Bitcoin’s Hashrate Climbs Again, but Silicon Sets the Ceiling

Bitcoin spent most of the summer trading below the cost of producing it. On September 22, 2026, that stopped being true. BTC changed hands near $86,400 on CoinGecko, up double digits on the week and roughly 14% above its September 15 close of $75,584, the day the US Senate voted down the CLARITY Act. US spot Bitcoin ETFs then pulled in $998.95 million in a single day, their largest haul of 2026, led by BlackRock’s IBIT. A wave of short liquidations did the rest, and Bitcoin cleared $86,000 for the first time since January. The timing was the striking part: Bitcoin had bottomed a week earlier, on the very days the US Senate killed the CLARITY Act and the Federal Reserve raised interest rates, then rebounded to close above its 50-week moving average for the first time in nearly a year.

For miners, price is oxygen. When Bitcoin trades above the marginal cost of production, idled machines switch back on, and the network’s total computing power, its hashrate, starts to climb. You can read it in the difficulty adjustments: the number that governs how hard it is to win a block rose 4.16% on September 19 to about 132.76 trillion, the largest single step up since late June, with another increase near 2.82% projected for early October. After a year defined by a stalling, sometimes falling hashrate, the machine is warming up again.

Here is the catch, and the subject of this piece. The easy part of that recovery is nearly free, and the hard part is not available at any price. Re-energizing mothballed rigs takes a phone call to a hosting site. Growing the network beyond its old peak takes new chips, and new chips are gated by a supply chain that three Chinese companies design, two foundries fabricate, and a tariff regime taxes on the way in. In 2026, the ceiling on Bitcoin’s hashrate is not power. It is silicon.

What hashrate actually counts

Hashrate is the number of SHA-256 guesses the entire network makes every second as miners race to find the next block. It is measured in hashes per second, and Bitcoin has climbed the whole metric ladder in fifteen years, from kilohashes on a laptop CPU to nearly a zettahash on a global fleet of purpose-built machines.

One thing to keep in mind throughout: no meter reports the true figure. Nobody can poll every anonymous machine, so hashrate is estimated from the one thing the blockchain does record, namely how fast blocks arrive relative to the current difficulty. The rough formula is difficulty times two to the thirty-second power, divided by the 600-second block target, which puts the network around 950 EH/s at September’s difficulty. Short windows are noisy, because block discovery is a random Poisson process; a quiet hour can read 700 EH/s and a lucky one 1,100 EH/s on the same day. That is why analysts smooth over seven or thirty days and treat difficulty, which moves only every 2,016 blocks, as the hard anchor.

The number is climbing again

Difficulty retargets are Bitcoin’s thermostat. Every 2,016 blocks, about two weeks, the protocol compares how long those blocks actually took against the two-week target and adjusts so the next stretch averages ten minutes per block. Add hashrate and difficulty rises; pull machines and it falls.

For most of 2026, the story was falling. Difficulty dropped in ten of the year’s adjustments and bottomed near 124.93 trillion in June, roughly 20% below the November 2025 record near 156 trillion. Then the direction changed. It rose 1.31% on September 5 and 4.16% on September 19, and the early-October estimate points up again. Two, likely three, consecutive increases is the clearest sign in a year that machines are coming back online.

To see why the turn matters, rewind twelve months. Hashrate set its record near Bitcoin’s own price high in October 2025, then bled lower through a brutal winter as prices fell below the cost of production; by mid-January the seven-day average had slipped back under one zettahash. The decline ran all the way into a summer floor, difficulty bottoming near 124.93 trillion on June 13, its lowest of the year and about 15% under where 2026 had started. What broke the slide was not a technical fix but a price: once Bitcoin cleared its production cost again, the machines that had gone quiet had a reason to run. The table below traces the whole year, and the September entries are the first sustained upturn on it.

JPMorgan’s Nikolaos Panigirtzoglou described the mechanism earlier this year, running in the other direction: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.” Flip the price move, as September did, and the loop runs in reverse. Price rises above production cost, dormant machines return, hashrate climbs, and difficulty follows it up. The recent retargets are that quote played backward.

