Bitcoin Hashrate Growth: From Kilohashes to a Zettahash

For roughly fifteen years, one chart in Bitcoin looked almost boring in its consistency. The network hashrate, the total computing power that miners point at the blockchain, went up. It went up through bull markets and bear markets, through China’s mining ban and four halvings, through exchange collapses and a global pandemic. A metric that began life measured in thousands of hashes per second on a laptop processor grew, by the autumn of 2025, into more than a sextillion hashes per second, a figure with twenty-one zeros in front of the decimal.

Then, in 2026, the curve did something it had never done outside of a government crackdown: it bent. Bitcoin’s hashrate posted its first first-quarter decline in six years, and mining difficulty fell below its year-earlier level for only the second time in history. The cause was not a ban or a blackout. It was money, and specifically the growing pull of artificial intelligence on the same power, capital, and real estate that Bitcoin miners depend on.

This is a history of that growth curve, told with the numbers rather than around them. It traces how hashrate climbed from kilohashes to a zettahash, what actually powered each leg of the ascent, and why 2026 is the first year the story stopped being a straight line up and to the right. Prices below are in US dollars, and the regulatory frame is the US Securities and Exchange Commission, since American operators now run close to a third of the network. Understanding the curve is really understanding the security budget of the largest proof-of-work network on the planet, because every hash on that chart is a dollar and a joule spent making the ledger harder to rewrite.

What hashrate growth actually measures

Hashrate is the number of guesses per second that the entire Bitcoin network makes as it searches for a valid block. Miners run the SHA-256 hashing algorithm over slightly different versions of a candidate block, billions of times per second, until one machine finds a hash below the network’s target. The rate of those guesses, summed across every machine on Earth, is the hashrate.

The subtle part is that nobody measures it directly. There is no meter bolted to the network. Hashrate is estimated backward from two things the blockchain does record: how much work the difficulty target demands, and how quickly blocks are actually arriving. The standard approximation is that hashrate roughly equals difficulty multiplied by 2^32 and divided by the 600-second target block time. Plug in the mid-August 2026 difficulty of about 127.5 trillion and you get an implied hashrate near 912 exahash per second. Because block timing is random, single-day estimates jump around by ten percent or more, which is why analysts smooth them into seven-day and thirty-day moving averages before drawing any conclusion.

So when people say hashrate grew, they mean one of three things happened, usually all at once: more machines were plugged in, each machine got more efficient at turning watts into hashes, or cheaper power let older machines stay switched on. Growth in the curve is a proxy for capital and energy committed to defending the ledger. That is why it is watched so closely, and why its recent stall carries more weight than a simple price wobble.

From kilohashes to zettahashes: reading the ladder

The units climb in factors of a thousand, and Bitcoin has walked up nearly the entire ladder in fifteen years. It helps to see the whole scale at once before tracing the history that filled it in.

Each rung is a thousand times the one below it. A single modern Antminer, a box roughly the size of a shoebox, now produces more hashrate on its own than the entire global network did in 2013. The network today, near 900 exahash per second, is on the order of a billion times more powerful than it was the day the first ASIC shipped. No other computing network in history has scaled its raw throughput this far this fast, and it has done so with no central planner deciding how much capacity to add. The market did the allocating, one power contract at a time.

The first decade: CPUs, GPUs, and the ASIC big bang

In January 2009, Bitcoin ran on ordinary computer processors. Satoshi Nakamoto mined the genesis block on a CPU, and for the first year and a half so did everyone else. Network hashrate sat in the kilohash and megahash range, small enough that a single enthusiast’s desktop represented a meaningful slice of it. Difficulty barely moved, because there was barely any competition.

The first arms race began in October 2010, when miners realized graphics cards could run SHA-256 far faster than CPUs. A GPU delivered roughly a hundred times the throughput, according to hardware histories of the era, and difficulty started its long climb. Field-programmable gate arrays, or FPGAs, arrived around 2011 as a brief intermediate step, more power-efficient than GPUs but quickly overtaken and never widely deployed.

