Bitcoin’s Hashrate Found a Second Job: Balancing the Grid

One year ago this week, Bitcoin printed the two biggest numbers in its history within days of each other. On 6 October 2025 the price closed at a record $126,080, and the network hashrate, the raw amount of computing power pointed at the chain, pushed to roughly 1.1 zettahashes per second, the most work ever spent securing a single ledger. Twelve months later the price sits near $85,500, about a third below that peak, and the hashrate has never been back.

On its own, a lower hashrate is not a crisis. The number has fallen before. What is new in 2026 is the reason. The drawdowns of 2018 and 2022 were capitulations, overleveraged miners going dark as a bear market dragged revenue under their electricity bills. The crash of 2021 was a ban, with China evicting more than half the network in a single quarter. The stall of 2026 is neither. Nobody went bankrupt in a wave, and no government kicked anyone out. The machines simply found something more lucrative to do with their power, and in doing so Bitcoin’s hashrate quietly took on a second job. It became one of the largest flexible electrical loads on the grid, a gigawatt-scale switch that utilities increasingly pay to flip off.

That shift, from reading hashrate as a security-and-sentiment gauge to reading it as an energy-market signal, is the real story behind the flat line. Here is how mining became a grid asset, what operators like Riot Platforms earn to power down, why artificial intelligence cannot yet do the same trick, and what all of it means for the question the hashrate chart was always supposed to answer: is Bitcoin still safe?

The number, one year on

Start with where the network actually sits. Luxor’s Hashrate Index put the third-quarter average at roughly 940 exahashes per second on a 30-day basis, down 6.3% from the second quarter and the second straight quarterly decline, leaving the network about 12% below its December high near 1,066 EH/s. Daily spot readings are noisier and have bounced around 1 zettahash, but the trend that matters is the one on the longer average, and it points sideways to down. (Be wary of chart widgets that flash an all-time high above 1.4 ZH/s; that figure is an artifact of a noisy single-block estimate, not a real sustained record.)

Difficulty, the self-adjusting target that keeps blocks arriving about every ten minutes, tells the same story. It stands near 132.7 trillion, roughly 15% under the record 156 trillion set in November 2025, after a summer of cuts and a partial rebound through September. Hashprice, the daily revenue a miner earns per unit of hashing, has recovered to about $40 per petahash per day from a five-year low near $27.70 in late June, per Hashrate Index data cited by news.bitcoin.com. Even so, that sits well below what miners earned per unit of work before the 2024 halving cut the block subsidy in half and before the network swelled to its current size.

Put those together and 2026 marks a first. Bitcoin’s hashrate has now spent a full year below a prior high without a ban or a mass bankruptcy forcing it there. Every earlier plateau had an obvious villain. This one has an opportunity cost.

What hashrate measures, and why it reads differently now

A quick refresher, because the metric is more slippery than it looks. Hashrate is the number of guesses per second that miners collectively throw at the puzzle of finding the next block. Crucially, no one meters it. The network only sees how fast blocks are arriving, so analysts back out an estimate from the pace of blocks and the current difficulty, using the relationship hashrate is approximately difficulty times 2 to the 32nd power, divided by the 600-second block target. That is why the daily figure is jumpy and why two trackers can disagree by a hundred exahashes on the same afternoon. The honest version of the number is a multi-day average, not a spot reading.

For most of Bitcoin’s life, the hashrate chart was read in two ways. It was a security gauge, because more hashing means a more expensive network to overpower. And it was a sentiment gauge, because miners add machines when they are bullish and switch them off when they are scared, so a rising line looked like confidence. Both readings are still valid. But in 2026 a third has moved to the front, as explored in our look at what Bitcoin’s hashrate growth really means: the line is now, above all, an energy-demand signal. It tells you how many megawatts the most price-sensitive buyer of electricity on earth is willing to consume at today’s Bitcoin price, and how quickly that buyer will stand aside when someone else bids higher.

A dispatchable gigawatt

Here is the property that makes mining strange among heavy industries: it can stop on a dime and lose almost nothing. A smelter that interrupts a pour ruins the batch. A chip fab that loses power can scrap weeks of wafers. A Bitcoin miner that powers down forfeits only its lottery tickets for the blocks mined while it was dark, and it can be hashing again within minutes of flipping the breakers. The work is stateless and infinitely interruptible. In grid terms, a mining site is a controllable load resource, a block of demand an operator can shed on command and restore just as fast.

