Bitcoin Hashrate Growth Is a Boardroom Call in 2026

On September 19, 2026, Bitcoin’s mining difficulty jumped 4.16 percent, its largest single step up since June, and the network’s estimated hashrate pushed back toward the 900 to 950 exahash range after a long flat stretch. Bitcoin itself had run to roughly $85,500, a 2026 high, up more than 12 percent on the week, while the difficulty print landed at 132.76 trillion. On the surface that reads like a clean growth story: higher price, more revenue per machine, idle rigs switching back on.

The reality is more interesting than that. The hard part of hashrate growth, adding brand new capacity rather than re-energizing rigs that were already built and sitting idle, is no longer settled on the network. It is settled in the boardrooms of a dozen or so listed mining companies, each of which now asks the same question about every megawatt it controls: should this power hash Bitcoin, or should it be leased to an artificial-intelligence lab for the next twenty years? Through 2026, the answer has kept coming back the same way, and it is not mining.

This is the demand side of a story whose supply side we covered separately in our look at the silicon ceiling: chips and foundries set the physical limit on how fast net-new hashrate can be built. The capital that would order those chips, and the power that would run them, is what this piece is about. Somewhere in the last year, the unit of analysis for Bitcoin’s growth stopped being the network and became the firm.

What hashrate growth actually measures

Hashrate is the total number of SHA-256 guesses the network makes every second while searching for the next block. Nobody can measure it directly, because miners are anonymous and there is no central meter. It is estimated from how fast blocks are actually found relative to the ten-minute target, anchored by the difficulty setting. When people say the network runs at 900 exahashes per second, they mean roughly 900 billion billion guesses a second, and they mean it as a smoothed estimate, not a reading off a dial.

Growth, then, has two very different components. One is cheap and fast: energizing machines that already exist but were switched off when they stopped being profitable. The other is slow and expensive: manufacturing and deploying net-new rigs, which means ordering chips, securing power, and building sites. The first kind of growth can happen in days when price rises, which is exactly what the September rebound looked like. The second kind takes 12 to 18 months and a capital-allocation decision that, in 2026, competes directly with AI. Keeping the units straight helps.

The 2026 stall nobody was forced into

The network first held one zettahash per second on a seven-day average in early September 2025 and peaked near 1,162 exahashes on October 19, 2025, when Bitcoin traded around its all-time high. It then fell back below a zettahash on January 17, 2026, at 988 exahashes on a seven-day basis, and spent most of the year on a plateau. That is where 2026 stops resembling any previous slowdown.

Bitcoin’s hashrate has declined only a handful of times, and until this year every decline was coerced. The 2018 and 2022 drops were capitulations, machines shutting off because they were underwater in a bear market. The 2021 drop was a ban, China forcing more than half the network offline in a single quarter. The 2026 plateau is the first that came from a voluntary choice by profitable operators to point capital and power somewhere else. As Jamie Redman of Bitcoin.com News put it when difficulty and hashprice diverged this autumn, “The flat hashrate indicates that miners are evolving very carefully, and they might be facing hardware limits.” Nobody switched the machines off. They just stopped buying more of them.

That distinction is the whole story. A capitulation reverses when price recovers. A ban reverses when policy changes. A capital-allocation decision reverses only when mining out-competes the alternative, and in 2026 the alternative is the most capital-hungry buildout in technology. Reference the eight increases and ten decreases the network logged through mid-September and you get a net decline of roughly 12 percent for the year, even as the price of Bitcoin climbed.

The megawatt is the real unit now

To understand the boardroom math, drop the exahash and think in megawatts. A miner does not really own hashrate; it owns contracted power and the machines that turn that power into either Bitcoin or heat. The question is what each megawatt earns. Mining revenue per megawatt-hour depends entirely on fleet efficiency and hashprice, the daily revenue a petahash of capacity produces. At the roughly $40 hashprice that prevailed after this autumn’s rally, the spread between a new fleet and an old one is stark.

Now put an AI colocation lease next to that. A megawatt rented to a hyperscaler or an AI lab under a long contract does not swing with Bitcoin’s price or the difficulty thermostat. It pays a fixed, much higher figure, and it pays it for a decade or two. HIVE Digital’s management has framed the gap bluntly, telling the market that 10 megawatts of NVIDIA H100 GPUs generate revenue equivalent to 100 megawatts of Bitcoin mining, a ten-to-one advantage in revenue density. John Todaro, an analyst at Needham and Company, put the finance version of the same point on the record: “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining.” Once a management team internalizes that ratio, growing the mining network becomes the thing you do with the megawatts nobody wants for AI.