The real inflection was the application-specific integrated circuit. When Canaan Creative shipped the Avalon 1 in January 2013, a chip built to do nothing but hash Bitcoin blocks, it delivered about 66 gigahash per second and rendered every other kind of hardware obsolete within months. Difficulty exploded. The network raced from gigahashes through terahashes and into petahashes in under two years, and hobbyists mining on laptops were priced out for good. From that point on, Bitcoin mining stopped being a pastime and became an industry with a supply chain, a capital cycle, and an unforgiving efficiency treadmill that has run without pause ever since.

Industrialization: crossing 1 EH/s and 100 EH/s

The ASIC era’s defining machine was Bitmain’s Antminer S9, launched in 2016. It produced 13.5 terahash per second at about 98 joules per terahash, and it was so durable that units were still running profitably in cheap-power regions years later. Around the start of 2016, with fleets of early ASICs multiplying across China, the network crossed one exahash per second for the first time, one quintillion hashes, a number that would have sounded like science fiction three years earlier.

From there the curve went fully industrial. Warehouses of S9s in Sichuan, Inner Mongolia, and later Texas pushed hashrate up another order of magnitude. On January 4, 2020, the network first crossed 100 exahash per second, a milestone it took roughly eleven years to reach from the genesis block. That eleven-year span to the first 100 EH/s is worth holding onto, because the next chapter would compress a comparable amount of growth into a matter of weeks.

The 2020 halving, which cut the block subsidy to 6.25 BTC, briefly squeezed margins, but the arrival of far more efficient hardware more than absorbed the shock. By the end of 2020 the newest Antminer S19 series was delivering around 95 to 110 terahash per unit at roughly 30 to 34 joules per terahash, a threefold efficiency gain over the S9 that made every incremental watt buy far more security. The pattern was set: each halving pressured revenue, and each hardware generation answered by squeezing more hashrate out of the same electricity.

The vertical years: 500 exahash to one zettahash

The stretch from 2023 to 2025 is where the growth curve went close to vertical. On November 22, 2023, the seven-day average hashrate crossed 500 exahash per second for the first time, as tracked by Luxor’s Hashrate Index. Less than a year later, in October 2024, it set fresh records above 770 exahash, and by mid-December 2024 it had pushed past 800, all while the April 2024 halving cut the subsidy to 3.125 BTC and should, in theory, have slowed the buildout.

Then came the zettahash. Daily estimates flickered above 1,000 exahash per second in the first months of 2025, but the number the industry treats as real, the sustained seven-day average, first held above one zettahash per second on September 15, 2025, per data compiled by TFTC. Five days later the network printed a single-day record of 1,091 exahash per second, reported by Bitcoin.com News. The smoothed average kept climbing into late October, with CoinShares putting the autumn peak near 1.16 zettahash per second.

The pace of that final leg is hard to overstate. In the six weeks after crossing one zettahash, the network added more than 100 exahash per second, roughly as much raw computing power as it had accumulated in its entire first eleven years of existence. A worthwhile caution for readers who check chart sites directly: some automated summaries report an all-time high near 1.44 zettahash for September 2025. That figure does not match any primary tracker or research report and appears to be an aggregation error; the credible autumn-2025 peak sits in the 1.1 to 1.16 zettahash range.

The efficiency engine behind the curve

None of this happens without the chips. Hashrate growth is really two curves stacked on top of each other: more machines, and better machines. The second curve, measured in joules per terahash, which is how much electricity it takes to produce a given amount of hashing, is what let the network multiply its throughput a thousandfold without multiplying its power bill a thousandfold.

The Antminer S21 XP’s rated 13.5 joules per terahash is roughly seven times more efficient than the S9 that defined the 2016 network, a gain achieved in under a decade. That improvement is the quiet engine under every milestone in the table above. It also explains a structural shift in where mining lives. Once a machine sips power this efficiently, the marginal advantage of chasing the world’s absolute cheapest and least reliable electricity shrinks, and the advantage of siting near stable grids that can also host other high-value computing loads grows. Hold that thought, because it turns out to be central to what happened in 2026.

