A Bitcoin mine can earn money by switching off. That was the case David Chernis, Senior Account Executive at CPower, made at Mining Disrupt Miami, held June 24–26, 2024. His subject was how automation, demand response and a clear view of a mine's marginal costs let an operator pick the moment when power is worth more to the grid than to the machines.
setting a strike price so that you go into curtailment mode when it's more profitable to give your energy back to the grid than it is to mine
David Chernis · 0:19
Chernis introduced CPower as the largest demand response provider in the United States, with 6.7 gigawatts across 27,000 sites. He pointed to the deregulated markets where it operates, among them ERCOT, PJM and the New York ISO, and said the company expects AI loads to start showing up in other markets as well.
Miners as energy companies
To frame the shift, Chernis quoted Dan Lawrence of OBM. Historically, Bitcoin miners made most of their revenue from mining. More of them now identify as energy companies, and some public miners, Chernis said, are making up to 40% of their revenue through demand response.
And some public miners are making up to 40% of their revenue through demand response.
David Chernis · 3:08
He followed it with the line that summed up his pitch to the room: in the US, demand response is no longer something a miner can leave on the table and stay competitive.
If you are not participating in demand response in the US now, you are not optimizing your revenue.
David Chernis · 3:16
CPower describes itself as a controls-agnostic automation partner. Chernis said it can integrate at the automation vendor level or at the firmware level, and that it has worked with Foreman since Foreman's start. The company's Enerwise site optimization platform was built to orchestrate microgrids with solar, wind and storage and to bid those assets into day-ahead energy markets. He said CPower uses it in PJM now and plans to bring it to the other markets, and that it is using AI to analyze market conditions so that complex systems can be automated.
Three ways automation pays
Chernis split demand response revenue into three streams. The first is classic programs such as capacity, where a site is paid for its ability to curtail whether or not the grid activates it. His example was a 30-megawatt site that earns passively every day and meets its commitment when the utility or ISO calls on it during a heat wave or extreme cold.
The second is energy. That covers avoiding price spikes, but also reducing a site's capacity tag and its transmission peak, the figures that set the rate it pays for power through the following year. He was careful to call this on-bill savings rather than revenue.
The third is ancillary services in the day-ahead markets, which he described as new territory opened up by the level of automation CPower and its partners now have.
A strike price in PJM
Chernis walked through an example from PJM. The strike price is set from the mine's marginal costs. When the market price rose above it during an unexpected spike, the mine shifted automatically into curtailment, cleared all the economic hours above the strike price, earned revenue and avoided the spike in its energy cost. In the hours it was not curtailing, the site was awarded synchronized reserves, a second layer of day-ahead revenue. When the economic program and the reserves overlap, the economic program takes priority.
Breaking the sound barrier
The other theme was speed. Chernis said CPower could not do this work without its integration partners and that curtailment times keep falling.
I call it the breaking the sound barrier.
David Chernis · 7:31
He said CPower has linked a mine to follow the New York ISO's six-second base point signals in ancillary services, which makes it eligible to earn in New York's regulation capacity program, similar to frequency regulation in PJM. Besides Foreman, he named Auradyne, integrated at the firmware level, and Awesome Miner among the automation partners, and pointed to six- and two-second curtailments, up and down, of the kind being done in Scandinavia.
Where AI compute fits
Chernis closed on AI. Foreman, he noted, had made an announcement about AI load the day before his session. His argument was that a mine is an energy shell with capacity and available rack space, which is why AI compute is moving into mining infrastructure.
AI compute load is increasingly dropping into mining infrastructure because if you think of a mine as an energy shell
David Chernis · 8:27
He cited an expected 300% growth in demand for data over the next seven years, and raised the problem that comes with it: AI workloads carry service level agreements, a large language model is not something an operator can stop, and even inference may or may not be curtailable. His proposal was to treat GPU compute the way miners treat hashrate, assuming a window of about 15% of the load that can be ramped up and down in small increments. That, he said, would let AI sites earn in real-time markets and take part in grid services, so that a utility or ISO sees them as an asset rather than a burden, which in turn helps those build-outs get approved.
He made a similar point about sustainability. Through CPower's partnership with WattTime, the company can quantify the emissions avoided through demand response participation, which it provides to its miners and which he said helps at the regulatory level.
Chernis ran out of time at that point and invited the audience to CPower's booth, 311. When this session was recorded, Bitcoin traded at $61,272; on October 16, 2026 it was $82,165.