Bitcoin Gold: One GPU, One Vote

The 2017 fork that tried to take mining back from the ASIC factories, got robbed three times for its trouble, and now exists twice

When Kurt Wuckert Jr and I spent three hours and forty minutes on the history of Bitcoin’s forks, Bitcoin Gold came up once, in half a sentence, as a chain on a family-tree chart that “was not so prominent.” That is roughly its standing across the whole industry. Bitcoin Cash has a constituency and Bitcoin SV has a villain; Bitcoin Gold has an anniversary.

Yet it was, for a few weeks in late 2017, the fifth largest cryptocurrency in the world, and it remains the first fork in this issue that changed the thing Bitcoin’s other schisms never touched: the proof of work itself. If it weren’t for Luke-jr’s experiment with Blake2b, Bitcoin Gold would have retained the title of being the only UTXO fork of Bitcoin to ”fire the miners”.

I did not interview anyone from the Bitcoin Gold team for this article, and by the end of it you will understand why that was difficult. The project currently has two websites, two founders’ factions and, depending on whom you believe, either one chain or a chain and a token. What follows is reconstructed from block explorers, the project’s own GitHub, court-of-public-opinion documents on both sides, and the reporting of the three occasions on which someone rented enough graphics cards to rewrite its history.

Vital statistics

Market figures are a snapshot from 17 September 2026. For a coin this thinly traded, treat them as an order of magnitude.

Field	Value
Ticker	BTG
Category	UTXO fork of Bitcoin with a new proof of work; holders at the snapshot received 1 BTG per BTC
Snapshot block	491,406, mined 24 October 2017 at 01:17:35 UTC; 2,343 transactions, just over 1 MB
First BTG block	491,407, mined 12 November 2017 at 13:34:01 UTC, nineteen days after the snapshot
Lineage	BTC (2009) to BTG (2017). No child chains; a disputed “merge back” token since 2025
Proof of work	Equihash (200,9) at launch, the Zcash algorithm; Equihash-BTG (144,5) with the personalization string BgoldPoW, also called Zhash, since block 536,200 on 3 July 2018
Difficulty adjustment	Per block: a DigiShield variant at launch, Zawy’s LWMA since July 2018
Replay protection	Two-way SIGHASH_FORKID with fork ID 79 (Bitcoin Cash uses 0)
Addresses	Base58 prefixes G and A instead of 1 and 3; Bech32 prefix btg; BIP44 coin type 156
SegWit	Kept, the first fork in this issue to do so
Supply	21 million cap on Bitcoin’s schedule; about 19.9 million issued at height 965,361; fourth halving at block 840,000 on 24 April 2024, four days after Bitcoin’s
Endowment	100,000 BTG from the first 8,000 blocks (491,407 to 499,406), mined at minimum difficulty to a 4-of-6 multisig; 40 percent unlocked, 60 percent vested monthly over three years
Market cap	About $8.3 million, roughly 0.00054 percent of Bitcoin’s; CoinMarketCap rank 1,006
Price	About USD 0.47; all-time high between $456 and $540 depending on the aggregator, in the last quarter of 2017
Chain status	Still producing blocks: height 965,361 on 20 September 2026, difficulty about 3,500, node software version 0.17.3 at the sampled tip

Why it split: the politics

Bitcoin Gold belongs to a different genre from the two chains before it in this issue. Bitcoin Cash and Bitcoin SV were civil wars, fought by people who had shared a chain and could no longer share a rulebook. Bitcoin Gold was a product of the autumn of 2017, when the success of the Bitcoin Cash airdrop taught the market that any snapshot of Bitcoin’s ledger could be sold as free money, and a dozen projects rushed to take one. Most of them (Bitcoin Diamond, Super Bitcoin, Bitcoin God, and so on) are footnotes. Bitcoin Gold survived because it attached itself to a genuine grievance.