What actually drives hashrate higher

Strip away the narrative and hashrate growth comes down to a handful of forces, each of which can accelerate the curve or, when it reverses, drag it back down.

• Price. Higher BTC prices raise mining revenue per unit of hashrate, which funds more machines. The 2025 run to an all-time high near $126,000 bankrolled the sprint to a zettahash.
• Hardware efficiency. Each ASIC generation that lands more terahash per watt lets operators expand hashrate without proportionally expanding power draw or operating cost.
• Cheap and stranded power. Flared gas, curtailed wind and hydro, and demand-response arrangements give miners electricity below prevailing industrial rates, keeping older fleets profitable and switched on.
• Public-company capital. Nasdaq-listed miners raised billions in equity and convertible debt through 2024 and 2025 to buy machines at scale, professionalizing what had been a fragmented cottage industry.
• The halving cycle. Each halving forces a hardware refresh, as operators running inefficient machines either upgrade or power down, which paradoxically tends to raise the average efficiency of the surviving fleet.

When several of these line up, as they did from 2023 to 2025, the curve goes vertical. When they reverse together, as they began to in 2026, growth stalls. That is exactly the setup for the year the trend finally broke.

Difficulty: the thermostat that absorbs the growth

Bitcoin has a built-in governor that keeps block production steady no matter how much hashrate joins or leaves. Every 2,016 blocks, roughly every two weeks, the protocol recalculates the difficulty target so that blocks keep arriving about every ten minutes. If hashrate surges and blocks come too fast, difficulty ratchets up; if hashrate leaves and blocks slow, it ratchets down, with each adjustment capped at a fourfold move in either direction.

Difficulty is therefore the mirror image of the hashrate curve, linked by that same approximation of hashrate equals difficulty times 2^32 divided by 600 seconds. As of mid-August 2026, difficulty sits at about 127.5 trillion after a small positive adjustment on August 8, with the next retarget due around August 22. The early estimate for that adjustment swung from roughly minus three percent to roughly flat as more blocks came in, a clean reminder that difficulty forecasts are noisy until an epoch is well underway. We unpack the mechanics and the 2026 numbers at length in our guide to Bitcoin mining difficulty; for this story, the key point is that difficulty is how the network absorbs hashrate growth, and how it reveals hashrate decline.

That is why a falling difficulty is such a loud signal. It does not mean machines merely failed to join. It means machines that were already hashing went dark.

2026: the year the curve finally bent

For almost its entire history, Bitcoin’s hashrate fell on a year-over-year basis exactly once, after China banned mining in 2021 and knocked more than half the network offline in a single summer. Everything else, every bear market and every halving, was a pause or a wobble, not a genuine reversal.

2026 changed that. Bitcoin’s hashrate posted its first first-quarter decline in six years, slipping around four percent in the opening months of the year while hovering near one zettahash, CoinDesk reported. By August, mining difficulty had fallen roughly 14 percent from its 2026 high and dropped below its year-earlier level for only the second time ever, a shift CoinDesk tied to plunging revenues and flagged by Hashrate Index as just the second such annual decline in the network’s history.

The economics behind the reversal were brutal. With BTC trading near $63,500 against production costs that CoinShares put in the $80,000 to $90,000 per coin range for many operators, and hashprice, the daily revenue a unit of hashrate earns, sitting near $31.73 per petahash per day per Hashrate Index, a meaningful share of the global fleet was running at or below its cash cost. Older machines were the first to power down. But unlike the shakeouts of 2018 or 2022, the capital that would once have simply idled and waited for the next bull market suddenly had somewhere more attractive to go.

The AI pivot: compute competing with itself

The somewhere is artificial intelligence. The same attributes that make a site good for Bitcoin mining, namely large interconnection to the grid, cheap power, cooling, and permits already in hand, also make it good for the data centers that train and serve AI models. And the economics are not close. Revenue per megawatt from high-performance computing and AI hosting runs far higher, and far more predictably, than from mining, which turns every operator’s power contract into a live question of whether that megawatt should be hashing at all.

