Bitcoin mining is rightly praised as a flexible electricity load. But flexibility is a state, not a label: how much load a fleet will actually shed depends on hashprice, power price, machine efficiency, contracts and restart constraints.
Flexible load is not the same as fixed flexibility
Grid operators and mining advocates repeat the same sentence: bitcoin mining is an interruptible, flexible load. It is true, and it is the reason mining keeps winning seats at energy-market tables.
The mistake is the silent upgrade of that sentence into something stronger: that every megawatt of mining is available flexibility, at any moment, at a predictable price. It is not. A miner curtails when shutting down is worth more than mining. That comparison has two moving sides, and both change daily.
On the revenue side sits hashprice: expected mining revenue per unit of hashrate per day. On the cost side sits your all-in power price for the next hour, including energy, transmission charges and program obligations. Between the two sits your fleet: its efficiency in joules per terahash, its restart behavior, and the contracts wrapped around it.
Flexibility lives in that comparison. When the spread is wide, mining absorbs cheap power and stays online through moderate price spikes. When the spread narrows, the same fleet becomes eager to curtail. Same machines, same site, different state.
The curtailment threshold, in plain English
Every operating fleet has a break-even line: the power price above which running the machines earns less than the electricity costs, adjusted for whatever the grid will pay you to stop. Cross the line and continuing to mine destroys money; below it, curtailment destroys money.
That line is not a constant you calculate once. It moves with:
- hashprice, which repriced your revenue while you slept;
- your all-in power cost, which in markets like ERCOT can swing violently within a day;
- the efficiency of the specific machines at the site: an efficient fleet still earns thin margin at power prices that make an old fleet bleed;
- what a curtailment event actually pays or saves under your PPA, hosting agreement or demand-response program;
- how long the site takes to come back, and what ramping does to hardware and program compliance.
None of this requires an invented universal formula. It requires knowing your own numbers on each line, per site, today.
What the 2026 Texas manuscript actually found
A 2026 preprint by Subir Majumder studied bitcoin mining load in the Texas market, where large flexible loads face wholesale prices plus transmission-charge incentives. It is a manuscript, not peer-reviewed consensus, and its scope matters.
The result matters even more: mining load declines as power costs rise, but the response is moderated by hashprice. When expected mining revenue is higher, miners tolerate higher power prices before shutting down. In other words, the implied curtailment threshold shifts upward.
The author’s caution is direct. Treating mining loads as stable demand-response resources may overstate their available flexibility, because the flexibility on offer varies with crypto-market conditions.
For grid planners, that is a warning about counting on mining megawatts as if they were a battery. For operators, it is close to a mirror: your own willingness to curtail weakens exactly when hashprice improves. Any energy strategy that ignores this will misprice both uptime and flexibility.
What Riot’s Q1 2026 numbers actually say
Riot Platforms’ Q1 2026 Form 10-Q is one of the clearest public windows into what energy strategy is worth to a large miner. Riot reported $21.0 million of power curtailment credits for the quarter, compared with $7.8 million a year earlier.
Read the filing’s definitions before repeating the number, because three different things hide under the word curtailment:
| Mechanism | Who triggers it | What it means in the filing |
|---|---|---|
| Manual curtailment | The operator | Riot powers down when potential power credits exceed the bitcoin-mining revenue it would otherwise generate. |
| Grid-directed curtailment | ERCOT / Oncor | Load reductions follow instructions or testing by the grid operator. |
| Program participation | Contracted | Credits include sales of unused PPA power and participation in ERCOT and MISO demand-response programs. |
Three honest observations follow. First, the credits are real and material: they are recognized outside cost of revenue but, in Riot’s words, significantly reduce the overall cost to mine bitcoin. Second, they are not free money for doing nothing. They are the paid side of a deliberate trade against foregone mining revenue, executed under specific contracts. Third, they are volatile by nature. The filing says the amount varies from period to period with power supply and demand factors. A quarter of $21.0 million follows a prior-year quarter of $7.8 million; neither number is a promise about the next one.
The operator decision stack
Before any shutdown decision, five layers of your own data have to agree:
- Establish the revenue side, datedRecord hashprice at a defined time with its source. Every curtailment decision inherits this timestamp; a threshold computed on stale hashprice is a guess.