That matters because the blocks are big. The US Energy Information Administration, in a 2024 analysis before an industry lawsuit forced it to withdraw its emergency survey and destroy the data it had gathered, estimated that crypto mining already drew somewhere between 0.6% and 2.3% of all US electricity, with 137 commercial-scale facilities across 21 states, a load comparable to the entire state of West Virginia, per the agency’s figures reported by Utility Dive. On some regional grids, a cluster of mining halls is the single largest interruptible load connected to the system. That combination, enormous and sheddable, is exactly what a stressed grid wants.

The Texas laboratory

Nowhere is the experiment more advanced than Texas, whose ERCOT grid is an island with cheap wind, volatile prices, and a chronic fear of blackouts. Miners there have long played two games at once. They earn credits through ERCOT demand response programs by agreeing to curtail on short notice, and they manage their bills around the grid’s Four Coincident Peak system, which charges big users based on their draw during the four highest-demand intervals of the year. Idle during those peaks and your transmission costs fall for twelve months.

In 2025 Texas made the arrangement official. Senate Bill 6, signed by Governor Greg Abbott on 21 June 2025 and effective immediately, rewrote the rules for large loads, which it defines as sites drawing 75 megawatts or more. As law firm Bracewell summarized, the bill directs ERCOT to competitively procure demand reductions from those customers and requires them to curtail during declared grid emergencies on at least 24 hours’ notice, with a new study fee of at least $100,000 just to get in the interconnection queue. The scale of the queue is the headline: ERCOT is now tracking more than three times as many large-load interconnection requests as in 2024, and nearly 69% of the roughly 189 gigawatts in the pipeline comes from data centers, according to reporting by Inside Climate News.

The irony is thick. The rules written to tame the coming wave of artificial-intelligence campuses lean on a flexibility that Bitcoin miners spent a decade proving out. Texas learned how to manage a giant, fickle load by watching mining sites switch on and off, and it is now trying to make the rest of its new demand behave the same way.

What the grid pays miners to stop

The money is not theoretical. The clearest ledger belongs to Riot Platforms, one of the largest public miners and a heavy user of ERCOT demand response. During a brutal Texas heat wave in August 2023, the state effectively paid Riot $31.7 million in a single month to power down and sell its contracted electricity back into the market, a sum that dwarfed the value of the Bitcoin it would have mined. That was a spike, but the pattern has become routine. In its second-quarter 2026 results, Riot reported about $31.1 million in power curtailment credits across the first half of the year, including $10 million in the second quarter alone, or roughly $6,335 for every Bitcoin it produced.

Across the industry, demand response and power sales have historically supplied somewhere between 2% and 10% of a large miner’s revenue. Read that back through the hashrate chart and something clicks. A line that drops because machines are being paid to sit idle is not a sign of a sick network. It is a sign that the network’s hardware has a second income stream, one that pays precisely when it refuses to hash.

Worth more to the grid than people realize

Luxor’s own analysts have started framing the hashrate decline in these terms rather than as weakness. In the company’s third-quarter heatmap, chief operating officer Ethan Vera called the drop “a structural shift, not just a cyclical low,” and argued that “miners are worth more to the grid than people realize.” His sharpest point was a comparison. Mining capacity, Vera noted, consists of flexible loads that can power down instantly, something the AI capacity now replacing them cannot yet do. In other words, the very machines being pushed aside by artificial intelligence have a grid virtue their replacements lack.

His colleague Kaan Farahani, a research analyst at Luxor, put the economics plainly: the least efficient machines switch off first while the rest migrate toward the cheapest power, and in a contracting network, as he put it, “holding flat is a win.” The subtext of both quotes is that hashrate is no longer a race to the top. It is a managed resource, dialed up and down against the price of electricity and the value of the next-best use for a megawatt.

Why AI cannot yet do a miner’s job

The flexibility gap is the crux of the whole story, so it is worth being precise about it. An artificial-intelligence data center wants firm, uninterrupted power, 24 hours a day. A training run that loses power can lose days of progress; an inference service that pauses breaks the latency guarantees it sold to customers. Those workloads are the opposite of sheddable, which is why the gigawatts flooding into ERCOT’s queue frighten grid planners: they add to the peak instead of flexing around it. A mining hall, by contrast, is indifferent to being switched off and can hand its power back in seconds.

