Alephium (ALPH): a complete guide to the coin and its mining

Alephium is a first layer proof of work network whose mainnet went live on 8 November 2021. The project builds a sharded blockchain on the stateful UTXO model, sUTXO for short: funds are tracked the Bitcoin way, through unspent outputs, with smart contract state added on top. Development is run by a small team led by Cheng Wang, and by the figures the project publishes itself, roughly 5 million dollars of raised funding has been spent over five years.

The network is not one chain but 16 parallel chains split across 4 address groups. The mechanism is called BlockFlow and it lets transactions be processed inside a group and between groups without a shared bottleneck. Smart contracts run on a purpose built virtual machine, Alphred, and the contract language is called Ralph. There is no EVM compatibility and none was ever planned.

The single most important point for anyone reading reference articles: the Danube fork of 15 July 2025 cut block time from 16 to 8 seconds and removed the cap of 1 billion ALPH entirely, replacing it with endless tail emission. The official tokenomics documentation puts it plainly: emission is no longer limited and supply is infinite. Most aggregators and reviews still publish a maximum supply of 1 billion, and that number is out of date.

This page is written as a reference for hardware owners: network parameters, the mechanics of the two emission curves, a ranking of Blake3 ASICs from our catalogue with a break even calculation by electricity price, pools, wallets, the Decred trap and a separate walk through the figures that large sources still serve in an outdated form.

• 2021Mainnet launches on 8 November. The network starts out sharded: 4 groups, 16 chains, block time 64 seconds. Genesis issues 140 million ALPH: 80 million for sales, 30 million for the ecosystem, 30 million for the team and the treasury, all of it with vesting from two to four years. Mining runs on graphics cards from day one.
• 2022A year of GPU mining and first listings. The official CUDA miner and open third party clients give access to any owner of an Nvidia or AMD card. The first pools appear, the network stays niche, and hashrate is measured in tens of terahashes.
• 2023The Leman fork on 30 March. This is the first major upgrade: the smart contract stack is completed and dApps can now be written in Ralph. The same release raises the burned share of fees from 50 to 100 percent: the miner gets no fee at all, it is destroyed.
• 2024The all time high price of 3.86 dollars is set on 27 February. Speculative capital rides the wave of interest in DeFi on Alephium, and hardware makers start paying attention along with it.
• 2024First half of the year: ASICs arrive. The IceRiver line runs from the compact AL0 to the industrial AL3, joined by the desktop Goldshell AL-BOX units. Hashrate grows several times over and graphics cards are pushed out of mining within a few months.
• 2024The Rhone fork on 12 June. Block time is cut from 64 to 16 seconds and virtual machine performance is reworked. The block reward is rescaled proportionally so that daily emission stays unchanged.
• 2024July brings the Antminer AL1 at 15.6 TH/s and 3,510 W, which works out to 225 J/TH. The arrival of Bitmain finally closes the question of GPU competition and sets the efficiency ceiling that still stands today.
• 2024Autumn: official support for Ledger and OneKey hardware wallets. October brings what is still the latest release of the open third party firmware build for IceRiver machines, version 0.8.4 dated 1 October.
• 2025The Danube fork on 15 July, the most important upgrade in the history of the network. Block time falls from 16 to 8 seconds, the cap of 1 billion ALPH is lifted and tail emission introduced, groupless addresses appear along with PassKey authentication, contract call chains and the new Sync V2 synchronisation protocol.
• 2026February opens the public testnet of Powfi, an application built by the project itself: a concentrated liquidity exchange plus staking through wrapped xALPH. The declared split of trading fees is half towards buying back and burning ALPH, half to stakers. At the end of July 2026 the Powfi mainnet is still not live.
• 2026The bridge hack on 30 May. The attacker deployed a contract that forged messages for the bridge validators and drained around 815 thousand dollars of collateral on Ethereum and BSC, minting 13.76 million unbacked wrapped ALPH along the way. The Alephium network itself and its consensus were not affected, the bridge was the vulnerable part.
• 2026In early June a contract upgrade burned 96.4 percent of the unbacked wrapped tokens, and the project published a detailed transaction level account of the incident. About half a million tokens had already reached the secondary market by then.
• 2026State of play at the end of July: hashrate around 5.4 PH/s, price around 0.038 dollars, market capitalisation of the order of 5 million dollars. From the February 2024 peak the coin has lost roughly 99 percent of its value, and mining at European electricity tariffs runs at a loss.