The easy hashrate and the hard hashrate

Not all hashrate is created equal, and the difference matters enormously right now. Roughly 150 EH/s of older equipment powered down during the winter and spring lows. That capacity did not disappear; it sat unplugged, waiting for revenue to clear its electricity bill. When Bitcoin doubled off those lows, most of it could come back within days, reconnect to a mining pool, and start hashing. Call it the easy hashrate: capacity that already exists, already sits in a data center, and only needs the economics to justify the power draw. The coordination layer that reconnects those rigs is itself being modernized through efforts like Stratum V2, the protocol rewrite reshaping how miners talk to pools, but the machines themselves are the ones already built.

The hard hashrate is different. To push the network above its October 2025 peak, you cannot just re-energize what is idle; you have to add machines that do not yet exist. That means ordering new ASICs, which means waiting on a chip supply chain that cannot flex on the timeline a price rally moves on. A rally can restore the network to where it was in a month. Growing it to genuinely new highs is a twelve-to-eighteen-month manufacturing question. The two are so different that treating hashrate as one number hides the most important fact about 2026: the cheap capacity is coming back, and the expensive capacity is stuck in a queue.

The bottleneck is a chip, not a megawatt

For a decade, the mining story was about electricity: who had the cheapest kilowatt, which government offered stranded hydro, when China banned and Texas welcomed. Power still matters, but the binding constraint on growth has quietly moved upstream, to the machines themselves. When hashrate went flat this month even as revenue jumped, Jamie Redman of Bitcoin.com News put it plainly: “The flat hashrate indicates that miners are evolving very carefully, and they might be facing hardware limits.”

Those hardware limits start with an unusually concentrated market. Almost every Bitcoin ASIC on Earth comes from three companies. An April 2025 University of Cambridge study, cited by Cointelegraph, put Bitmain at 82% of production, MicroBT at 15%, and Canaan at about 2%, a combined 99%. All three are Chinese. Analysts call it an oligopoly, which understates it; this is closer to a single dominant design house, Bitmain, with two much smaller rivals. When one firm’s roadmap slips or one foundry reallocates capacity, the entire network’s growth rate moves with it.

Three designers, two foundries

Designing an ASIC is not the same as building one. Bitmain, MicroBT, and Canaan draw the chips; they do not own the fabrication plants that etch them. For the leading-edge nodes that make a modern miner efficient, the world has essentially two options, TSMC and Samsung, and TSMC dominates the top end because its advanced processes deliver the best performance per watt, the single number miners obsess over. Samsung sits a distant second, historically slowed by yield problems. China’s SMIC would be a third path, but US export controls cap it near seven nanometers because it cannot buy the extreme-ultraviolet lithography needed to go further; it is stuck on older deep-ultraviolet tools.

That leaves miners competing for fab time against the richest buyers in technology. TSMC’s cutting-edge three-nanometer and two-nanometer capacity is spoken for years in advance by Apple, Nvidia, and the hyperscale cloud providers building AI accelerators. A Bitcoin ASIC designer does not outbid Apple for wafers. Switching foundries or nodes is no quick fix either; requalifying a chip on a new process is a twelve-to-eighteen-month redesign. So even with Bitcoin at $86,000 and margins the fattest they have been all year, the network cannot simply order its way to a higher hashrate. The wafers are not there to be bought, which is why price and hashrate, once tightly linked, have come apart in 2026.

The geopolitics reach past commercial scheduling. In January 2025 the US Commerce Department added Sophgo, a chip-design firm tied to Bitmain’s co-founder, to its Entity List after a TSMC-made part matching a Sophgo design turned up inside a Huawei AI processor. The action did not ban a single Antminer, but it showed how exposed the mining supply chain is to the machinery aimed at China’s AI sector: the lithography tools, the foundries, and now the designers all sit inside a national-security contest that has nothing to do with Bitcoin and everything to do with who is allowed to buy advanced silicon. A miner planning a fleet three years out has to price in export controls it has no influence over.

Why a better chip does not always mean more hashrate

There is a subtler reason new hardware does not translate cleanly into network growth. The newest machines are so much more efficient than the fleet they replace that miners increasingly use them to swap out old rigs rather than to add capacity.

Bitmain’s Antminer S23 Hydro, shipping in 2026, runs at about 9.7 joules per terahash, the first mainstream miner under 10 J/TH. Set that against the roughly 1,200 J/TH of the first ASICs in 2013 and you have a hundredfold efficiency gain in thirteen years. But TheMinerMag analysts flagged the twist: “Instead of fueling network expansion, the S23 Hydro and similar rigs may be used to upgrade existing fleets. That’s a notable shift.” A miner with a fixed power contract who replaces old S19s with S23s gets far more hashrate per megawatt, but the network as a whole does not necessarily gain, because that operator’s total power draw is capped, and the retired machines’ hashrate comes off the top. Efficiency and hashrate growth used to move together. In 2026 they have partly decoupled.