The grievance was ASICs. By 2017 SHA-256 mining had consolidated into a few manufacturers, chiefly Bitmain, and a few pools, most of them Chinese, and the scaling war had shown that those miners had opinions and votes. The Bitcoin Gold pitch was to swap SHA-256 for Equihash, the memory-intensive algorithm Zcash used, which at the time ran best on consumer graphics cards. The marketed ideal was simple and catchy: “make Bitcoin decentralized again.” The hashtag was #1CPU1VOTE, a phrase lifted from the whitepaper. This idea was not new; Satoshi’s own writings assumed CPU mining, and every altcoin since Litecoin had tried some version of it. What was new was applying it to Bitcoin’s own ledger.

The irony was noted immediately: the founder and public face, Jack Liao, was the chief executive of LightningASIC, a Hong Kong seller of mining hardware. He assembled a board of six co-founders, most of them pseudonymous at launch, the lead developer among them going by h4x3rotab. The code was not public when the snapshot was announced, which prompted Coinbase and Bittrex to publish warnings that they would not support a coin whose replay protection they could not audit.

The official website was knocked offline by a denial of service attack on fork day. Within a month a fake wallet promoted through the project’s own channels had drained about $3.3 million from users who trusted it. And then there was the endowment: 8,000 blocks mined privately at minimum difficulty before the public launch, producing 100,000 BTG for the project’s multisig.

The team called it a post-mine, insisted it was 0.6 percent of supply, and locked most of it on a three-year vesting schedule. Critics called it a premine and never stopped. BitMEX Research, which tracked the fork’s coin flows, concluded the allocation had damaged the project’s integrity relative to Bitcoin Cash, which had none.

None of this stopped the price. BTG traded above USD 450 within weeks of launch on the strength of the free-airdrop mania that peaked that December. It has never been within a factor of ten of that price since 2018.

What changed at the fork: the technical delta

Bitcoin Gold is the cleanest example in this issue of how to fork Bitcoin without hurting anyone by accident, which makes the security disaster that followed all the more instructive.

1. A new proof of work. Equihash (200,9) replaced SHA-256. The block header was reformatted to match Zcash’s, with a 256-bit nonce, a height field and a solution field, so that existing Zcash mining software worked out of the box. Bitcoin ASICs were useless on the new chain; Bitcoin’s hashrate could not follow, defend or attack it.
2. Full replay protection. Every post-fork transaction signs with SIGHASH_FORKID and fork ID 79, so a BTG transaction is invalid on Bitcoin and vice versa.
3. Unique addresses. BTG addresses begin with G or A, and the Bech32 prefix is btg. Bitcoin Cash did not fix its address confusion until CashAddr in January 2018; Bitcoin SV never did.
4. Per-block difficulty adjustment. Bitcoin retargets every 2,016 blocks. A minority chain with a rentable algorithm cannot afford that, so BTG adjusts every block. This decision saved the chain from stalling and, as we will see, did nothing to save it from anything else.
5. SegWit kept. Where the big blockers rejected SegWit as the reason for their departure, Bitcoin Gold kept it and promised Lightning compatibility. The team announced a Lightning payment on its testnet in 2019. I could find no evidence of a functioning mainnet Lightning network on BTG at any point.

Everything else, block size and subsidy and halving schedule included, was Bitcoin’s.

The roadmap since

Bitcoin Gold’s history is not written in upgrades. It is written in attacks and delistings, with one upgrade in the middle that was a response to the first attack.