Alephium inherits the Bitcoin accounting model and then extends it. In plain UTXO, coins sit in unspent outputs and a transaction spends some outputs while creating others. Alephium adds state to that: a contract lives in separate storage, and a call to it remains part of the same UTXO transaction. Hence the name stateful UTXO. The advantage is that parallel transactions which do not overlap on inputs never conflict and need no global ordering, unlike an account model.

The second layer of the idea is BlockFlow sharding. The address space is split into 4 groups. Between every pair of groups, including a group paired with itself, there is a chain of its own: 4 by 4, that is 16 chains. Each chain advances its own blocks independently, and dependencies between chains are recorded as references in the header. The miner picks which chain to work on, and the pool distributes the jobs by itself.

A transaction inside one group is confirmed by a single chain. A transaction between groups touches two chains and requires both to reach agreement, which is why cross group transfers confirm more slowly than intra group ones. Before the Danube fork a user had to keep track of the group of their own address by hand, and that was the main usability complaint. Danube introduced groupless addresses, and for the end user sharding became invisible.

Smart contracts run on the Alphred virtual machine, designed specifically for sUTXO. The Ralph language is deliberately restricted: it lacks a number of constructs behind the routine exploits seen in the EVM ecosystem, and ownership of assets inside a contract is stated explicitly. The price of that is total incompatibility with the Ethereum toolchain: a Solidity contract cannot be ported directly, it has to be rewritten.

1. The wallet builds a transactionSpends of unspent outputs are signed on the secp256k1 curve, the same one Bitcoin uses. Address hashing goes through Blake2b at a length of 32 bytes, which is why Alephium addresses are noticeably longer than Bitcoin ones.
2. The transaction reaches the right chainThe sender group and the recipient group decide which of the 16 chains accepts the transaction. An intra group transfer goes into its own chain, a cross group one engages two.
3. The node assembles a block templateThe header takes in the 24 byte nonce, the version, references to dependent blocks in other chains, the state hash, the transaction list hash, a timestamp and the difficulty target. The nonce comes first in the serialisation, ahead of every other field.
4. The ASIC iterates over noncesThe machine serialises the header and runs it through Blake3 twice until the result falls below the target value. The double pass and the position of the nonce are the details that make Alephium hardware incompatible with other networks built on Blake3.
5. The block spreads across the networkA block that is found goes into its own chain. Target block time is 8 seconds per chain, so with 16 chains the network delivers roughly 2 blocks per second and 172,800 blocks per day.
6. The reward is credited, the fee burnsThe miner receives only the emission part, about 0.1433 ALPH at the end of July 2026. Transaction fees never reach them: since the Leman fork they are burned 100 percent.
7. Difficulty is retargeted chain by chainThe target adjusts independently for each of the 16 chains in order to hold 8 seconds. The reward, meanwhile, depends not only on timing but also on current hashrate: two curves operate at once.

Alephium computes proof of work on Blake3, but not in the shape other networks use the function in. Checked against the protocol source: the PoW hash is a double Blake3 over the serialised block header. There is no single pass here, and that is fixed in the implementation rather than in the documentation.

The second peculiarity is the 24 byte nonce. In the protocol code the nonce length is a constant equal to 24, and the nonce itself sits first in the header structure, ahead of the version, the dependency references, the hashes and the timestamp. In other words, the part being iterated is at the start of the message and not in the middle or at the end, as it is in most PoW networks. For an ASIC designer that matters fundamentally: the pipeline is built around one specific byte layout.

Why Blake3 gave way to ASICs so quickly. The function was designed from the outset to be very fast and easy to parallelise, with a compact state and simple operations: additions, rotations, exclusive or. Nothing in it stands in the way of a dedicated circuit. There is no memory hardness of the Ethash or RandomX kind and no heavy internal structures. That is why it took roughly two and a half years from network launch to a production ASIC, and graphics cards dropped out of mining in the single season of 2024.