The implication for the network is uncomfortable for anyone expecting a straight line up. Cheaper, better machines are available; what is scarce is the fab capacity to make enough of them to move a network already close to 950 EH/s. Each new generation mostly renews the fleet. Expanding it is a separate, slower project.

Tariffs and the customs choke point

The supply chain has a second pinch point, and it sits at the US border. The United States hosts roughly 38% of global hashrate but imports something like 99% of the machines that produce it, almost all from China. That mismatch is now a policy variable, not a footnote.

Tariffs on Chinese imports stand at about 25%, down from a peak above 100% earlier in the trade dispute but still a real tax on every rig landed in America. Enforcement adds uncertainty on top of the rate. In November 2024, US Customs and Border Protection halted delivery of thousands of Bitmain ASICs during an investigation into an affiliate’s alleged links to sanctioned chipmaker Huawei; roughly 10,000 machines were seized and only released in early March 2025, per Cointelegraph’s reporting. A four-month customs hold is, in effect, four months of hashrate that never came online.

Bitmain, MicroBT, and Canaan have all announced US assembly lines to blunt the tariffs, but assembly is not fabrication; the chips are still etched in Asia, and the metal inputs are still dutiable, so the move softens the tax without removing the dependency. Jaran Mellerud, chief executive of Hashlabs Mining, warned that the tariffs “could lead to a collapse in US demand for Bitcoin mining rigs,” with surplus inventory flowing to cheaper jurisdictions abroad. For a network whose center of gravity is now North America, a tax on the hardware is a tax on hashrate growth.

The onshoring is already underway, and its limits are instructive. Bitmain opened its first US production line in late 2024 and shipped scores of tons of Antminer components to a Delaware affiliate through 2025; MicroBT advertises tariff-free machines assembled on American soil; Canaan is running early US trials. Every one of these projects is final assembly, workers bolting together boards and power supplies that arrive from Asia, rather than fabrication. The chips at the heart of each rig are still etched an ocean away, so the moves trim the tariff bill without touching the dependency that actually caps growth.

AI is bidding for the same wafers and the same watts

Miners cannot simply outlast the silicon shortage, because their biggest competitor for it is also their most tempting alternative business. Artificial-intelligence data centers want the exact two inputs mining needs: advanced chips and large blocks of cheap, reliable power.

On the silicon side, AI accelerators and Bitcoin ASICs are fabricated on the same leading-edge nodes at the same two foundries, and the AI buyers have far deeper pockets. The race to build out decentralized and centralized AI training compute alike is pulling wafers toward GPUs and away from hashboards. On the power side, the pull is just as strong. CoinShares research head James Butterfill has called 2026 “one of the most challenging periods” for miners, noting that listed operators could earn as much as 70% of revenue from AI and high-performance computing by the end of the year, up from roughly 30%, against more than $70 billion in announced AI and HPC contracts across the sector.

Every megawatt redirected to an AI tenant is a megawatt not hashing, and with interest rates higher after the September Fed hike, the opportunity cost of tying up capital in mining rigs rather than in something that throws off a steadier real yield is sharper than it has been in years. The result is a network that can re-energize its idle rigs when price recovers but struggles to expand, because the marginal chip and the marginal megawatt both have a higher bidder.

The choice is playing out unevenly across the public miners, and the split is worth watching, because it decides how much of the returning hashrate is permanent. A shrinking group are still pure-play accumulators. American Bitcoin, the venture that pairs Hut 8’s fleet with the Trump family, has kept adding machines and reactivating older rigs to mine below the spot price; HIVE Digital is racing to bring a hydro-powered campus in Paraguay online. The larger cohort is diversifying. Riot Platforms is converting a big block of its Texas capacity into a twenty-year AI lease; CleanSpark, after spending much of the year calling AI too uncertain, reversed and began bidding for high-performance-computing sites; Core Scientific, TeraWulf, IREN, and Hut 8 have all signed multi-billion-dollar compute contracts.