That moat is real, but it is not permanent, and honest coverage has to say so. Researchers are now showing that AI compute can be made partly flexible too. As energy-industry trade press has reported, a June arXiv study drove a 130-kilowatt GPU cluster through more than 200 grid events with full compliance, cutting 30% of its draw within 40 seconds, while a field trial published in Nature Energy trimmed 25% of a 256-GPU cluster’s power for three hours during peak stress without breaking its service guarantees. Galaxy’s head of power, Blake King, is among the executives arguing the industry can learn to curtail. The catch is scale: those pilots are tiny next to the 100-megawatt-plus campuses under construction, and the most valuable AI work, real-time inference, remains the least interruptible. For now, no-penalty, instant curtailment is still something miners do better than anyone, and that is the race unfolding in the wider market for decentralized GPU power.

The Duke headroom argument

Flexibility is not just a revenue trick for miners; it may be the key that unlocks the AI buildout without a decade of new transmission. A widely cited study from Duke University’s Nicholas Institute, “Rethinking Load Growth” by Tyler Norris and co-authors, found that the existing US power system could absorb roughly 76 gigawatts of new demand, and up to 126 gigawatts across all balancing authorities, if those new loads agreed to curtail for just 0.25% of their maximum uptime. One quarter of one percent. The grid has enormous latent headroom, the paper argues, as long as the new tenants are willing to step aside for a handful of hours a year.

Bitcoin miners are the working proof of concept for that idea. They are already the flexible load the Duke model imagines, already wired into demand-response markets, already demonstrating that gigawatts can be parked and repowered on command. The policy conversation about how to power the AI era is, in effect, a conversation about how to make everyone else behave more like a Bitcoin mine.

Flexibility is also a feature of weakness

It would be too tidy to end there, because the same property that makes miners good grid citizens is what makes their hashrate fall. Mining can afford to be interrupted because hashing earns relatively little per megawatt-hour. When a buyer who values the same power more, an AI tenant on a 20-year lease, shows up, the miner is the one who yields. Fred Thiel, chief executive of MARA Holdings, has been blunt about where that leads, telling CoinGeek that mining is “a zero-sum game” in which “the floor is your energy cost,” and that “by 2028, you’ll either be a power generator, be owned by one, or be partnered with one.” The economics are not subtle: revenue per megawatt and profit margins are far higher for high-performance computing and AI colocation than for mining.

The emblem of the trade is Riot itself. In August 2026 the company signed a 20-year lease for 191 megawatts at its Rockdale, Texas campus, a deal worth about $9.1 billion that Bloomberg reported was struck with the AI lab Anthropic, as covered by CoinDesk. That capacity does not flex in and out of mining; it leaves hashing for good. Multiply that across the sector, where CoinShares estimates AI could supply up to 70% of listed miners’ revenue by the end of 2026, and the hashrate stall stops looking like an accident. It is the sound of the most flexible load on the grid being steadily outbid for its own electricity.

How grid economics bent the curve

The mechanism that translates all of this into a falling line is the difficulty adjustment, Bitcoin’s thermostat. When revenue per unit of hashing drops below a miner’s running cost, the rational move is to switch off, which is exactly what the flexible-load story predicts. JPMorgan’s Nikolaos Panigirtzoglou described the loop to TFTC: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.” His team pegs all-in production cost near $78,000 a coin, estimates 15% to 20% of the fleet runs underwater at times, and measures a difficulty-to-price sensitivity of about 0.62, meaning the network’s work tracks price with a lag rather than leading it.

Through 2026 that thermostat ran cold, then warmed again. Difficulty posted its first year-on-year decline since the 2021 China ban, falling about 14% from its 2026 high by midsummer, as CoinDesk reported, before a rebound as Bitcoin rallied back toward $85,000 and idled machines came back online. James Butterfill of CoinShares has called it “one of the most challenging periods” for miners even as the sector signed more than $70 billion in AI and HPC contracts. The point is that the hashrate curve is now a residual. It is whatever is left over after the energy market, the Bitcoin price, and the AI bid have each taken their cut of the available power, which is why understanding it increasingly means understanding the economics laid out in our piece on difficulty and the 2028 halving.

Does a flatter network still mean secure?

This is the question the hashrate chart was invented to answer, and the flexible-load era reframes it. Security has never actually been about the raw hashrate number. It is about the dollar cost to assemble enough hardware and power to overwrite Bitcoin’s history, and that cost scales with price far more than with the EH/s headline. At today’s levels the annual security budget, hashprice times hashrate, runs to roughly $14 billion a year, an amount an attacker would have to credibly threaten to burn. Research by Campbell Harvey of Duke’s Fuqua School put the cost of a one-week majority attack at about $6 billion, later revising it toward $8 billion, roughly half a percent of Bitcoin’s market value, once you account for financing the attack with a short position.