A separate word on the Decred trap. Decred also moved to BLAKE3 for proof of work, and on that basis some hardware catalogues and calculators put both coins in one list, and occasionally offer Alephium machines outright for mining DCR. That is a mistake. In Decred, per the DCP-0011 specification, the PoW hash is a single BLAKE3 over a serialised header 180 bytes long, and the nonce sits inside a 36 byte extra data field rather than at the start. On top of that, the Decred block hash itself is still computed with BLAKE-256, and BLAKE3 is applied only to proof of work. The schemes differ at pipeline level, so an Alephium ASIC does not mine Decred, and Decred machines do not mine Alephium.

The practical conclusion: when you buy a Blake3 ASIC, you buy hardware for one specific coin. Alephium has no merged mining and no sibling networks on the same hashing scheme to switch to when the price falls. That makes the risk concentrated: the economics of your fleet depend entirely on one network and one exchange rate.

Full description of the Blake3 algorithm →

Genesis issued 140 million ALPH: 80 million for sales with vesting from two to four years, 30 million for ecosystem development with four year vesting, 30 million for the team and the treasury with three year vesting. Everything else enters circulation through mining alone.

Alephium has no classic halving. The block reward is computed as the minimum of two values: a time curve and a hashrate curve. The time curve lowers the reward smoothly from 15 to 5 ALPH over the first four years, then holds it and, on a horizon of decades, drives it towards zero. The hashrate curve ties the reward to the current power of the network: the smaller the hashrate, the lower the payout. Because of that second curve, a weak network loses reward with no schedule at all, which is exactly what we are seeing today.

The Danube fork changed the design fundamentally. Emission used to be laid out over 81 years and ran into a cap of 1 billion ALPH. Danube removed that cap: the schedule of the first 81 years is preserved unchanged, but after it mining does not stop and instead moves into endless tail emission along the lines of Monero. The point is to leave miners a permanent incentive to defend the network, even when fees are burned and never reach them. Maximum supply is now correctly recorded as not limited, and CoinMarketCap already displays it that way.

The second economic loop is burning. Since the Leman fork of March 2023, 100 percent of transaction fees are destroyed, whereas half of them were burned before that. The miner receives emission and nothing else. With network activity low the burned volume is small and does not offset tail emission, so there is no ground for calling the coin deflationary today.

The third loop, Proof of Less Work, is not running yet. The mechanism assumes that once network power is high enough, part of the miner reward starts being destroyed, so security gets paid for in coin and not in electricity alone. The activation threshold is 1 EH/s. At the end of July 2026 the network delivers around 0.0054 EH/s, less than one percent of the threshold, so PoLW burning is not active and the network runs as ordinary PoW. Claims in reference articles that Alephium already saves energy thanks to PoLW are factually wrong today.

Our catalogue carries 6 Blake3 models from three manufacturers, and every one of them mines Alephium and nothing else.

Browse miners in the catalogue →

The ranking is built on real efficiency from the catalogue, from the best watts per terahash ratio to the worst, and not on how popular a model is.

The main point first: we have no firmware for Blake3. The algorithm is not on the support list of HashCore Toolkit, and AsicBoost builds are released for Bitmain machines on SHA-256 and Scrypt, that is for the Antminer S, T and L series. Installing them on an Antminer AL1 or on any IceRiver or Goldshell machine is impossible, because the platform and the firmware image of those devices are different.

The only third party firmware that exists on Blake3 today is the open build for IceRiver machines that an independent developer publishes in a public repository. It offers a frequency offset over a wide range, manual voltage control applied without a reboot, automatic fan regulation by target chip temperature, extended metric charts, a dark theme and token based API access over HTTPS. There is also separate pool health monitoring with an automatic switch back to the primary pool once it is reachable again.

The developer fee in that build is 1 percent. A commercial variant for hosting sites exists with a fee of 1.33 percent, adding multi user management, hashrate splitting across several addresses and custom interface branding. The last release, version 0.8.4, came out on 1 October 2024, and no new versions have appeared since. That is worth keeping in mind: the project is effectively frozen, and fixes for new board revisions should not be counted on.

For the Antminer AL1 and for the whole Goldshell AL-BOX series, no third party firmware could be found in the public domain. Only the factory firmware of the manufacturer is available. If somebody offers you an overclocked build for these models, ask for a link to a public repository and check it yourself: on the Blake3 market this is the classic way a malicious image that swaps the payout address gets distributed.