Date	What happened	Nature
24 Oct 2017	Snapshot at Bitcoin block 491,406; website hit by denial of service	Fork
12 Nov 2017	Public mining begins at block 491,407 after the 8,000-block endowment window	Launch
Nov 2017	Fake wallet promoted through official channels steals about $3.3 million	Incident
18 to 22 May 2018	Majority-hash attack with rented Equihash power; about 388,000 BTG (roughly $18 million) double-spent against Bitinka, Bithumb, Bittrex, Bitfinex, Binance and HitBTC	Attack
3 Jul 2018	Hard fork at block 536,200: Equihash-BTG (144,5) with BgoldPoW, LWMA difficulty, difficulty cut to one hundredth at activation	Consensus
14 Sep 2018	Bittrex delists after the team refuses to pay 12,372 BTG (later 6,000) in compensation	Market
Nov 2018	Jack Liao leaves the board for an advisory role; lead developer h4x3rotab reveals himself as Hang Yin	Governance
23 to 24 Jan 2020	Two reorganizations of 14 and 15 blocks double-spend 1,900 and about 5,267 BTG (roughly $72,000); cost estimated at 0.2 BTC per reorg via NiceHash	Attack
1 to 10 Jul 2020	Attacker mines a 1,300-block private chain; developers circulate a checkpoint at block 640,650 in v0.17.2 on 2 July and the chain is rejected when released	Attack thwarted
24 Oct 2022	Binance delists BTG, five years to the day after the snapshot	Market
2023	Bitfinex delists	Market
24 Apr 2024	Fourth halving at block 840,000	Schedule
15 Jul 2024	The bitcoingold.org domain changes hands; the development team calls it theft and moves to btgofficial.org	Governance
22 Jan 2025	Liao’s faction declares the “merge back to Bitcoin” and begins swapping exchange balances into a Bitcoin-native token	Disputed
23 Jan 2025	Upbit, by then the dominant venue, delists; price falls about 60 percent in the preceding month	Market

The price of ASIC resistance

Here is the argument Bitcoin Gold made, in its own words, in 2018: Zhash “uses more memory than an ASIC can muster, but runs fine on many graphics cards.” Here is what that meant in practice. If the hardware that mines your chain is the same hardware that mines a dozen other chains and sits idle in gaming PCs the world over, then anyone can rent it by the hour. NiceHash exists precisely to let them. And a chain worth a few hundred million dollars with a few gigahashes of Equihash securing it is a chain whose entire hashrate can be rented for less than the value of one exchange deposit.

That is what happened in May 2018. Between 15 and 18 May the network’s hashrate roughly tripled as an unknown party brought rented power to bear, mined a private chain, deposited BTG at exchanges on the public one, withdrew other assets, then released the private chain and erased the deposits. The team published the attacker’s address and begged exchanges to raise their confirmation requirements.

By the time the attacker’s wallet went quiet it held 388,201 BTG, about $18 million at the time. Six weeks later the team hard-forked to Equihash (144,5), a parameter set that needs about 2.5 GB of memory per mining thread, chosen to break compatibility with Bitmain’s newly announced Equihash ASIC and with the shared (200,9) rental market. It did the second, briefly. It could not change the underlying economics.

The Bittrex episode that followed is worth dwelling on because it asked a question no fork had faced before: who pays when a blockchain is attacked? Bittrex had absorbed 12,372 BTG of the losses and told the team that “taking responsibility for your chain” meant reimbursing it, later offering to eat half.

The board refused, arguing that it was “not empowered to make a disbursement from the Endowment for the purpose of covering a private company’s losses from their own security failures,” offered a loan instead, and was delisted on 14 September 2018. Both positions were coherent. A team that had kept a 100,000 BTG war chest for the project’s benefit was a strange team to plead poverty. An exchange that accepted deposits on a minority chain after six confirmations had, in fact, misjudged its own risk.

In January 2020 it happened again, on a smaller scale and with better forensics. James Lovejoy, then at MIT’s Digital Currency Initiative, documented two reorganizations of 14 and 15 blocks on 23 and 24 January that redirected 1,900 and about 5,267 BTG. Using NiceHash’s published prices for Zhash he estimated each reorganization cost about 0.2 BTC, roughly $1,700, and that the attacker recovered about the same in block rewards. In other words, the attack was free if the double spends failed and profitable if they succeeded.

Binance, the counterparty, had been crediting deposits after six confirmations and allowing withdrawals after twelve; it moved to twenty. Lovejoy’s conclusion was that “a confirmation requirement on the order of tens of blocks for BTG is still far too few.” After the Upbit delisting in 2025 the Liao site itself advertised that exchanges by then required 300 confirmations, about two days, which it offered as a reason to leave the chain.