A network down 15% from its peak work is still defended by that wall of capital, and Bitcoin has never suffered a successful 51% attack in its history. There is even an argument that a fleet of flexible machines is more robust, not less: capacity that can be idled for economics can be repowered for defense, and much of the hardware diverted to AI still sits in facilities that could, in extremis, be pointed back at the chain. That is a very different security model from proof of stake, where protection comes from bonded capital rather than burnt energy, a contrast we drew out in validator economics. Bitcoin’s security, by contrast, is something the power grid is actively paying to keep flexible.

Who regulates a power plant that mines?

As the meaning of hashrate migrated from finance to energy, so did the regulators who matter. In Washington, the question of whether mining is a securities activity has effectively been settled: the SEC’s Division of Corporation Finance stated in March 2025 that proof-of-work mining, whether solo or pooled, does not involve the offer or sale of a security, a posture consistent with the quieter enforcement climate we described in the year after the crackdown. The federal energy regulator’s attempt to even count miners collapsed when the industry sued and the EIA scrapped its survey.

What binds a miner now is not the SEC but the public utility commission and the grid operator. The interconnection agreement, the curtailment obligation under a law like Texas SB6, the demand-response contract, the transmission-cost formula: those are the rules that decide whether a site is built, how much it pays, and when it must go dark. The compliance surface of a modern mining business looks less like a broker-dealer’s and more like a merchant power plant’s. The machines are the same SHA-256 hashers they always were. The oversight has moved to the people who keep the lights on.

What to watch into 2028

The near-term signposts are familiar. The next difficulty retarget, due in mid-October, and the Federal Reserve meeting on 27 and 28 October will nudge the price, and the price will nudge how many machines stay on. If Bitcoin holds its recovery, expect the thermostat to keep warming and some idled hashrate to return; if it slips back under production cost, the flexible load will shed again, as it is designed to.

The structural clock is the one that reframes everything. The 2028 halving will cut the block subsidy from 3.125 BTC to 1.5625, halving the energy-funded portion of the security budget at a constant price and raising the stakes on both transaction fees and that second income stream from the grid. Watch three things: how fast Texas and other grids operationalize their large-load curtailment markets under rules like SB6, whether AI compute genuinely learns to flex and erodes the miner’s unique pitch, and whether the hashrate line ever reclaims its 2025 record. Increasingly, that last line will tell you less about whether traders are bullish and more about who is winning the auction for the world’s cheapest electricity. For a decade, Bitcoin’s hashrate was a scoreboard for a bull market. In 2026 it became a readout of the power grid, and that is a far more interesting thing to watch.

Frequently Asked Questions

Why is Bitcoin’s hashrate still below its 2025 record?

Because the decline is economic, not forced. With hashprice near multi-year lows and Bitcoin about a third below its October 2025 high, the least efficient machines stay switched off, and a growing share of mining capacity has been redirected to artificial intelligence and high-performance computing, which pay far more per megawatt. The difficulty adjustment then lowers the bar automatically, so blocks keep arriving roughly every ten minutes even with less power online.

What does it mean that Bitcoin mining is a flexible load?

A flexible or interruptible load is one that can cut or stop its power use on short notice without serious harm. A miner can idle its machines in seconds, hand the electricity back to the grid during a heat wave or a price spike, and restart minutes later, losing only the chance to find blocks in between. That makes large mining sites valuable to grid operators, who pay them to curtail when demand is high.

How much do miners get paid to power down?

Enough to matter. Riot Platforms earned about $31.1 million in power curtailment credits in the first half of 2026, including $10 million in the second quarter, or roughly $6,335 for every Bitcoin it mined. During a Texas heat wave in August 2023 it collected $31.7 million in a single month for pausing. Demand response has historically supplied between 2% and 10% of a large miner’s revenue.

Does a lower hashrate make Bitcoin easier to attack?

Not meaningfully at these levels. The cost to rewrite Bitcoin’s history still runs into the billions, with one Duke study putting a one-week majority attack at roughly $6 billion to $8 billion. Security scales with the dollar value of the hardware and energy committed, which price drives more than the raw hashrate figure, and a network of flexible machines that can repower quickly is arguably more resilient, not less.

Can AI data centers become flexible grid loads like miners?

Partly, and not yet at scale. Field trials have shown GPU clusters trimming 25% to 40% of their power for hours during grid stress, but those tests are tiny next to the 100-megawatt-plus campuses being built, and latency-sensitive work such as real-time inference cannot pause. For now, instant, no-penalty curtailment is still something Bitcoin miners do better than anyone, which is the heart of their pitch to the grid.

By Marcus Okafor, mining and energy desk, HOGE Wire.