30 holdout signal days was not met: only 6 days had mining revenue per hash more than 8% above the previous rental price.

Across 2,475 days the mean ratio of mining revenue per hash to the rental price was 0.994 in both the discovery period and the holdout, so the market priced rented hashrate at what it earns.
Renting on every day, without a signal, returned -6.85% of rental cost per day net. That matches the fee and stale-value assumptions: at a ratio of 1.0 the model costs 3% buyer fee plus 3.5% pool and stale allowance, about -6.3% of cost.
Only 19 days had a signal above 8%: 13 in the discovery period and 6 in the holdout, against the 30 required.
Discovery signal days returned -1.08% of rental cost on average (n 13).
The 6 holdout signal days fall in two years, five in 2024 and one in 2023. They are too few to say anything about their return.
Signal days within a day of a downward difficulty adjustment (n 6) returned -1.3% on average, so no gain after a difficulty cut was visible.
A 4% pool fee instead of 2.5% gave the same picture.

Net return on rental cost, by group of days

Mean net return per rented day. Renting on any day loses 6.85%. The 6 holdout signal days are too few to show.

Chart data (% of rental cost, mean)

	Mean net return
All days (n 2,475)	-6.85
Signal days, discovery (n 13)	-1.08
Signal days within a day of a downward adjustment (n 6)	-1.3

Mining revenue per hash against the rental price

Mean ratio of computed mining revenue per hash to the rental price, before fees. A ratio of 1 means renting costs exactly what mining earns before fees. Both periods sit at 0.994.

Chart data (ratio of revenue to price, mean)

	Revenue / rental price
Discovery	0.99
Holdout	0.99

Days left after each gate

Of 2,475 days with at least 0.1 EH/s rented, 19 passed the 8% signal and 6 of those fall in the holdout, against 30 required.

Chart data (days)

Stage	Count
Days with at least 0.1 EH/s rented	2,475
Days with a signal above 8%, both periods	19
Signal days in the holdout (30 required)	6

04 · What it means

Why it failed Interpretation · not established

The rental price follows mining revenue per hash almost one for one, and the same-day repricing leaves the 8% gap open on about six days in nearly five years of holdout. The all-days loss is what the fees and the stale allowance predict, so the market gave no discount to pay for them. The same-day repricing may explain why no lag after difficulty changes was visible.

Lesson

When a price can be compared with a number computed from public data, the gap tends to be about the size of the frictions. A pre-registered minimum count turns that into a clear result instead of a story about a handful of large days.

Scope of this result

Covers daily average prices of rented SHA-256 hashrate on one marketplace from 2020 to 2026, revenue modelled from difficulty and block rewards with an FPPS lag, an 8% signal and the stated fee assumptions. The daily price averages all orders, so it may sit above the marginal price; depth for renting 1 EH/s was not measured. Account, pool and payout mechanics were not tested.

A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.

1. Test design

Pre-registered

Yes: rule and criteria fixed before the test data was examined

Data

Daily average price, speed and spend of rented SHA-256 hashrate — public API of the largest hashrate rental marketplace; 1 day; 2019-07..2026-10; public, anonymous | Daily average block reward including subsidy, and every difficulty adjustment with its time — public Bitcoin block explorer API; 1 day; per difficulty adjustment (468 adjustments); 2019-07..2026-10; public, anonymous

Universe

Days with at least 0.1 EH/s of rented speed: 2,475 days from 2020-01 to 2026-10, in one SHA-256 hashrate market.

Signal

Revenue per EH/s per day at difficulty D and reward R (sat per block): H = 86400 x R x 1e10 / (D x 2^32). The FPPS lag is modelled by using the previous UTC day's average reward. The signal is S = H(difficulty at 00:00 UTC of day d) / P(d-1) - 1, with P the rental price in BTC per EH/s per day. Signal when S is above 8%.

Execution

Rent for the whole day d at that day's realised average price P. The realised value V is the time-weighted mean of H over the day, using the actual difficulty adjustment times.

Costs

Net e = V x (1 - 2.5% pool fee - 1% stale allowance) - P x (1 + 3% marketplace buyer fee); return = e / (P x 1.03). Stress: pool fee 4%.

Controls

The unconditional mean over all days; the subset of signal days after downward difficulty adjustments.

Sample split

Discovery 2019-07 to 2021-12; holdout 2022-01 to 2026-10, split into two halves at the median holdout signal date.

Pass criterion

At least 30 holdout signal days, holdout mean net at least +2% of rental cost, week-block bootstrap interval with lower bound above 0, median above 0, and both holdout halves above 0.

Kill criterion

Holdout mean at or below 0, fewer than 30 holdout signal days, or a positive result only before 2022.

Power

6 holdout signal days against the 30 required; the 6 days fall in two years (1 in 2023, 5 in 2024), so no interval or half-split could carry weight.

2. Result

Observations

Instruments

Net mean

—

95% interval

Not reported

Median

—

Worst

—

Finding 1

Across 2,475 days the mean ratio of mining revenue per hash to the rental price was 0.994 in both the discovery period and the holdout, so the market priced rented hashrate at what it earns.

Finding 2

Renting on every day, without a signal, returned -6.85% of rental cost per day net. That matches the fee and stale-value assumptions: at a ratio of 1.0 the model costs 3% buyer fee plus 3.5% pool and stale allowance, about -6.3% of cost.

Finding 3

Only 19 days had a signal above 8%: 13 in the discovery period and 6 in the holdout, against the 30 required.

Finding 4

Discovery signal days returned -1.08% of rental cost on average (n 13).

Finding 5

The 6 holdout signal days fall in two years, five in 2024 and one in 2023. They are too few to say anything about their return.

Finding 6

Signal days within a day of a downward difficulty adjustment (n 6) returned -1.3% on average, so no gain after a difficulty cut was visible.

Finding 7

A 4% pool fee instead of 2.5% gave the same picture.

3. Interpretation

Why it failed

Lesson

Scope

4. Reproducibility

Steps

1) Download daily SHA-256 rental price (sat per H/s per day), speed and spend from the public API of the largest hashrate rental marketplace, and the daily average block reward (sat per block, including subsidy) and all difficulty adjustments with times from a public Bitcoin block explorer API. 2) Convert the rental price to BTC per EH/s per day (times 1e10); hashprice H = 86400 x R x 1e10 / (D x 2^32). 3) For each day with at least 0.1 EH/s, take R from the previous UTC day, compute V as the mean of H over the day at the actual difficulty (10-minute grid), and the signal S = H at 00:00 / P(previous day) - 1. 4) Where S is above 8%, net = V x (1 - 2.5% - 1%) - P x 1.03, divided by P x 1.03. 5) Split at the start of 2022, bootstrap by week blocks, and compare with the all-days mean.

Grade

COMPLETE