The third attempt, in July 2020, is the one the team counts as a success and its critics count as the end of the pretense. Developers noticed an attacker mining a private chain from 1 July, circulated a client with a hard-coded checkpoint at block 640,650 to exchanges and pools on 2 July… and when the attacker released a 1,300-block chain on 10 July, the honest network ignored it. The chain was saved by a small group of developers deciding which blocks were valid in advance, which is to say by the thing proof of work was designed to make unnecessary. Bitcoin Gold’s website reported 108 reachable nodes that week.

I want to be fair to the argument here, because it is a real one. The team maintained throughout that the attacks were not bugs, that the code was Bitcoin’s code, and that the security of a minority proof-of-work chain is a matter of economics that exchanges must price. All true.

But the founding claim of Bitcoin Gold was that commodity hardware would make mining more decentralized and therefore safer. What it demonstrated instead is that commodity hardware makes hashrate a rental market, and a rental market has no loyalty to the chain that happens to be using it this afternoon. Bitcoin’s ASICs are ugly, concentrated and useless for anything else. That last property turns out to be a security feature.

How it differs from Bitcoin today

On consensus. Bitcoin Gold shares Bitcoin’s history to block 491,406 and its rulebook almost entirely afterward: same 1 MB base and 4 million weight units, same SegWit, same subsidy, same script. What differs is the algorithm miners must hash, the header they solve it in, the fork ID in every signature, plus the letters at the front of every address. It is, of all the forks in this issue, the one that made the least significant changes to the original design.

On scaling philosophy. There is none to speak of. BTG adopted Bitcoin’s position wholesale: small base blocks, SegWit, and Lightning as the eventual answer. The blocks at the tip of the chain in September 2026 typically carry one transaction, the coinbase.

On security. This is the whole story and it is told above. Bitcoin Gold’s security budget is independent of Bitcoin’s, which protects it from the fate Bitcoin Cash and Bitcoin SV theoretically face, a redirection of a fraction of Bitcoin’s ASICs. It is exposed instead to the entire GPU rental market, and it has been exploited three times for a combined loss in the tens of millions of dollars. Since 2020 it has lived with hard-coded checkpoints and 300-confirmation deposit policies, which is to say it is secured by the assumption that nobody bothers to attack.

On decentralization. The team’s 2020 count of 108 reachable nodes is the highest number I could find in the BTG contest. Mining is by definition whoever is renting Zhash at the moment. Governance, since July 2024, is contested between two groups each claiming to be the project.

On adoption. Bitcoin Gold’s last meaningful market was South Korea. In December 2024 Upbit alone was reporting daily BTG volume of about $147 million against a market cap of $226 million, a ratio that tells you the coin was a trading instrument for Korean retail and nothing else. When Upbit flagged it for “transparency of operations and sustainability of the project” and delisted it a month later, the price fell from about $15 to, eventually, under fifty cents. CoinGecko stopped updating its BTG feed in July 2026. When the aggregators stop bothering, the market has spoken.

The two Bitcoin Golds

Which brings us to the strangest chapter, and the reason this article cannot end with a clean verdict.

For six years after Jack Liao left the board in November 2018, Bitcoin Gold was run by the remaining co-founders: Hang Yin as lead developer, Martin Kuvandzhiev on operations, Franco Niebles on design, with Edward Iskra as communications director and the author of most of the project’s public statements. Development slowed to a crawl. The last node release with new features, v0.17.3, shipped in August 2020; a v0.21.3 beta appeared in December 2024 and the explorer I sampled still reports 0.17.3 at the tip. The last roadmap was published in 2019. Hang Yin, for his part, co-founded Phala Network, a privacy-computing project on Polkadot, and announced a partnership between it and BTG in 2021.