- Establish the true cost sideAssemble the all-in power price for the decision window: wholesale or contract energy, transmission charges, program penalties or payments. Site-level, not company-average.
- Segment the fleet by measured efficiencyGroup machines by wall-measured J/TH, not by nameplate. In a mixed fleet, evaluate the least-efficient machine groups first rather than assuming one site-wide threshold.
- Price the transition, not just the stateEstimate the cost of going down and coming back: lost ramp time, restart labor, program verification and thermal stress. A one-hour price spike can be shorter than your round trip.
- Write the trigger rule and log every eventDefine who decides, at which spread, for which fleet segment. Record every curtailment with its numbers. The log is how thresholds improve and how you prove program performance.
What firmware and control layers can - and cannot - do
Automation is entering this space visibly. On July 14, 2026, Enel North America and Braiins announced a service integrating mining software with a demand-response platform in the PJM and ERCOT markets, so participating facilities can reduce load automatically during grid stress and be compensated for it. That is an industry signal worth watching: curtailment logic is moving from a human with a spreadsheet into the control stack.
Be precise about what any control layer, from any vendor, can honestly offer. Software can expose per-machine power and efficiency data, execute staged shutdowns and restarts, and act faster and more consistently than manual operations. It cannot repeal the economics above: no firmware makes curtailment profitable when the spread says otherwise, and no automation removes the volatility of program revenue.
Any operation beyond manufacturer defaults, including firmware-level tuning, automated power management or rapid cycling, carries real risk. It can affect manufacturer warranty, machine stability and hardware lifespan, while obligations differ by hosting contract, program and jurisdiction. Validate on a small segment first, keep restart procedures tested, and read your contracts before automating anything. Nothing here is a promise of savings or revenue.
Three questions operators keep asking
Is curtailment always worth it when power prices spike?
No. It is worth it when the spike exceeds your site’s current threshold: expected mining revenue at today’s hashprice for the affected machines, adjusted for what the event pays and what the round trip costs. At high hashprice, a fleet can rationally mine through prices that would have justified shutdown a month earlier.
Did Riot really earn $21 million by turning miners off?
Riot reported $21.0 million of power curtailment credits for Q1 2026, versus $7.8 million a year earlier. Per its 10-Q, the credits include sales of unused power under PPAs and participation in ERCOT and MISO demand-response programs: a managed energy strategy with foregone mining revenue on the other side of the trade, not passive income.
Can grid operators count on mining as demand response?
Partly. The 2026 Texas manuscript finds that mining responds to prices, but the response weakens when hashprice rises. Treating mining load as a fixed demand-response resource may therefore overstate available flexibility. Measured, contract-backed flexibility can be priced; a slogan cannot.
Flexibility is a state, not a label
The industry line “mining is a flexible load” earned its place. The operators making money on it in 2026 use a longer version: mining is a flexible load whose flexibility has a price - and we know ours, today, per site and per machine group.
Review six source-linked performance reports before converting a reported TH/s, watt or J/TH figure into an economic assumption. Open the independent field reports.
Sources and data freshness
- Subir Majumder, “Hashprice moderates the electricity demand response of Bitcoin miners,” arXiv:2606.00587, manuscript v2, revised June 3, 2026. Preprint, not peer-reviewed. arXiv (checked August 19, 2026).
- Riot Platforms, Form 10-Q for the quarter ended March 31, 2026. U.S. SEC filing (checked August 19, 2026).
- Enel North America and Braiins, automated demand response for bitcoin mining in PJM and ERCOT, July 14, 2026. Press release (checked August 19, 2026). Industry signal; not a statement about VNISH products.
- Hashrate Index, hashprice snapshot of $34.15/PH/day on August 19, 2026. Hashrate Index. This metric changes continuously; refresh before republication.
Published by the ROI ASIC Analytics Desk. ROI ASIC is part of the VNISH ecosystem; VNISH develops ASIC firmware and fleet-management tooling. This article is operator analysis built on the cited public sources; it contains no product claims and no revenue promises. Material claims were checked against the cited public sources before publication.