For hashrate, the distinction is everything. A pure-play that re-energizes idle rigs adds capacity that stays on the network as long as the machines earn their power. A hybrid that leases a data hall to an AI tenant takes those megawatts off the network for the length of the contract, often a decade or more. So even a strong price recovery produces a lopsided result: the easy hashrate floods back, but a slice of the fleet’s future growth has been quietly signed away to model training. The network can refill toward its old level without ever setting a new record, because it is being rebuilt and partly repurposed at the same time.

Hashprice, the paycheck that decides who plugs in

Whether a given machine comes back online is decided by hashprice, the miner’s revenue per unit of computing power per day. It is the cleanest read on mining economics: multiply the daily block rewards and fees by the Bitcoin price, divide by the network hashrate, and you get dollars per petahash per day.

The last official weekly print from Luxor’s Hashrate Index was $39.36 per petahash per day at the end of August, struck when Bitcoin was near $78,500. Run the same math at $86,000 and a hashrate near 940 EH/s, and hashprice lands around $41 to $42, a first-principles estimate rather than a published figure, but a meaningful lift. That is enough to put a large share of the fleet comfortably back above its electricity cost, which is why the idle rigs are returning and difficulty is rising.

The recovery is not evenly shared, and that too points back to hardware. At the end-of-August hashprice, an efficient sub-14 J/TH fleet earned roughly $136 of revenue per megawatt-hour, while a tired machine in the 25 to 38 J/TH range earned closer to $51, using Hashrate Index’s efficiency tiers. The efficient rigs clear their power bill with room to spare and switch on first; the oldest ones sit at the margin, waiting for either a higher price or a difficulty dip. That ordering is why the machines coming back in September skew toward newer hardware, and why simply owning idle capacity does not guarantee it returns.

The forward market is less convinced the good times last. Six-month hashprice contracts have traded near $37.59, below spot, a backwardation that prices the rally to partly fade as difficulty catches up. And the catch-up is the point. As Panigirtzoglou’s loop runs in reverse and hashrate returns, difficulty climbs and competes the windfall back down. Hashprice is the accelerator that brings idle capacity back; it does nothing to conjure the new chips that genuine growth requires.

What the peak really was, and the phantom record

Because hashrate is estimated rather than measured, bad numbers circulate freely, and one of them is worth flagging. Several popular trackers, including CoinWarz, display an all-time high above 1.4 zettahashes per second, dated to early 2026. It is an artifact of noisy spot readings, not a record.

The mechanism behind the phantom is worth understanding, because it recurs. A spot hashrate reading is really a measure of how quickly the last handful of blocks arrived; a lucky streak of fast blocks, which happens purely by chance, makes the network look far more powerful than it is for a few hours. Feed those spikes into an automated all-time-high field and a tracker will happily immortalize a number the network never sustained. The fix is to lean on smoothed averages and on difficulty, which cannot spike on luck because it changes only every two weeks.

The best-corroborated peak is about 1.162 EH/s, that is 1.16 ZH/s, reached on October 19, 2025, near Bitcoin’s own price high. The seven-day average first held above one zettahash in early September 2025 and fell back below it around January 17, 2026, at roughly 988 EH/s. Everything since has traded under that October 2025 mark. So the honest way to describe late September 2026 is that the network sits somewhere around 890 to 950 EH/s depending on the window, climbing, but still meaningfully below the record it set almost a year ago. When a chart shows a suspiciously round or suspiciously high all-time high, check it against the difficulty-implied figure before repeating it.

Security still scales with the number

Hashrate is not just an industrial statistic; it is Bitcoin’s defense budget. Multiply hashprice by hashrate and you get the network’s annual security spend, the money an attacker would have to match to rival it, currently on the order of $14 billion a year. A higher, price-driven hashrate makes the chain more expensive to attack even when the growth is just idle rigs returning.

How expensive is a live debate. Duke finance professor Campbell Harvey has modeled a one-week 51% attack at around $8 billion, “about 50 basis points of the value of bitcoin,” once you account for pairing the hashrate with a large short position in Bitcoin derivatives, so the attack profits from the crash it causes. Skeptics think the threat is academic. Matt Prusak, an executive at American Bitcoin, argues that “economic feasibility kills the 51% thesis,” since accumulating that much hardware quietly is close to impossible and exchanges would freeze suspicious activity first. Either way, the arithmetic ties security to the same silicon supply chain: an attacker faces the identical foundry bottleneck as an honest miner, which is part of why Bitcoin has never suffered a successful 51% attack while smaller chains have. The longer-run pressure is the 2028 halving, which will cut the block subsidy from 3.125 to 1.5625 BTC and, at a constant price, halve the subsidy portion of that budget, a math problem explored in our look at the halving cycle in a world without cheap money. Fees, still well under 1% of miner revenue, will eventually have to carry more of the load.