Then, according to the BTCGPU wiki and the project’s X account, on 15 July 2024 Liao “stole the domain bitcoingold.org and redirected DNS to his own account.” Every service behind that name went dark for the existing team, which moved to btgofficial.org. Liao’s version, published on a Medium account with leaked WeChat logs, is that after Binance delisted the coin he tried to push a dormant team to do something, anything, and when they would not, he reclaimed the project he had founded. Both accounts agree on the facts: one man took the website, the rest kept the code.

What Liao did with the website is the part I cannot fully verify because the website appears to be down. So I will present Liao’s claims. From April 2024 his site describes a mission to “merge back to Bitcoin.” On 22 January 2025, the day before Upbit’s delisting took effect, it announced a “strategic upgrade” after which balances of the old coin, which it labels BTGO, on cooperating exchanges would be swapped one for one into a new BTG, apparently a Bitcoin-native token of the Ordinals or Runes variety, since the site lists it on Magic Eden, UniSat and OKX’s Bitcoin marketplaces. The pitch: the same security as Bitcoin, “zero inflation,” a stablecoin called USDG with a 70 percent collateral ratio, and a future for GPU miners doing “AI reasoning.” At least one Indian exchange delisted BTG in early 2025 citing a token migration whose plans “were later updated and not supported by offshore crypto exchanges.” Meanwhile the Equihash chain at btgofficial.org continues to produce a block every ten minutes… most of them empty, as if it switched places with Blockstream’s Liquid federation.

So there are two Bitcoin Golds. One is a proof-of-work chain with nine years of history, three attacks, a dozen active volunteers and a market cap you could raise at a small angel round. The other is a token issued on Bitcoin by the man who founded the first one, claiming its name, its ticker and its holders. Neither can stop the other. I can tell you which one the block explorer follows. I cannot tell you which one the aggregators are pricing at $0.47… and I suspect they cannot either.

Who is who

The founders

• Jack Liao, chief executive of LightningASIC and the “first founder” who recruited the other five. Left the board for an advisory role in November 2018; returned in 2024 by taking the domain and launching the merge-back campaign. Posts as @realSatoshiGold. Listed as “Co-Founder, Inactive” on the team page of the site he does not control.
• Hang Yin (h4x3rotab), lead developer, co-founder and board member. Fudan computer science graduate and former Google engineer who de-anonymized himself in November 2018. Later co-founded Phala Network.
• Martin Kuvandzhiev, operations lead, co-founder and board member, Bulgaria.
• Franco Niebles, lead designer, co-founder and board member, Colombia.
• Alejandro Regojo and Robert Kuhne, co-founders listed as inactive.
• Edward Iskra, communications director and board member; author of the “Responding to Attacks” and Bittrex statements in 2018 and of the July 2024 domain notice on the wiki. The nearest thing the project has to an institutional memory.

The other side of the ledger

• The attackers, never identified. The May 2018 wallet ended with 388,201 BTG; the January 2020 blocks were mined to a single address beginning GWrW5d.
• James Lovejoy, then of MIT’s Digital Currency Initiative and lead maintainer of Vertcoin, whose reorganization tracker documented the 2020 attacks and priced them at 0.2 BTC each.
• Bittrex, the first exchange to ask a fork’s developers to pay for its losses, and also the first business to delist BTG.
• Upbit, the exchange whose January 2025 delisting removed the last liquid market.
• NiceHash, the hashrate marketplace named by the team in 2018 and 2020 as the likely source of the attacking power; it has never been accused of wrongdoing, only of existing.