A general rule for any third party firmware: it voids the warranty, requires a backup of the factory image before installation, and requires the machine to be kept off the open internet. Installing firmware taken from a forum, from a private message or from an archive with no public version history is never acceptable, under any circumstances.

Getting started is simple. You need an Alephium wallet and its address, a chosen pool, the machine itself and power. In the web interface of the ASIC you enter the pool address, the worker field usually holds your ALPH address separated by a dot from the device name, and the password is mostly arbitrary. Solo mining is technically possible through a full node, but at the current difficulty a single machine would search for a block for an unacceptably long time, so only pool mining makes practical sense.

What you need physically. An industrial machine of the Antminer AL1 or IceRiver AL3 class draws about 3.5 kW, which means a dedicated circuit and proper wiring, since a household socket will not hold that load. Noise of around 75 dB is comparable to a working vacuum cleaner, so placing it in a living space is out of the question. The desktop Goldshell AL-BOX and IceRiver AL0 units fit within 100 to 480 W and live on an ordinary socket, though even they are louder than household appliances.

About heat. Every kilowatt consumed turns into a kilowatt of heat. A machine at 3.5 kW heats a room like three good radiators, and removing the hot air matters more than it seems at the start: without an organised exhaust loop the room warms up, the ASIC throttles and hashrate drops. In winter that heat can be partly counted as a benefit, in summer it becomes a pure cost.

Home against an industrial site. On end of July 2026 metrics the difference is fundamental: at an industrial tariff below 3 cents per kilowatt hour the machine breaks even, while at a European household tariff it is guaranteed to lose money. Home mining of Alephium today makes sense as learning, as supporting the network out of conviction, or as heating, but not as a source of income. Writing anything else would mislead the reader.

Miner income is made up of four quantities: the share of your hashrate in the network, daily emission, the ALPH price and the pool fee. Transaction fees do not enter the calculation at all, because they are burned in full. The formula is transparent: the network produces 172,800 blocks a day, each carrying a reward of about 0.1433 ALPH at the end of July 2026, and your share is proportional to your hashrate.

The key feature of Alephium is that the reward depends on network hashrate. In an ordinary PoW network a drop in total power raises your share and partly offsets the losses. Here two curves work at once, and when hashrate falls the second one kicks in: the block reward drops along with the power. That weakens the usual stabilising mechanism and makes the downward spiral steeper than in Bitcoin or Kaspa.

The second factor is the price. From the high of 3.86 dollars in February 2024 the coin has lost roughly 99 percent, and the entire economics of a fleet today is set by that. Daily turnover on exchanges is measured in hundreds of thousands of dollars against a market capitalisation of about 5 million, so selling even a modest amount of mined coin moves the price: real proceeds come out below the calculated figure because of slippage.

What to calculate before buying. Take the consumption of the machine in kilowatts, multiply by 24 and you have the daily draw. Divide daily revenue in dollars by that draw and you get your own break even electricity price. For the Antminer AL1 it is about 0.032 dollars per kilowatt hour, for the desktop Goldshell units about 0.015 dollars. If your tariff is higher, the machine runs at a loss, and no firmware setting will turn that around: tuning shifts the numbers by percentages, while the gap here is a multiple.

Open the profitability calculator →

A practical note on address groups. Before the Danube fork a user had to remember which of the 4 groups their address belonged to, and that was a frequent source of mistakes when withdrawing from an exchange and when setting up pool payouts. Groupless addresses, introduced in July 2025, close that problem, but old addresses keep working and some venues and tools were slow to update. Before your first large transfer, check that your pool and your exchange handle the chosen format correctly.

Keep a separate address for mining, not the one holding your savings. The pool payout address is visible in the ASIC configuration, and the machine sits on a network and is potentially exposed to attack, especially if it runs third party firmware. Move what you accumulate to a hardware wallet from time to time.

This section is informational, buying the coin is not a service we provide. At the end of July 2026 most ALPH turnover sits on a handful of mid tier centralised venues: MEXC and Bitget together account for roughly 90 percent of volume, with noticeably less on Gate and CoinEx. There are no direct pairs against the euro or other fiat currencies, and trading goes through USDT, less often through BTC and ETH.