The reallocation is already enormous. Publicly listed miners have announced more than $70 billion in AI and HPC-related contracts, and CoinShares expects some operators to earn as much as 70 percent of their revenue from AI by the end of 2026, up from roughly 30 percent today, according to the firm’s head of research James Butterfill. When CoinDesk reported the first-quarter hashrate drop, the explanation was blunt: many listed miners are switching to AI and high-performance computing infrastructure, where returns are higher and more predictable.

This is the deep reason the growth curve bent. For fifteen years, capital that entered the mining industry mostly stayed there, because a warehouse full of ASICs could not easily become anything else. Now it can, or at least the building and the power contract can. The broader repricing of decentralized compute, which has hammered tokens like Gensyn’s $AI even as demand for GPUs soars, is the same tide pulling megawatts away from SHA-256 and toward matrix multiplication. Hashrate growth now competes for resources with the most capital-hungry buildout in modern computing, and it does not always win.

What all that hashrate actually buys: security

All that computing power is not an end in itself. It is the price of rewriting Bitcoin’s history. To reverse a confirmed transaction, an attacker needs to out-hash the honest network long enough to build a longer competing chain, the so-called 51 percent attack. The more hashrate defends the chain, the more it would cost to marshal a majority against it, which is why the growth curve doubles as a security-budget curve.

Duke University finance professor Campbell Harvey put a number on it. In an October 2025 paper, he estimated that a one-week majority-hashrate attack would cost roughly $6 billion, mostly hardware and data-center construction, or about 0.26 percent of Bitcoin’s network value at the time, as reported by crypto.news. By mid-2026 he had refined the model. By pairing the hashrate acquisition with a large short position in offshore Bitcoin derivatives, an attacker could profit from the very price crash the attack would cause, bringing the effective cost to around $8 billion, or roughly 50 basis points of Bitcoin’s value. As Harvey told Bitcoin.com News, the cost is now “about 50 basis points of the value of bitcoin,” and “the difference today is the derivatives markets.”

Not everyone is alarmed. Matt Prusak, president of American Bitcoin Corp, has argued that economic feasibility kills the 51 percent thesis, pointing out that quietly accumulating that much hardware would take years and that exchanges would freeze suspicious activity long before an attacker could ever cash out. The honest summary is that Bitcoin’s base layer has never suffered a successful 51 percent attack, while smaller proof-of-work coins like Bitcoin Gold and Ethereum Classic have been hit repeatedly, precisely because their hashrate, and therefore their attack cost, is a rounding error next to Bitcoin’s. The growth curve is what keeps Bitcoin out of that category.

The other security curve: Ethereum staking

Bitcoin is not the only network whose security budget has its own growth curve. Ethereum, which abandoned mining for proof of stake in 2022, secures itself not with energy but with capital: validators lock up ether as collateral that can be slashed if they misbehave. Roughly a third of all ether is now staked, and the count of active validators runs into the hundreds of thousands, though that number has started to decouple from the total staked after the May 2025 Pectra upgrade raised the maximum effective balance per validator and let large operators consolidate many validators into one.

The contrast is instructive. Bitcoin’s security scales with joules and silicon; Ethereum’s scales with dollars of staked collateral and the yield paid to attract them. That yield has been under pressure as more ether competes for the same issuance, a squeeze we cover in our look at validator economics, and Wall Street’s growing appetite for staking exposure is reshaping the market, as we detail in our piece on institutional liquid staking. Both networks are buying the same product, the improbability of a successful attack, but they pay for it in different currencies. And in 2026 both of those currencies, energy for Bitcoin and yield-bearing capital for Ethereum, got more expensive to supply at the same time.

Fees, halvings, and the long-run security budget

Here is the long shadow over the hashrate curve. The revenue that pays for all that security comes overwhelmingly from the block subsidy, currently 3.125 BTC per block since the April 2024 halving, and that subsidy is programmed to halve again around 2028, to 1.5625 BTC. Transaction fees, which are supposed to eventually replace the subsidy as the security budget, still make up less than one percent of block rewards on a typical day.