What Bitcoin Gold proved

• You can fork Bitcoin’s proof of work cleanly. The header format, the fork ID, the address prefixes and the per-block retarget all worked exactly as designed. Nobody lost coins to replay or to sending BTG to a BTC address. That is a better technical implementation than Bitcoin SV’s.
• ASIC resistance is rental exposure. Three attacks in 26 months, two of them profitable, established the economics that every small proof-of-work chain now lives with: if your hashrate can be rented, your confirmation depth is a negotiation.
• Checkpoints are the price of survival for a minority chain. The July 2020 defense worked, and it worked by developers deciding in advance which chain was honest. Bitcoin Gold was the first Bitcoin fork to do this openly; it will not be the last.
• Exchanges will eventually ask who pays. The Bittrex demand was unprecedented in 2018. The question has not gone away.
• A brand can outlive its chain. The 2024 domain dispute and the 2025 “merge” show that a fork’s name, ticker and holder list have value independent of the blocks, and that whoever holds the website holds the narrative.
• Hold your forks anyway. Anyone who held Bitcoin on 24 October 2017 still holds Bitcoin Gold, on both of its chains, and it cost them nothing. As Kurt said in the previous article, that entitlement to every branch of the tree is the one piece of fork-era advice that has aged well.

Will it flip Bitcoin?

At 0.00054 percent of Bitcoin’s market value, a hashrate that can be rented for the price of a used car, a founding team split across two websites, and its last major exchange gone, Bitcoin Gold is not going to flip anything, and I am not aware of anyone who claims it will. The interesting question is the other one. Bitcoin Gold set out to answer whether Bitcoin’s ledger could be made safer by making it minable on ordinary hardware. Nine years of data say no, and they say it more conclusively than any argument ever did. The most valuable thing the chain has produced is that answer.

Scorecard inputs

Criterion	Evidence
Technical ambition	Proof-of-work swap with a Zcash-compatible header; two-way replay protection; unique addresses; per-block difficulty; one further PoW hard fork in 2018; SegWit retained. Modest by design and executed cleanly
Political radicalism	Anti-ASIC mining politics inside 2017 fork mania; a 100,000 BTG endowment; a founder who sold ASICs; a 2024 domain seizure and 2025 token migration by that founder against the remaining team
Adoption	Briefly top five by market cap in 2017; delisted by Bittrex in 2018, Binance in 2022, Bitfinex in 2023, Upbit in 2025; market cap 0.00054 percent of BTC; CoinGecko feed stale since July 2026
Scalability claims vs measurement	Claims: Bitcoin’s roadmap with fairer mining. Measured: near-empty blocks at the tip; no mainnet Lightning network found
Decentralization	108 reachable nodes in 2020; mining is whoever rents Zhash today; governance contested between two factions since 2024
Security model	Independent Equihash budget, immune to BTC ASIC redirection but exposed to GPU rental; three majority-hash incidents (May 2018, Jan 2020, Jul 2020); hard-coded checkpoints since 2020; exchanges requiring up to 300 confirmations
Long-run viability	Chain still producing blocks at height 965,361; last feature release 2020, beta in Dec 2024; no roadmap since 2019; brand split
Speculative context	Upside: none articulated by anyone outside the two factions. Downside: the coin may already exist in two incompatible forms

Sources

mempool.space API for Bitcoin block 491,406; BTCGPU README and Technical Spec wiki (edited 3 February 2025); BTCGPU chainparams.cpp; btgexplorer.com Blockbook status, 20 September 2026; btgofficial.org team page; Bitcoin Gold project posts “Responding to Attacks,” “A Successful Network Upgrade,” “Statement Regarding Bittrex” and “Leadership for Another Year” (2018); James Lovejoy’s gist “Bitcoin Gold (BTG) was 51% attacked” (January 2020); CoinDesk, 24 October 2017, 27 January 2020 and 11 July 2020; BitMEX Research, “Bitcoin Gold investment flow data”; ZDNet and The Next Web on the Bittrex delisting, September 2018; Binance delisting notice, October 2022; Bitfinex support notice on the BTG delisting; Upbit notice of 24 December 2024 and coverage by crypto.news and Cryptonews; bitcoingold.org “BTG Roadmap 2025” and “Merge back Bitcoin” pages and the bitcoingoldorg/mergebackbitcoin repository (Liao faction); CoinMarketCap data API for BTG and BTC, 17 September 2026; Bitcoin Takeover Podcast S17 E39.