GPU Mining in 2027: Two New Coins Pay Real Money — Every Classic Loses It

GPU mining did not come back. Every classic GPU coin — Ravencoin, Ergo, Ethereum Classic, Kaspa — loses money at $0.10/kWh, and at $0.15/kWh every classic on every card we checked is negative except Conflux, by 3 to 9 cents a day. All of the actual profit sits in two coins that did not exist a year ago: Pearl and Quantus. Both pay real money today. Both carry warnings as big as the numbers.

So the honest 2027 question is not "which coin do I mine?" It is a three-way choice: mine the two new coins while they last, rent the card to AI workloads, or sell it — a used RTX 4090 fetches $2,342–$3,010, roughly 470 to 815 days of mining or rental income, collected today, with no power bill.

Every coin below gets the same two questions: where does the miner's money actually come from, and does it survive the electricity bill? All figures checked October 31, 2026 — a dated snapshot, not a rate.

The ruler we used

The reference card is an RTX 4090, the card most home miners built rigs around. Net is gross revenue minus power: watts × 24 hours × your rate. Pool fees and wear excluded — every net is slightly optimistic.

Electricity gets its own line:

• A 4090 mining Pearl draws ~370W → 8.88 kWh/day → $0.89/day at $0.10/kWh, $1.33/day at $0.15
• A 4090 mining Quantus draws ~300W → $0.72/day at $0.10, $1.08/day at $0.15
• A 4090 on a classic coin draws 249–299W → $0.60–$0.72/day at $0.10, $0.90–$1.08/day at $0.15

Net $/day on an RTX 4090 at $0.10/kWh, checked October 31, 2026 (WhatToMine and hashrate.no; rental scenarios below). Selling the card is not on the chart — it is not a daily number: $2,342–$3,010, once.

Pearl: the only real paycheck, with four warnings attached

PRL is a coin — Pearl is its own Layer-1, a fork of Bitcoin's codebase with the hash puzzle swapped for matrix multiplication, the core arithmetic of AI work (consensus: "Proof of Useful Work"). Mainnet launched April 27, 2026, built by Pearl Research Labs from a 2025 cryptography paper by co-founders Ilan Komargodski and Omri Weinstein. Max supply 2.1 billion PRL; ~364.7 million (17.4%) were circulating on October 31 at $1.45.

The money, first. A 4090 grosses $5.99–$6.42 a day mining Pearl — the range is WhatToMine versus hashrate.no, printed rather than picked. Net of power: +$5.10 to +$5.75 a day at $0.10/kWh, +$4.66 to +$5.41 at $0.15. Nothing else in GPU mining is close.

Where the money comes from: 100% inflation and emissions. Mining is the only source of PRL — no premine, no founder allocation, no investor block, per the project, consistent across the coverage we checked. No users pay fees underneath this; holders pay miners through dilution, ~$1.57 million a day of it. The whitepaper itself concedes that without future AI-compute adoption, "the value of the token and the amount of mining should clearly be zero."

Now the warnings, at full weight:

• The "useful work" is mostly unverified. An independent preprint (June 3, 2026) examined 8,012 workers in one pool — about 21% of network hash — and found the dominant mining software contained no inference or ML code at all. The researcher's own miner, submitting randomly generated matrices, had 44 shares accepted. His estimate: ~320,000 GPU-equivalents producing output nobody requested. The protocol verifies the math happened; it cannot verify anyone wanted the result. One real counter-example: Together AI's Pearl-linked inference endpoint, priced 25%+ below standard — the discount funded by expected PRL emissions.
• "Fair launch" still had insiders. More than one-third of all PRL in existence was mined within the first six days. No premine is not the same as no head start.
• The earnings are already collapsing. A 5090's estimated daily Pearl earnings fell from ~$33.80 to ~$17.19 within weeks of the May 2026 boom, and revenue was down roughly 75–85% from the May peak by late October. The official pool also charges a 20% fee.
• The price just slipped, and the field is not who you think. Aggregators date the all-time high to mid-to-late October 2026 at $2.04–$2.22; by October 31 PRL was $1.45 — down roughly a third in about eleven days. Your competition is not other home miners, either: Pearl's own economics are modelled on datacenter H100 costs (~$1.70 per GPU-hour). You are racing AI datacenters at their own arithmetic. And selling is a chore: wrapped PRL on Bitrue Alpha, perpetuals on ApeX, and a Coinbase listing roadmap entry — which is not a listing.

Score: profitable today, durability unproven. The number and the warnings are the same story.

Quantus: seven weeks old, and the float is a burning match

QTC is also a coin — a standalone proof-of-work Layer-1, built around post-quantum signatures (ML-DSA, the NIST FIPS 204 standard) from block one. Mainnet launched September 9, 2026 — seven weeks before our check date. Behind it: Quantus Labs, co-founder and CEO Christopher Smith, pre-launch security reviews and an Immunefi audit competition, and backers including a Balaji-led round and AngelList co-founder Babak Nivi.

