Bitcoin YearsThe Atlas
Menu

Bitcoin Years › Network › Bitcoin mining history

Bitcoin mining history 2009 → 2026

From a button to a power plant.

In 2009 anyone could mine Bitcoin by ticking a box in Satoshi’s software. Today the same puzzle is solved by warehouses of machines built for nothing else. This is how that happened, step by step, from the people and sources who were there.

983 EH/snetwork hashrate, 7-day average (18 Nov 2026)
135 trilliontimes harder than the first block (difficulty 1 in 2009)
3.125 BTCnew coins per block since April 2024 (50 in 2009)
×571,429less energy per hash than a 2009 desktop CPU

Figures read from /data/network/ when this page was built

  1. “Generate Coins” in Bitcoin v0.1
  2. First difficulty rise, block 32,256
  3. GPU mining introduced
  4. First public OpenCL miner
  5. First mining pool announced
  6. First open-source FPGA miner
  7. getblocktemplate (BIP 22)
  8. First consumer ASICs ship
  9. Hashrate passes 1 EH/s
  10. CPU miner removed from Bitcoin Core
  11. Stratum V2 announced
  12. China crackdown statement
  13. Reward falls to 3.125 BTC
  14. Hashrate passes 1 ZH/s

Prologue · January 2009

A button called Generate Coins

When Satoshi Nakamoto announced the first version of Bitcoin on 8 January 2009, mining did not have a name yet. It was a menu option. “You can get coins by getting someone to send you some, or turn on Options->Generate Coins,” the announcement said.[1] Every copy of the program could make new money, and every copy competed on equal terms: the computer that happened to find a valid block first was paid 50 bitcoin.

Satoshi was frank about the starting point. “I made the proof-of-work difficulty ridiculously easy to start with,” he wrote, so that “a typical PC will be able to generate coins in just a few hours.” He also said what would come next: “It’ll get a lot harder when competition makes the automatic adjustment drive up the difficulty.”[1] Almost everything on this page is the story of that one sentence coming true.

In those first months the competition was tiny. Most early blocks appear to come from very few machines, and one of them left a recognisable fingerprint in the blocks it mined. That pattern, and whether its coins ever moved, is told on our Patoshi page.

“I made the proof-of-work difficulty ridiculously easy to start with.”Satoshi Nakamoto, 8 January 2009

Before the story

What a miner actually does

Every few minutes, somewhere in the world, a machine finds a number that nobody can guess. That is all mining is. A miner takes the 80-byte header of a candidate block, which holds the previous block’s fingerprint, a summary of the transactions and the time, and runs it through SHA-256 twice. The result is a 256-bit number. If that number is below a target set by the network, the block is valid and the miner is paid. If not, the miner changes one field, the nonce, and tries again.

There is no shortcut. SHA-256 gives no hint about which input will produce a small output, so the only strategy is to try, billions of times a second. That is why the work is a fair lottery: your chance of finding the next block is your share of all the guesses being made. And because anyone can check a winning header with a single hash, a block that took the whole network ten minutes to find takes a phone a millisecond to verify.

The target is the dial. Every 2,016 blocks the network compares how long they took with the two weeks they should have taken, and moves the target so that blocks keep arriving about every ten minutes, however many machines join or leave. “Difficulty” is simply how much harder today’s target is than the first one. Try it below with a toy block in your own browser.

Mine a toy block in your browser

Your browser hashes a small demo header with SHA-256, twice, changing the nonce each time, until the result starts with the number of zeros you choose. Each extra zero makes it about 16 times harder.

Nonce0
Your hashrate–
Time–

Latest hashpress “Mine a block”

The real network today expects about 579 sextillion hashes per block, because its difficulty is 135 trillion.

Demo only: the header text is made up and nothing is sent anywhere. Real mining hashes the 80-byte binary header; this toy hashes readable text, which works the same way.

Chapter 1 · 2009

One computer, one vote

For most of 2009, mining meant leaving an ordinary computer on. The white paper had described proof-of-work as “one-CPU-one-vote”, and for a while that is literally what it was: whoever ran the software used the processor they already owned.

