A bitcoin mine that stops mining is normally a mine losing money. In Texas it is frequently a mine having its best afternoon of the year.
The mechanism is demand response, and once you understand it the whole industry looks different. Large flexible loads are paid to reduce consumption when the grid is tight, and a mining facility is close to the perfect flexible load: it can drop from full draw to nothing in seconds, it does not care about the interruption, and nothing spoils while it is off.
Which means the revenue model has two legs rather than one. Hashing when power is cheap. Curtailing when power is expensive, and being paid for the curtailment.
It exists at all because Texas asked the industry for nothing beyond a power contract. That is worth holding against how the state treats other sectors: the constitution restricts gambling to the lottery, charitable bingo and pari-mutuel racing, with no commercial casinos and no authorised online play, and changing it requires a constitutional amendment rather than a bill. Anyone comparing online casinos available to Texas players is therefore looking entirely at platforms licensed in other jurisdictions, because no domestic casino framework exists to be licensed under. Same legislature, same decade. One industry needed permission it could not obtain, the other needed a substation.
Why Texas Specifically
Three things had to line up, and only one state had all three.
A deregulated grid with real-time pricing. ERCOT runs its own interconnection, largely separate from the rest of the country, with wholesale prices that move by the interval and can spike dramatically under stress. Most grids smooth that away. Texas exposes it, which creates both the cheap troughs and the expensive peaks a two-legged model needs.
Enormous intermittent generation. West Texas wind produces heavily at night when demand is low, and at times has pushed prices to zero or below. A load that can absorb power precisely when nobody else wants it is solving a problem the grid genuinely has.
No permission required. Setting up a data centre needs land, power and an interconnection agreement. It does not need a licence from anyone deciding whether the activity should exist.
How the Payment Actually Works
Worth separating the routes, because they are not the same thing and the reporting conflates them constantly.
Voluntary curtailment on price. No programme involved. The miner watches the real-time price, and when it exceeds the revenue per megawatt-hour from hashing, it powers down and sells nothing rather than buying expensive electricity. This is just arithmetic.
Ancillary services. The miner commits capacity to the grid operator in advance and is paid for standing ready to drop load, whether or not the call comes. That is a payment for availability, which is why it appears as revenue in quarters with no major grid events.
Demand response programmes. Structured arrangements with defined notice periods and compensation for responding.
The second one is the interesting one commercially. A facility earning availability payments has converted a variable-cost business into something with a contracted revenue line, which is a very different proposition to lend against.
The Part That Attracts Criticism
Worth stating plainly rather than skipping.
Critics argue the arrangement amounts to paying an industry to not operate, funded through charges that ultimately reach consumers, and that the load only exists because the payments made it viable. The counter-argument is that flexible load genuinely stabilises a grid with heavy intermittent generation, and that the alternative to a curtailable data centre is an inflexible one.
Both positions have merit and the dispute is live. What is not disputed is the mechanism, which is that a substantial share of the sector’s margin comes from not producing anything.
What the Halving Did
The economics shifted materially when block rewards last halved, and the response was instructive.
Hashing revenue per unit of work fell by half overnight while power costs did not move. That compressed margins across the industry and made the curtailment leg proportionally more important, because a revenue stream unaffected by the halving becomes a larger share of the total when the other stream shrinks.
It also accelerated a pivot that was already under way. Several operators began converting capacity toward AI and high-performance computing hosting, which is a considerably better business per megawatt and does not halve every four years.
The asset that made those companies valuable turned out not to be the mining hardware. It was the power contracts, the interconnection agreements and the substations, none of which care what the racks are computing.
What to Watch
Three things, for anyone following the sector.
Whether availability payments survive scrutiny. They are politically exposed in a way energy prices generally are, and any grid event tends to reopen the question.
How much capacity actually converts to AI hosting. The announcements are numerous and the completed conversions are fewer, and the gap between the two is where the interesting analysis sits.
Whether other grids copy ERCOT’s structure. Flexible load is valuable anywhere with heavy renewable penetration, but most markets are not built to price it by the interval.
The durable lesson is smaller than the headlines and more useful. An industry that looked like it was selling computation was substantially selling something else: the ability to stop instantly, which is a service the grid pays real money for and which almost no other large consumer can provide.