That arithmetic sets up a slow-motion question. If the subsidy keeps halving while fees stay thin and BTC’s price does not compensate, miner revenue per block shrinks, and with it the incentive to keep growing hashrate. On other chains, ordering value and priority fees, the phenomenon we explain in our guide to MEV strategies, have grown into a meaningful supplement to block rewards; Bitcoin’s fee market remains far thinner and more sporadic, spiking during Ordinals and Runes manias and then subsiding to almost nothing. Whether Bitcoin can grow a durable fee market before the subsidy fades toward insignificance is arguably the single most important open question for the next decade of the hashrate curve. The 2026 stall is a small preview of what happens when revenue and hashrate stop reinforcing each other.

How Washington sees mining, and where the curve goes next

For US operators, who now run close to a third of the network, the regulatory picture turned friendlier in 2025. The SEC’s Division of Corporation Finance stated in March 2025 that proof-of-work mining, whether solo or through a pool, does not involve the offer or sale of securities, because miners are rewarded for their own computational effort rather than the managerial efforts of others. That removed a long-standing legal cloud over domestic operators. The same month, the White House established a Strategic Bitcoin Reserve, signaling that the federal posture had shifted from suspicion toward accommodation.

So where does the curve go from here? CoinShares forecasts hashrate reaching 1.8 zettahash per second by the end of 2026 and 2 zettahash by the end of March 2027, in its Q1 2026 mining report, but with a large asterisk: that path assumes BTC recovers toward $100,000. If the price instead stays near current levels, the more likely outcome is a plateau, with efficient fleets holding steady while a swelling share of megawatts drifts toward AI hosting. The two-scenario spread is unusually wide. For the first time, the honest forecast for Bitcoin hashrate is not a straight line up, but a fork: a return to record growth if price cooperates, or a genuine ceiling if the AI trade keeps winning the bidding war for power. The straight line that held for fifteen years is, at least for now, a thing of the past.

Frequently Asked Questions

What is Bitcoin hashrate, and why does it grow?

Hashrate is the total number of SHA-256 guesses per second that Bitcoin miners make while searching for valid blocks, estimated from difficulty and block times rather than measured directly. It grows when more machines join the network, when new ASICs produce more hashing per watt, or when cheap power keeps older machines profitable. Rising hashrate signals more capital and energy committed to securing the ledger.

What is the highest Bitcoin hashrate ever recorded?

Bitcoin set a single-day record of 1,091 exahash per second, about 1.09 zettahash, on September 20, 2025, and the smoothed seven-day average peaked near 1.16 zettahash per second that autumn, per CoinShares. Some chart sites display an all-time high near 1.44 zettahash, but that figure does not match primary trackers and appears to be an aggregation error.

Why did Bitcoin’s hashrate fall in 2026?

Weak mining economics, with BTC trading well below many operators’ production costs, pushed unprofitable machines offline, and capital shifted toward artificial intelligence and high-performance computing, where returns are higher and more predictable. The result was Bitcoin’s first first-quarter hashrate decline in six years and only the second year-over-year drop in mining difficulty in the network’s history.

Does higher hashrate make Bitcoin more secure?

Yes. More hashrate raises the cost of a 51 percent attack, the only way to reverse confirmed transactions. Duke professor Campbell Harvey estimated in 2026 that such an attack would cost roughly $8 billion, about 50 basis points of Bitcoin’s value. Bitcoin’s base layer has never been successfully attacked this way, unlike smaller proof-of-work coins with far less hashrate behind them.

How high will Bitcoin’s hashrate go?

CoinShares projects 1.8 zettahash per second by the end of 2026 and 2 zettahash by the end of March 2027, but only if BTC recovers toward $100,000. If the price stays near current levels, growth is more likely to plateau as miners divert power and capital to AI hosting. The outlook is now a genuine fork rather than a straight line up.

Written by Marcus Okafor, mining and markets correspondent at HOGE Wire.