The scale of that problem is easy to see if you hold the price still and let the subsidy fall on schedule. At a constant $86,000, the subsidy alone pays miners on the order of $14 billion a year today; each halving cuts that in half, all else equal.

Price growth or a real fee market can offset the decay, but neither is guaranteed. This is the deeper reason the silicon story matters: security ultimately rests on miners being paid enough to keep buying and running machines, and every link in that chain, the subsidy, the fee market, the chips, and the power, is under pressure at once.

How Washington sees mining

For US readers, the regulatory frame around mining is calmer than the headlines suggest, with one clear exception. The Securities and Exchange Commission settled the securities question in March 2025, when its Division of Corporation Finance stated that proof-of-work mining, whether solo or pooled, does not involve the offer or sale of securities, because miners rely on their own effort and computing power rather than the efforts of others. That removed a cloud that had hung over the business.

The pressure that remains does not come from the SEC. Tariffs are set by the executive and administered by Customs and Border Protection; export controls on chip-making equipment run through the Commerce Department; and the Entity List actions that entangled a Bitmain affiliate are national-security tools, not securities law. In other words, the biggest regulatory risk to US hashrate growth in 2026 is written in tariff schedules, not securities filings. Miners still face ordinary tax questions on the BTC they produce, an area Congress keeps reworking, but the securities overhang that once shadowed the industry is, for now, lifted.

What to watch next

The recovery is real but partial, and a few markers will tell you how it resolves. The early-October difficulty retarget, estimated up near 2.82%, will confirm whether idle capacity keeps returning or the easy hashrate has already come back. Watch whether the seven-day average reclaims one zettahash for the first time since January; that would signal the network is not just recovering but growing again. On the supply side, the tells sit further upstream: any sign that TSMC or Samsung frees advanced-node capacity for miners, the trajectory of US tariffs, and how aggressively the big listed miners keep trading mining megawatts for AI leases.

The through-line is that 2026 broke the old assumption that hashrate follows price with a short lag. Price came back fast, and the easy hashrate is following. But the network’s ability to set new records now runs through a fab in Taiwan and a customs desk in California as much as through the Bitcoin price. For the first time, Bitcoin’s security curve has a ceiling that money alone cannot lift. The rally decides how quickly the network refills. Silicon decides how high it can go.

Frequently Asked Questions

What is Bitcoin’s hashrate in September 2026?

Depending on the averaging window, roughly 890 to 950 EH/s, climbing after difficulty rose 4.16% on September 19 to about 132.76 trillion. That is still below the record of about 1.162 EH/s set on October 19, 2025. No meter measures it directly; it is estimated from difficulty and block times.

Why is Bitcoin’s hashrate rising again?

Bitcoin’s rally to about $86,000 pushed hashprice back above many miners’ electricity costs, so machines idled during the summer are being switched back on. That extra computing power shows up as consecutive upward difficulty adjustments.

What is really limiting Bitcoin hashrate growth in 2026?

Silicon supply, not power. New ASICs come from three Chinese designers and are fabricated by TSMC or Samsung, whose leading-edge capacity is booked by Apple, Nvidia, and AI chip buyers. Add roughly 25% US tariffs and customs delays, and the network cannot simply order its way to a higher hashrate.

How does AI compete with Bitcoin mining?

AI data centers want the same advanced chips and the same cheap power miners need, and they pay more for both. CoinShares estimates listed miners could earn up to 70% of revenue from AI and high-performance computing by the end of 2026, so every wafer and megawatt redirected to AI is one not producing hashrate.

Does a stalling hashrate make Bitcoin less secure?

Security tracks hashrate times hashprice, roughly $14 billion a year now. A higher hashrate raises the cost of a 51% attack, estimated by one Duke study at around $8 billion. Because attackers face the same chip bottleneck as honest miners, Bitcoin has never suffered a successful 51% attack.

By Marcus Okafor, mining and Bitcoin-infrastructure desk, HOGE Wire.