Curtailment and demand response: getting paid to switch off

Electricity grids have to match supply and demand continuously. Most large consumers cannot help with that: a smelter, a data centre or a hospital cannot simply stop. A mining site can, instantly, with no product spoiled and no process to restart. That flexibility is a service with a market price, and operators who treat it as a product rather than an inconvenience change their own economics.

The three things people mean by curtailment

• Contractual curtailment. Your power contract is cheaper on the condition that the supplier can require you to stop for a stated number of hours a year. You are paid in the tariff rather than in cash.
• Demand response programmes. You commit to reduce load when the grid operator calls, and you are paid for being available whether or not they call, plus more when they do.
• Economic curtailment. Nobody asked you to stop. The spot price of power rose above what mining earns, so you switch off and either save the cost or sell the power back. This is a decision you make hourly, not a contract.

They stack. A site can hold a curtailable tariff, be enrolled in a programme, and still switch off economically at other times, provided the contracts do not conflict.

The arithmetic that makes it work

Mining is unusual because your opportunity cost of stopping is precisely knowable. It is the machine's contribution: gross revenue at current hashprice minus the power it would have burned. If the payment for curtailing exceeds that contribution, stopping is the profitable choice, and the calculation takes seconds.

On sites with volatile wholesale prices, the hours where power costs more than mining earns are not rare. A fleet that runs through them because nobody built a shutdown process is destroying value in a completely measurable way.

What the grid asks of you

• Speed. Programmes specify how quickly you must be off. Miners are excellent at this, but only if the shutdown is automated. A person with a login is not a response time.
• Depth. How much load you can shed, verified against a baseline. Understating it wastes money, overstating it means penalties.
• Verification. Interval metering the operator can audit. Your own dashboard is not evidence.
• Availability. Most programmes pay largely for being available. Missing a call because a switch failed costs more than the event was worth.

Building for it

The mechanism matters. Switching machines off in firmware is slow and hard on hardware if repeated. Dropping breakers is fast but hard on power supplies. Most sites end up with a tiered approach: an automated curtailment profile that underclocks the fleet for shallow, frequent events, and hard shutdown reserved for deep ones. That is one of the practical uses of the tuning covered in underclocking and overclocking.

Test the drop before you need it, and test the restart, which is the half everyone forgets. A fleet that switches off in thirty seconds and takes four hours to come back has not solved the problem, it has moved it.

What it costs you elsewhere

Curtailment reduces uptime, and uptime is an input to every payback model. A site curtailed ten percent of the year earns ninety percent of the mining revenue, so the curtailment payments have to more than cover the difference. Put the real expected uptime into the model in payback on a mining rig rather than the hundred percent the calculators assume.

It also constrains other deals. If you have promised heat to a neighbour, as in selling the heat, your ability to curtail is no longer entirely yours. Read those two contracts side by side before signing either.