How a Bitcoin miner becomes an AI landlord

The mechanism that redirected all this power is a lease. The template looks the same across the sector: a listed miner takes a site it already permitted, powered, and cooled for mining, signs a multi-year colocation agreement with an investment-grade counterparty, collects a large upfront prepayment, and finances the fit-out with non-recourse debt secured against the contract. The miner keeps the real estate and the interconnection, which is the scarce asset, and hands over the compute business to a tenant.

Riot Platforms provides the cleanest worked example. In August it disclosed a 20-year lease with the AI lab Anthropic for 191 megawatts at its Rockdale, Texas campus, locking in $9.1 billion of contracted revenue in the base term, and up to $16.1 billion with two five-year extensions. Deployment starts in December 2027. The Anthropic deal sits on top of an earlier lease with the chipmaker AMD, bringing Rockdale’s contracted AI capacity to 241 megawatts. For scale, Riot’s entire Bitcoin mining segment produced $113.7 million of revenue in the second quarter. A single 20-year AI lease is worth many times its annual mining business, and it does not care what Bitcoin does next week.

That is the repricing in one sentence: a long-dated stream of payments backed by investment-grade credit is a fundamentally different asset than hashrate-dependent block rewards, and capital markets value it accordingly. The demand pulling these megawatts is the same demand chasing every GPU cluster and decentralized compute network, a race we track in our coverage of the decentralized AI training market. Bitcoin miners happen to own something that buildout desperately needs: sites with power already flowing.

The 2026 scoreboard: who stacks hashrate, who leases it away

Public miners now sort into two camps. The pure-plays still treat hashrate as the product and keep adding it. The hybrids treat their power and land as the product and are steadily converting the best of it to AI. The line is blurring fast, because even the pure-plays are exploring AI, but the posture still tells you where each firm’s marginal megawatt is going.

The pure-plays: still racing the machine

A shrinking group still runs mining as the main event. MARA Holdings lifted its energized hashrate about 22 percent year on year to roughly 70 exahashes, one of the largest fleets on the network, and continues to chase its own power generation. Bitdeer sits near the top of the public-miner hashrate ranking at around 73 exahashes, with the added twist that it designs and sells its own ASICs, so growing the network also sells hardware. HIVE Digital is one of the few clearly expanding mining capacity, building out more than 300 megawatts of hydro-powered sites in Paraguay toward a 35-exahash target.

The logic that keeps these firms mining is the mirror image of the AI trade. Every megawatt a rival leases away is a megawatt that stops competing for blocks, so the difficulty thermostat quietly hands the miners who stay a larger share of the same subsidy. A pure-play is, in effect, a leveraged bet that Bitcoin’s price will reward patience more than a fixed lease would, and that the steady exit of hybrid capacity will keep its own machines profitable at a lower difficulty. MARA has leaned into that reasoning by chasing owned power generation rather than rented grid load, on the view that the cheapest long-run megawatt is one you control outright.

The cleanest pure-play is American Bitcoin, a majority-owned subsidiary of Hut 8 that is also backed by the Trump family. It runs a highly efficient fleet of roughly 28 exahashes and positions itself as an accumulator, producing coins below the going market price and holding them rather than diversifying into leases. Its very existence is a tell: to run a pure-play mining strategy at scale in 2026, it helped to spin up a dedicated vehicle for it, because the parent’s other megawatts were headed to AI. That contrast is the whole industry in one corporate structure.

The hybrids: cashing out the megawatts

The larger camp is converting. The deal book announced in 2026 is enormous, and most of it belongs to companies that started the decade as Bitcoin miners. TeraWulf signed a 20-year lease with Anthropic for about 401 megawatts at its Kentucky campus, worth roughly $19 billion over the term, with two five-year renewal options that could stretch it to thirty years. Core Scientific, fresh off its shareholders rejecting a $9 billion all-stock buyout from CoreWeave, leased 377 megawatts of critical IT capacity to AMD across multiple sites. IREN, formerly a pure Bitcoin miner, signed a five-year, $9.7 billion cloud contract with Microsoft, complete with a 20 percent prepayment and NVIDIA GB300 systems at its Texas facility.

Hut 8 offers the most telling case. It signed a 15-year, $9.8 billion lease for 352 megawatts at a Texas campus that it had originally underwritten to serve its own mining subsidiary, American Bitcoin, then repositioned to AI as demand and pricing shifted. Megawatts earmarked for hashing were redirected to a tenant before a single miner shipped. CleanSpark, long seen as one of the last big pure-plays, told the market that new mining investment no longer made much sense at prevailing hashprices and opened talks to lease part of its Georgia data center to Meta. Across the sector, CoinShares counts more than $70 billion of announced AI and HPC contracts, and public miners sold about 32,000 BTC in the first quarter of 2026 alone, more than in all of 2025, partly to fund the pivot. One industry tracker put it starkly: listed miners shed roughly a fifth of their combined Bitcoin hashrate as AI revenue accelerated.