The money: a 4090 grosses $4.32 a day, netting +$3.60 at $0.10/kWh and +$3.24 at $0.15. All block rewards and fees go to miners — no developer tax.

Where the money comes from: emissions again — a steady drip of new QTC against a 21 million hard cap. But the supply side is where this gets sharp. Only 73% of QTC is mineable. The other 27% — 5.67 million QTC — was created at genesis for investors, founders, team and company (4.83 million to investors and team, 840,000 to the company), most of it locked. Circulating supply on October 31 was just 316,105 QTC at $73.39 — a $23.2 million market cap on a float you could fit in a spreadsheet. QTC only became properly tradable on October 6, 2026, via NEAR Intents, not a native exchange order book.

The dated event hanging over 2027: the genesis allocations begin unlocking in September 2027. Even a partial unlock, partially sold, dwarfs a 316K-coin float. A chain pitched as "Bitcoin, but quantum-safe" starts with a 27% insider allocation where Bitcoin had zero. Locked is not gone. The #2 GPU paycheck of October 2026 rests on a seven-week-old chain, a tiny float, one intent-based venue, and that unlock date: profitable today, handle like a burning match.

The classics: pennies at $0.10, dead at $0.15

Every classic below is a coin on its own chain, and every one pays miners almost entirely in new issuance — subsidy, not fees. The menu, on a 4090:

Coin	Gross $/day	Net at $0.10/kWh	Net at $0.15/kWh	The one-line problem
Conflux (CFX)	$1.03	+$0.36	+$0.03	Best classic — and the pay is inflation; fees are ~zero
Zano (ZANO)	$0.86	+$0.14	−$0.22	Miners share the block stream with stakers
Firo (FIRO)	$0.84	+$0.12	−$0.24	Miners were cut to 5% of the block in Sep 2024
Vertcoin (VTC)	$0.40	+$0.05	−$0.12	Total daily trading volume: $7.94. Unsellable
Ergo (ERG)	$0.32	−$0.12	−$0.34	Whole network pays miners ~$622/day, worldwide
Ravencoin (RVN)	$0.29	−$0.43	−$0.79	Halving landed, chain got rewound, exchanges left
Ethereum Classic (ETC)	$0.05	−$0.55	−$0.85	ASICs own it; a GPU is a space heater here
Kaspa (KAS)	$0.00	−$0.58	−$0.86	You pay 58 cents a day to earn 0.0093 KAS

Conflux is the cleanest "still a GPU coin" story: no Octopus ASIC found, GPUs genuinely run the chain (Nvidia only, in practice). But the PoW reward was already cut from 0.8 to 0.4 CFX during 2026, settings are governance-chosen with no fixed cap, and usage is thin — ~8,300 daily transactions in an August 2026 snapshot. Fees do not fund the payout; new issuance does, partly clawed back by burns. Founded by Dr. Fan Long, with Turing Award laureate Andrew Chi-Chih Yao as chief scientist — and one flag: DWF Labs bought $10 million of CFX in 2023 directly from team and foundation reserves. Disclosed, yes. Still a team-to-market-maker sale.

Zano has the healthiest token in this group — $3.76, ~$58 million market cap — and a hybrid catch: since the Zarcanum fork (March 2024), PoW and private proof-of-stake blocks alternate, and the single 1 ZANO per block reward goes to whoever produced that block. GPU miners compete for the PoW half of issuance — 720–1,440 ZANO a day (sources conflict on block time; the range is the honest figure), ~$2,700–$5,400 a day of subsidy, with all fees burned. The 3.69M ZANO premine at launch was disclosed.

Firo is genuinely GPU-only — no FiroPoW ASIC exists — and miners still got cut: a September 2024 hard fork set the block split to 70% masternodes / 5% miners / 15% development fund / 10% community fund. A GPU miner's slice fell 80% by governance vote, not by hashrate. Binance delisted FIRO in April 2025; an inflation bug in its Spark privacy system was disclosed in August 2026 and patched within weeks; from July 10, 2027 EU firms are barred from offering anonymity-enhancing coins — a dated headwind for every privacy coin here.

Ergo kept its ASIC-resistance promise — Autolykos v2's growing memory table means no commercial ASIC — and pays a flat 3 ERG per block all through 2027. The catch: those 3 ERG come out of the EIP-27 re-emission box — yesterday's subsidy, saved up by 2022-era miners who took smaller rewards — topped with small fees and storage rent. Total miner pay: ~$622 a day, worldwide. Break-even power ~$0.06/kWh; the Foundation's own page says its liquid treasury is "effectively depleted." The algo kept its promise. The market didn't.