Hal Finney, who believed he was the first person besides Satoshi to run the software, remembered it years later: “I mined block 70-something… Those were the days when difficulty was 1, and you could find blocks with a CPU, not even a GPU. I mined several blocks over the next days. But I turned it off because it made my computer run hot, and the fan noise bothered me.”[20] The first miners were hobbyists, and the first cost of mining was noise.

The first sign of competition arrived on 30 December 2009. Block 32,256 was the first to carry a higher difficulty than the first block: 1.18 instead of 1. You can still read it in the block’s own header.[3] Every 2,016 blocks the network measures how fast blocks are arriving and adjusts the target so they come roughly every ten minutes; more miners simply means a harder puzzle.

Satoshi already saw the next step coming. Earlier that month, answering a suggestion on the Bitcoin forum, he asked for restraint: “We should have a gentleman’s agreement to postpone the GPU arms race as long as we can for the good of the network.” His reason was practical: “It’s much easer to get new users up to speed if they don’t have to worry about GPU drivers and compatibility.”[2]

“We should have a gentleman’s agreement to postpone the GPU arms race as long as we can for the good of the network.”Satoshi Nakamoto, Bitcoin Forum, 12 December 2009

Chapter 2 · 2010–2012

The graphics card arms race

The agreement held for about half a year. Graphics cards are built to do thousands of small, identical calculations at once, which is exactly what hashing block headers needs. In May 2010 Laszlo Hanyecz, a developer who had built the first Mac version of Bitcoin, introduced GPU mining to the community as a patch to the Bitcoin source, getting about ten times more hashing out of a MacBook.[21] Satoshi worried, according to Bitcoin Magazine’s account of their emails, that GPUs would “prematurely limit the incentive to only those with high-end GPU hardware”, and Laszlo stopped advertising the code.[21]

He is better known for what he did with the coins: on 22 May 2010 he paid 10,000 BTC for two pizzas, the first known purchase of goods with bitcoin (Bitcoin Pizza Day). By July 2010 a miner known as ArtForz was among the earliest to mine with GPUs, and the secret was out.[4]

Then the knowledge went public. On 18 September 2010 Jeff Garzik paid the developer puddinpop 10,000 BTC, about $625 at the time, to release a CUDA mining client as open source. On 1 October 2010 m0mchil published poclbm, a Python miner for OpenCL that ran on AMD and other graphics cards.[4][5] Anyone with a gaming card could now out-mine a room full of ordinary computers.

The gap was enormous. A fast desktop processor of the time managed about 19 million hashes a second; a Radeon HD 5870 did around 421 million, and did so about twenty times more efficiently per joule.[7] Difficulty, which had been 1 a year earlier, passed 10,000 on 9 December 2010.[4] The Generate Coins button still worked; it had simply stopped mattering.

~19 MH/sCore i7 920 desktop processor, 0.10 Mhash per joule
~421 MH/sRadeon HD 5870 graphics card, 1.96 Mhash per joule
×20more hashes for the same energy

Chapter 3 · November 2010

Mining together

Rising difficulty created a new problem. A single miner might now wait weeks or months for a block, and then receive all 50 coins at once. The income was real on average and almost useless in practice.

On 27 November 2010 a developer known as slush announced a cooperative mining server on the Bitcoin forum: miners would work on the same block template, prove their effort with easier “shares”, and split each reward by work done.[6] On 16 December 2010 it found the earliest documented pool-mined block.[4] It is described as the first publicly known Bitcoin mining pool, and it still operates today under the Braiins name.[6]

Pools solved the waiting problem and created a new question that has followed mining ever since: if thousands of miners point their machines at a handful of pools, who decides what goes into the next block? The answer would come years later, in the protocols themselves.