Liquidity has to be called what it is: thin. Daily turnover across all venues stays in the range of 140 to 170 thousand dollars against a market capitalisation of around 5 million. That means even selling a few thousand dollars into the book moves the price noticeably, and a market order for a month of output from an average farm will produce visible slippage. Plan your exit with limit orders and in parts.

There is a decentralised route as well. Wrapped ALPH trades on Uniswap version 3 in pairs with ETH and USDT, and the network itself hosts its own exchange ecosystem. Here it is worth remembering the incident of 30 May 2026: it was precisely the bridge between Alephium and outside networks that became the point of an attack worth 815 thousand dollars. The network itself was not affected, but wrapped versions of the token deserve to be treated as a separate risk.

General safety rules. Do not keep the coin on an exchange longer than the trade requires. Check that the venue has not suspended ALPH deposits and withdrawals: such suspensions happen regularly with small coins, and one exchange has already halted trading. Before your first withdrawal, make a test transfer of a minimal amount and wait for all confirmations, which usually number from 30 to 110.

The stated purpose of Alephium is to be a platform for decentralised finance on proof of work. The idea is to combine the Bitcoin model of security and accounting with full smart contracts without moving to proof of stake. The Alphred virtual machine and the Ralph language are designed so that ownership of assets inside a contract is stated explicitly and the routine exploit patterns of the EVM world become syntactically inexpressible.

What works today. By project figures from May 2026 the mainnet carries about 28 native applications: exchanges, lending protocols, NFT marketplaces, games and social services. Roughly 146 million dollars have passed through the AlphBridge to Ethereum and BSC over the whole period. The audience is estimated at more than 70 thousand wallet users. Compared with large networks that is very little, and the honest word for this ecosystem is niche.

The main bet of the project for 2026 is Powfi, an application of its own opened on testnet on 19 February 2026. It combines a concentrated liquidity exchange with staking through wrapped xALPH. The declared mechanics: half of trading fees go towards buying back and permanently burning ALPH, half is distributed to stakers. The percentages are announced as preliminary and may change before release. At the end of July 2026 Powfi is still not on mainnet, so judging its effect on the price and on miner economics is premature.

The bridge deserves a separate note. The incident of 30 May 2026 was not a break of consensus but a forgery of messages for the bridge validators: the attacker deployed a contract that emitted fake events and drained around 815 thousand dollars of collateral on Ethereum and BSC. The 13.76 million unbacked wrapped tokens minted in the process were 96.4 percent burned by a contract upgrade in early June. It is an important lesson that a bridge and a network are different levels of trust.

On a 51 percent attack. At a hashrate of around 5.4 PH/s the cost of renting comparable power is low in absolute terms, but there is nowhere to rent it: Blake3 in the Alephium variant is not traded on hashrate markets, and all existing hardware consists of narrowly specialised ASICs held by a known circle of owners. An attacker would have to buy machines physically, which makes no economic sense when the network is capitalised at about 5 million dollars: the cost of the attack would exceed any gain from it. Sharding is an extra barrier, since what has to be attacked is not one chain but a coordinated structure of 16.

On mining decentralisation. Trackers identify only about 30 percent of network hashrate, spread between HeroMiners, Kryptex, K1Pool and several small pools, while the rest is opaque. Share estimates from different sources contradict each other so strongly that one tracker puts HeroMiners first with 16 percent while another gives F2Pool half the network. That opacity is itself a risk factor: the real concentration of power cannot be judged from outside.

On nodes and resilience. Alephium publishes no official public statistics on the number of full nodes, and we will not substitute somebody else figure here. What is known is different: the network has run without interruption since 8 November 2021, which is a non trivial result for a sharded architecture. The Danube fork of July 2025 additionally reworked the synchronisation protocol by introducing parallel block download, which lowered the time needed to bring a node up for the first time.

On incidents. The only major incident is the bridge hack of 30 May 2026, worth around 815 thousand dollars. Technically it was not a consensus failure: the attacker deceived the bridge validators with forged messages from a purpose deployed contract. The network kept running normally, and 96.4 percent of the unbacked wrapped tokens were burned in early June. Even so, the event showed that the perimeter around Alephium is protected less well than the protocol itself.