Why the market pays more for an AI megawatt

The share prices explain the behavior. Through 2026, mining equities have traded less on Bitcoin production and more on contracted AI capacity. Todaro captured the shift for Needham clients: “Investors are almost exclusively valuing Bitcoin miners for their HPC and AI opportunities at this point.” When a lease with an investment-grade tenant can be underwritten, discounted, and borrowed against, it is worth more per megawatt to equity holders than a volatile stream of block rewards that rises and falls with hashprice.

Matthew Sigel, head of digital-asset research at VanEck, described miners as sitting on a gold mine in terms of the value their pre-energized power holds for AI, a line widely reported through the year. The appeal is the same one drawing capital toward every contracted, cash-flowing corner of crypto, the pattern we described in our piece on DeFi’s higher yield floor after the Fed’s rate hike: in a world of higher rates, a durable, contracted return beats a variable one that depends on a token price. For a mining company, the AI lease is the closest thing to a bond it can manufacture from the assets it already owns.

None of this is free money. The counterparty risk is concentrated in a handful of AI labs and hyperscalers, the buildouts require billions in fresh debt, and a cooling in AI demand would leave miners holding half-converted sites. But as of this autumn, the market is pricing the AI story and discounting the mining one, and management teams are responding to the signal their own share prices send.

The difficulty thermostat rebalances whatever they decide

Here is the reassuring part for the network. Bitcoin does not need any particular company to keep mining, because difficulty automatically rebalances the game. Every 2,016 blocks, roughly two weeks, the protocol adjusts the target so blocks keep arriving about every ten minutes. When firms pull capacity, blocks slow, difficulty falls at the next retarget, and the miners who stayed earn more per unit of hashrate. That higher margin pulls capacity back. Nikolaos Panigirtzoglou of JPMorgan described the loop cleanly: “When Bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.” Run it in reverse and you get September, when a price rally lifted margins and idle rigs came back fast enough to print a 4.16 percent difficulty jump.

So the network’s total hashrate is best understood as a residual: the sum of whatever capacity the boardrooms leave in mining, plus whatever the thermostat pulls back by lifting margins. The AI pivot does not necessarily shrink the number. It slows net-new growth, changes who owns the marginal machine, and makes the whole curve more sensitive to price than to any single company’s plans.

The price bet under every boardroom model

Every one of these leases was signed against a spreadsheet, and the key variable in that spreadsheet is the price of Bitcoin. Hashprice, the daily revenue per petahash, moves almost one for one with price when the network is flat, which is why this autumn’s rally lifted it more than 22 percent even as machine counts barely changed. When hashprice is high enough to beat both a miner’s operating cost and the return on an AI lease, capital flows back to mining. When it is not, the megawatts keep going to AI. That is the entire decision, compressed.

The recovery in price that made September’s difficulty jump possible was itself a macro event, tied to record ETF inflows and the market’s read on Federal Reserve policy, which we covered in our analysis of the post-hike rally. CoinShares has been explicit that the resumption of hashrate growth toward its 1.8 zettahash year-end forecast is contingent on Bitcoin recovering toward $100,000. Below that, hashprice stays soft and the AI lease keeps winning the megawatt. James Butterfill, the firm’s head of research, has called this one of the most challenging periods for miners since the last halving, with the weighted-average cash cost to produce a Bitcoin around $79,995 in late 2025 and some firms projected to earn up to 70 percent of revenue from AI by the end of 2026.

Does a shrinking miner base weaken security?

If profitable companies are leasing capacity away from mining, the fair question is whether Bitcoin is getting less secure. The security budget, the total value miners earn for defending the chain, is roughly the block subsidy times price times the number of blocks. At 3.125 BTC per block, about 144 blocks a day, and a price near $85,000, that is on the order of $14 billion a year. An attacker who wanted to rewrite recent history would need to out-hash that entire apparatus.

Campbell Harvey of Duke University has modeled the cost. His updated estimate puts a majority-hashrate attack at around $8 billion, about 50 basis points of Bitcoin’s value, once you pair the hardware and power with a large short position that profits from the crash the attack would cause. He notes that “the difference today is the derivatives markets.” Matt Prusak, an executive at American Bitcoin, pushes back on the practicality, arguing that economic feasibility kills the 51 percent thesis because acquiring that much hardware quietly would take years and exchanges would freeze suspicious activity first. Either way, two things protect the budget as miners rotate: efficiency, which keeps total hashrate high even as machine counts thin, and difficulty, which lifts survivor margins whenever capacity leaves. The AI pivot changes who defends the network far more than it changes whether the network is defended.