Chapter 4 · 2011

Reprogrammable silicon

In 2011 hobbyists borrowed a tool from chip designers: the FPGA, a chip whose circuits can be rewired by software. On 20 May 2011 the first open-source FPGA Bitcoin miner was released.[8] By August a board built around a Spartan-6 LX150 chip was doing 100 million hashes a second on 6.8 watts.[9]

That was about seven times more work per joule than a Radeon HD 5870, but FPGAs were expensive and slow to set up; two graphics cards at half the price still produced more raw hashes.[9] FPGAs did not take over. They proved something more important: a chip designed around SHA-256 alone would leave everything general-purpose behind.

Chapter 5 · 2012–2013

Chips that do one thing

An ASIC is an application-specific integrated circuit: a chip that can do exactly one job. For Bitcoin that job is double SHA-256 of an 80-byte header, billions of times a second. ASICMiner’s BE100 chip, dated December 2012, used about 6 joules per gigahash, roughly eighty times less energy per hash than a Radeon HD 5870.[10]

In January 2013 Avalon began shipping the first consumer ASIC miners from China: an initial batch of 300 machines, each doing 60 gigahashes a second and priced at $1,299.[11] One box replaced a shelf of graphics cards. Within months, mining with anything else stopped paying for its electricity.

The original miner inside the Bitcoin software lived on for years as a relic. Bitcoin Core finally removed it in version 0.13.0, in 2016, noting that CPU mining “has been useless for a long time”.[18]

Chapter 6 · 2012–2019

Who builds the block?

As pools grew, the way miners talked to them mattered more. In 2012 Luke Dashjr’s BIP 22 defined getblocktemplate, which hands miners the whole block structure instead of a bare header to hash, so they can see and shape what they mine.[12] Late that year the Stratum protocol, first published on slush’s pool site, replaced the older “getwork” protocol for pooled mining.[13]

Stratum was efficient, but in it the pool chooses the transactions. In August 2019 Braiins announced Stratum V2, designed with ideas from Matt Corallo’s BetterHash proposal: a binary protocol in which miners can send their own block templates, so the people running the machines can choose which transactions to include.[14] It answers the same question pools raised in 2010.

Chapter 7 · 2016–today

From hobby to industry

Once mining became a chip business, it became an energy business, and the network’s computing power began to climb by factors of ten.

By the network’s own data, daily hashrate first passed 1 exahash per second (a billion billion hashes) in January 2016, 10 EH/s in October 2017, 100 EH/s in September 2019 and 1,000 EH/s, one zettahash, in April 2025.[19]

The machines kept improving. The Antminer S9, released in 2017, used about 98 joules per terahash;[15] the Antminer S21, unveiled in 2023, about 17.5.[16] From a desktop processor in 2009 to a modern ASIC, the energy needed for the same amount of hashing fell by a factor of roughly 571,429.

Energy per terahash, 2009 to 2023

How many joules each generation needed for a trillion hashes. Log scale: every gridline is ten times less energy.

Energy needed per terahash, from a desktop CPU to a modern ASIC, log scale10¹ J10² J10³ J10⁴ J10⁵ J10⁶ J10⁷ J10⁸ JIntel Core i7 920CPU · 2009–1010 million J/THATI Radeon HD 5870GPU · 2010–12510,699 J/THSpartan-6 LX150 boardFPGA · 2011–1268,000 J/THASICMiner BE100 chipASIC · Dec 20126,000 J/THAntminer S9ASIC · 201798 J/THAntminer S21ASIC · 202317.5 J/TH

Computed from sources [7], [9], [10], [15], [16]. Typical device figures, not lab measurements.

Network hashrate since 2013

Seven-day average of the daily estimate, log scale. Latest: 983 EH/s (18 Nov 2026).

Bitcoin network hashrate since 2013, exahashes per second, log scale100 TH/s1 PH/s10 PH/s100 PH/s1 EH/s10 EH/s100 EH/s1 ZH/s10 ZH/s2014201620182020202220242026HalvingHalvingHalvingChina

Data: Firsthand Bitcoin network data on this site, updated daily.

Chapter 8 · 2021

The great migration

On 21 May 2021 China’s State Council Financial Stability and Development Committee said it would “crack down on bitcoin mining and trading activities”.[17] What followed is visible in the network’s own numbers.