Maximum supply is 1 billion ALPH. The most widespread outdated figure. The cap was lifted by the Danube fork of 15 July 2025: the official tokenomics documentation states directly that emission is no longer limited and supply is infinite. The schedule of the first 81 years is preserved, but after it mining moves into endless tail emission along the lines of Monero. CoinMarketCap already shows maximum supply as unlimited, while dozens of reference sites and mining blogs still publish 1 billion.

Block time is 16 or 64 seconds. Both values are historical. 64 seconds applied at launch in 2021, and 16 seconds after the Rhone fork of 12 June 2024. Since 15 July 2025 the target block time is 8 seconds, the network delivers about 2 blocks per second and 172,800 blocks per day. Calculators that still count on 16 seconds are out by a factor of two on the block count.

Blake3 in Alephium and in Decred is the same algorithm. It is not. In Alephium the PoW hash is a double Blake3 pass over the serialised header, where the 24 byte nonce comes first. In Decred, per the DCP-0011 specification, it is a single pass over a header 180 bytes long, where the nonce sits inside the extra data field, and the block hash itself is computed with BLAKE-256. The machines are incompatible in both directions. Hardware catalogues that put both coins in one list of supported networks mislead the buyer.

Alephium saves energy thanks to PoLW. The Proof of Less Work mechanism is in the protocol, but it is not activated. The threshold for burning part of the reward is 1 EH/s of network hashrate. At the end of July 2026 the network delivers around 0.0054 EH/s, that is less than one percent of the threshold. Today Alephium runs as ordinary proof of work, and phrasing such as 87 percent more efficient than PoW describes a design goal, not the current state.

Throughput is 20 thousand transactions per second. A marketing figure that made its way into the cards of large aggregators. The official documentation of the project speaks of over 800 transactions per second on the current 16 chains and of a potential above 10 thousand once the shard count is expanded to the maximum of 32. The difference is an order of magnitude.

A classic halving every four years. Alephium has no halving at all. The block reward is the minimum of two curves, one by time and one by hashrate, and it declines smoothly. The practical consequence matters more than the terminology: when miners leave, the reward falls along with hashrate, so the familiar mechanism where departing competitors raise your share works far more weakly here.

The miner receives transaction fees. They do not. Since the Leman fork of March 2023, 100 percent of fees are burned, whereas 50 percent were burned before that. Any profitability calculation that adds fees to the block reward is wrong by construction for Alephium.

Circulating supply is about 122 million ALPH. Aggregator estimates at the end of July 2026 diverge: CoinMarketCap gives about 122 million in circulation against a total supply of about 216 million, while CoinGecko and PoolBay show roughly 135 million. There is no single correct number right now, and any specific figure should be quoted with its source and date.

Third party firmware exists for the Antminer AL1. No public open builds could be found for the Antminer AL1 or for the Goldshell AL-BOX series: only the factory firmware is available. The only third party firmware that really exists on Blake3 is made for IceRiver machines, and its last release is dated 1 October 2024. Offers of overclocked builds for Bitmain and Goldshell in this niche are almost always either a repack of the factory image or malware.

The Alephium network was hacked in 2026. That wording is incorrect. On 30 May 2026 the bridge between Alephium and outside networks was attacked: the intruder deployed a contract that forged messages for the validators and drained around 815 thousand dollars of collateral onto Ethereum and BSC. Consensus, sharding and the network itself worked normally, with no downtime. Of the 13.76 million unbacked wrapped tokens minted, 96.4 percent were burned in early June.

Powfi is already live and pays stakers. At the end of July 2026 Powfi is on a public testnet opened on 19 February 2026. There is no mainnet. The declared fee split, half towards buying back and burning ALPH and half to stakers, is marked by the project itself as preliminary and subject to change before release.

Mining Alephium pays off on a household tariff. It does not pay off at any European tariff. By the calculation of 31 July 2026 the best machine in the catalogue, the Antminer AL1, earns on the order of 2.7 dollars a day while consuming 84.2 kWh, which puts the break even point at about 0.032 dollars per kilowatt hour. In Spain a household kilowatt hour costs several times more, and the loss comes to around 10 euros a day per machine at a tariff of 0.15 euros.