The 2028 clock changes the math again

The capital-allocation decision gets harder in about eighteen months. At the 2028 halving the block subsidy drops from 3.125 to 1.5625 BTC, and at a constant price the security budget halves with it. Transaction fees are supposed to fill the gap, but they still run well under one percent of the block reward on a typical day. Fred Thiel, chief executive of MARA, has been among the most direct about what that implies. He calls mining “a zero-sum game” in which adding capacity makes it harder for everyone else, and predicts that “by 2028, you’ll either be a power generator, be owned by one, or be partnered with one.”

Read against that clock, the AI pivot looks less like a distraction and more like a hedge. Firms with a second, contracted revenue stream are better placed to survive a subsidy cut that arrives whether or not price doubles. The 2028 halving lands in a very different macro setting than past cycles, a theme we unpack in our look at halving cycle math in the first cycle without cheap money. Miners that have diversified their megawatts are, in effect, buying insurance against the moment the subsidy math turns.

What to watch: which way the megawatts flow

Because hashrate growth is now a corporate decision, the useful signals are corporate, not just on-chain. A short watchlist for the rest of 2026 and into 2027:

• Difficulty retargets. A run of increases means the thermostat is pulling capacity back and mining margins are winning at the margin; a run of decreases means the megawatts are still flowing to AI.
• Hashprice versus the forward curve. When spot hashprice trades above the six-month forward, the market expects the rally to fade; the gap is a live read on whether new mining capacity pencils out.
• Megawatts under contract. In each quarterly report, watch the split between mining capacity and AI or HPC capacity. That single line describes where the firm’s growth is going better than any hashrate figure.
• New lease announcements. Each multibillion-dollar, decade-long lease is a block of power leaving the mining pool, usually for good.
• Price and flows. ETF inflows and a move back toward $100,000 are the fastest way to make mining competitive again with a signed AI lease.

On the policy side, the ground is comparatively settled in the United States. The Securities and Exchange Commission’s Division of Corporation Finance stated in March 2025 that proof-of-work mining, solo or pooled, does not involve the offer or sale of securities, which keeps the mining side of these hybrid companies outside the securities perimeter even as their AI leasing businesses grow. That clarity is one reason the pivot has been able to move as fast as it has: the miners converting to AI landlords are not fighting a regulatory war on the mining side while they do it.

The bigger picture is a quiet inversion of what Bitcoin mining was supposed to be. For years the story was that cheap, often stranded energy would chase the block reward to the ends of the earth. In 2026 the block reward is chasing energy, and losing the bid to artificial intelligence at the best sites. Bitcoin’s hashrate will keep climbing when price makes mining the better use of a megawatt, and it climbed this September. But the decision now runs through a boardroom, and the boardroom has options it did not have a year ago.

Frequently Asked Questions

Is Bitcoin’s hashrate still growing in 2026?

Yes, but unevenly. After a flat, first-of-its-kind voluntary plateau through most of 2026, hashrate climbed again in September as a price rally to about $85,500 brought idle rigs back online, pushing difficulty up 4.16 percent on September 19 to 132.76 trillion. Net-new capacity, the kind that requires ordering machines, is growing slowly because listed miners are directing capital toward AI instead.

Why are Bitcoin miners switching to AI?

Because a megawatt earns far more, and far more predictably, leased to an AI lab than mining Bitcoin. HIVE Digital estimates 10 megawatts of NVIDIA H100 GPUs match the revenue of 100 megawatts of mining. Miners already own the scarce inputs AI needs, permitted sites with power flowing, so companies like Riot, TeraWulf, IREN and Hut 8 have signed multibillion-dollar, decade-plus leases with AI labs and hyperscalers.

Does the AI pivot make Bitcoin less secure?

Not obviously. The security budget is still around $14 billion a year, and difficulty automatically raises the margins of miners who stay whenever capacity leaves, which pulls hashrate back. Duke’s Campbell Harvey estimates a 51 percent attack would cost roughly $8 billion. The pivot changes who mines and slows net-new growth more than it changes whether the network is defended.

Which Bitcoin miners are still pure-play in 2026?

Few remain fully focused on mining. American Bitcoin, a Hut 8 subsidiary backed by the Trump family, runs a roughly 28-exahash accumulator strategy. HIVE Digital is still expanding mining in Paraguay, and MARA and Bitdeer keep very large fleets while beginning to explore AI. Most other listed miners, including Riot, Core Scientific, TeraWulf, IREN and CleanSpark, are now hybrids converting power to AI.

What is hashprice and why does it matter for hashrate growth?

Hashprice is the daily revenue a unit of hashing power earns, quoted per petahash per day, and it sat near $40 after the autumn rally. It matters because it is the number every mining boardroom runs its models on: when hashprice beats both operating costs and the return available from an AI lease, capital flows back into mining and hashrate grows. When it does not, the megawatts go to AI.

By Priya Reddy, HOGE Wire mining and markets desk.