The network felt it. Average hashrate fell from about 162 EH/s in May 2021 to about 100 EH/s in July, a drop of 39%.[19] On 3 July 2021 the difficulty adjusted down by 28%, the largest fall in our data since 2013.[19]

And then nothing broke. Blocks slowed for a few weeks, the difficulty adjustment did what it was built for in 2009, and by December 2021 average hashrate was back to about 172 EH/s.[19] The protocol had not changed at all.

162 → 100 EH/saverage hashrate, May to July 2021
-28%difficulty adjustment, 3 Jul 2021
172 EH/saverage hashrate, December 2021

Chapter 9 · Every four years

Halvings and the fee question

Miners are paid twice for each block: new coins, and the fees of the transactions inside it. The new coins halve every 210,000 blocks, exactly as the 2009 announcement promised: 50, then 25 in 2012, 12.5 in 2016, 6.25 in 2020 and 3.125 since April 2024.[1] Each halving cuts the main income of every miner in half overnight, and so far the network has carried on each time.

Fees are the part that is meant to take over. So far they have mostly been a small share, with bursts during busy periods. By our estimate the highest yearly share since 2013 was in 2017, about 13% of miner income; in 2026 so far it is about 0.6%.[19] Whether fees will one day pay for the security that new coins pay for today is one of Bitcoin’s long-running open questions; this page does not try to answer it.

Every era’s new coins

Each era lasts 210,000 blocks, about four years. The schedule is fixed by the protocol; the last column is the share of the 21 million cap issued by the end of each era.

EraBlocksRewardNew coinsCap issued
2009–20120–209,99950 BTC10,500,00050.00%
2012–2016210,000–419,99925 BTC5,250,00075.00%
2016–2020420,000–629,99912.5 BTC2,625,00087.50%
2020–2024630,000–839,9996.25 BTC1,312,50093.75%
2024–2028 now840,000–1,049,9993.125 BTC656,25096.88%
2028–20321,050,000–1,259,9991.5625 BTC328,12598.44%
2032–20361,260,000–1,469,9990.78125 BTC164,06299.22%
…and so on, halving every 210,000 blocks until the reward rounds down to zero, around the year 2140.

Supply today: 20,113,354 BTC (95.78% of 21 million), 17 Nov 2026, from Bitview. Exact cap: just under 21 million because rewards are rounded down to whole satoshis.

How much of miner income came from fees

Fees paid to miners as a share of fees plus new coins, by year. New coins estimated at 144 blocks a day at each period’s reward.

Share of miner income from transaction fees, by year0%5%10%15%2013: 1.1% of miner income from fees132014: 0.4% of miner income from fees142015: 0.6% of miner income from fees152016: 2.2% of miner income from fees162017: 13.3% of miner income from fees172018: 3.8% of miner income from fees182019: 2.9% of miner income from fees192020: 5.6% of miner income from fees202021: 6.1% of miner income from fees212022: 1.6% of miner income from fees222023: 6.7% of miner income from fees232024: 6.6% of miner income from fees242025: 1.0% of miner income from fees252026: 0.6% of miner income from fees26

Data: Firsthand Bitcoin network data; halving dates from the halving page.

Questions people ask

Who mined the first bitcoin?

Satoshi Nakamoto mined the genesis block on 3 January 2009. Hal Finney, who believed he was the first person besides Satoshi to run the software, mined several early blocks in January 2009 and later recalled mining “block 70-something”.

When did GPU mining start?

Laszlo Hanyecz introduced GPU mining to the community in May 2010. Open-source GPU miners followed in September and October 2010, after which graphics cards quickly replaced processors.

What was the first Bitcoin mining pool?

Slush’s pool, announced on the Bitcoin forum on 27 November 2010, is described as the first publicly known mining pool. It found its first block on 16 December 2010 and still operates under the Braiins name.

Can you still mine bitcoin with a CPU?

Technically yes, in practice no. At today’s difficulty of about 135 trillion, a desktop processor doing about 19 million hashes a second would need on average about 956 million years to find one block.

What is an ASIC miner?

An application-specific integrated circuit: a chip that does only one thing, here double SHA-256 of a block header. The first consumer ASIC miners shipped in January 2013 and made every other kind of mining hardware obsolete within months.

How often does mining difficulty change?

Every 2,016 blocks, about every two weeks. The network compares how long those blocks took with the intended ten minutes each and moves the target up or down to match.

What happens when all 21 million bitcoin are mined?

The block reward halves every 210,000 blocks until it rounds down to zero, around the year 2140. After that, miners are paid only from transaction fees.

Epilogue

The same 80 bytes

Everything changed except the puzzle. A modern mining farm does exactly what Satoshi’s computer did in January 2009: it takes an 80-byte block header, hashes it twice with SHA-256 and checks whether the result is below a target. The difficulty is now about 135 trillion times higher and the machines are about 571,429 times more efficient, but a block from 2026 and the genesis block are checked by the same rule, in the same way, by every node.

You can check that rule yourself: our Proof of History page hashes real block headers in your browser, and the Cryptography Lab shows how SHA-256 works.

Sources

  1. Satoshi Nakamoto, “Bitcoin v0.1 released”, Cryptography Mailing List, 8 Jan 2009
  2. Satoshi Nakamoto, “Re: A few suggestions”, Bitcoin Forum, 12 Dec 2009
  3. Blocks 32,255 and 32,256, read from their headers (Bitview)
  4. Bitcoin Wiki, 2010 timeline
  5. m0mchil, “python OpenCL bitcoin miner”, Bitcoin Forum, 1 Oct 2010
  6. Bitcoin Wiki, Slush Pool (announced 27 Nov 2010)
  7. Bitcoin Wiki, Non-specialized hardware comparison
  8. Bitcoin Wiki, Open Source FPGA Bitcoin Miner (released 20 May 2011)
  9. Hackaday, “FPGA Bitcoin miner is probably the most power efficient”, 22 Aug 2011
  10. Bitcoin Wiki, List of Bitcoin mining ASICs
  11. Bitcoin Magazine, “Avalon ships Bitcoin’s first consumer ASICs”, 23 Jan 2013
  12. BIP 22: getblocktemplate (Luke Dashjr, 2012)
  13. Bitcoin Wiki, Stratum mining protocol
  14. Bitcoin Magazine, “With Stratum V2, Braiins plans big overhaul in pooled Bitcoin mining”, 5 Aug 2019
  15. Hashrate Index, Antminer S9 specifications
  16. Hashrate Index, Antminer S21 unveiled at the World Digital Mining Summit 2023
  17. Reuters via Malay Mail, “China vows to crack down on bitcoin mining, trading activities”, 22 May 2021
  18. Bitcoin Core 0.13.0 release notes (internal miner removed)
  19. Firsthand Bitcoin network data, copied daily to this site
  20. Hal Finney, “Bitcoin and me”, Bitcoin Forum, 19 Mar 2013
  21. Bitcoin Magazine, “The man behind Bitcoin Pizza Day is more than a meme: he’s a mining pioneer”, 22 May 2019
  22. Bitview (bitview.space), circulating supply, CC BY 4.0, copied daily to this site

Quotes are short and verbatim from the linked originals; Satoshi’s line about GPUs and incentives is quoted from Bitcoin Magazine’s account of his emails. Device figures are the published specifications of typical hardware. Network figures are daily estimates from Firsthand Bitcoin. Nothing here is financial advice or a mining profitability estimate.

Cite this page

Hashrate and difficulty series are Firsthand Bitcoin's (firsthandbitcoin.com); their reuse terms are not confirmed, so we do not relicense them. Who mined each block: The Pool Ledger (CC BY 4.0).

Bitcoin Years. (2026). Bitcoin Mining History: From CPUs to ASICs and Industrial Farms. https://bitcoinyears.com/bitcoin-mining-history/

All our data sets, licences and downloads: Open data.