Ravencoin is the saddest file here. KawPow resistance worked — GPUs really are the network. Then 2026 happened: the halving cut the reward to 1,250 RVN in January; in August a flaw in KawPow's header validation let forged blocks skip the real work, and two pools — 2Miners and RavenMiner — rebuilt the chain from a checkpoint, discarding roughly three days of transactions, over a core developer's stated preference for a shallower fix. Upbit, which carried ~57% of RVN's volume, delisted it in October; KuCoin and Bitvavo left too, and RVN printed a new all-time low on October 30, 2026. Total miner revenue: ~$2,475/day.

Ethereum Classic and Kaspa are ASIC coins now; stop calling them GPU coins. One Jasminer X44-P does the Etchash work of ~370 RTX 4070s. ETC still pays out ~$74,800 a day, its reward flat at 1.6384 ETC through all of 2027 (next cut ~September 2028) — a GPU captures $0.05 a day of it, and the planned Olympia upgrade would redirect the base fee to a protocol treasury, away from miners, date still TBD. Kaspa is worse: one Antminer KS7 equals 20,000 RTX 4090s, the subsidy steps down 5.61% every month through 2027, and fees cover ~0.2% of miner pay. Junk-data warning: roundup sites claim a 4090 does 12 GH/s on Kaspa. WhatToMine's measured figure is 2.00 GH/s — wrong by 6×, and any earnings math built on it is fiction. Radiant and Aleo are ASIC-gone too: one IceRiver RX0 (260 GH/s at 100W) outruns ~145 top GPUs on Radiant, whose whole network issuance is worth ~$91 a day; Aleo proving belongs to IceRiver and Goldshell boxes, gated by a per-solution stake that rises every quarter — the venue's own guide says GPU mining there "isn't viable on mainnet."

Dead: do not buy hardware for these

• Iron Fish (IRON) — shut down October 31, 2026. A flaw in a 2019 Merkle-tree commit let the same private note be spent repeatedly; ~2.98 billion wrapped IRON were minted on Base with no backing. The foundation had already wound down to maintenance mode in April 2026, after Coinbase acquired the team (not the network) in March 2025. Support ended and the repo was archived the day we checked.
• Flux (FLUX) — mining abolished October 25, 2025. The PoUW v2 fork replaced GPU mining with Proof of Nodes: block rewards now go to collateralized node operators, zero to miners, and the supply cap was raised from 440 million to 560 million at the fork.
• Clore (CLORE) — the PoW chain was killed in December 2025. Holders were migrated 1:1 to an ERC-20 token on Ethereum; the claim window closed in February 2026 and, in the project's own words, unclaimed legacy coins are "permanently unrecoverable." The GPU rental marketplace survives (below) — the mining does not. The docs even state why: miner sell pressure was a built-in headwind.
• Pyrin (PYI) — abandoned December 2024. Per 2Miners, the developers walked away and deleted the project's resources. No verified live market remains.
• Dynex (DNX) — a zombie, and Spectre (SPR) — never a GPU coin. Dynex blocks still tick under a ~$106,000 market cap; SpectreX is CPU-only by design. Neither belongs on a 2027 GPU list.

The 2027 decision: mine it, rent it, or sell it

Mine it. Pearl nets +$5.10 to +$5.75 a day on a 4090 at $0.10/kWh, Quantus +$3.60 — paid in brand-new coins, from emissions. The catch is durability: Pearl's earnings already fell ~75–85% from May to October, and Quantus has the September 2027 unlock. Best daily number, worst foundation under it.

Rent it. On Vast.ai, a 4090's median renter price is ~$0.428/hour and the platform takes ~20%, leaving the host ~$0.342/hour while a job runs. At 50–70% utilization that nets ~$3.69 to $5.16 a day at $0.10/kWh ($3.48–$4.87 at $0.15); Clore.ai's lower fees land in the same band, ~$3.85 to $5.40. Salad's published host pay is lower — $0.173/hour, netting $1.66 to $2.32 a day. Two honest caveats: renter price is not host income (RunPod's $0.34/hour is a renter price; io.net's $0.40–$0.80 renter range has no verified home-host payout behind it, so neither is quoted as income), and utilization is not yours to control — unmanaged home cards realistically run 20–50% rented, cutting rental nets to roughly $1.50–$3.00 a day. The payer here is real, though: renters paying for compute — real revenue, in dollars or stablecoins, not emissions.

Sell it. A used 4090 sold for $2,342–$3,010 (the trackers disagree; the range is the honest figure), held up by AI demand, not mining. That equals roughly 470–590 days of Pearl mining at $5.10 a day, or 630–815 days of renting at 50% utilization — paid immediately, with zero power bill, zero token risk and zero utilization risk. A used 4070 fetches $525–$599, about 250–330 days of its own Pearl net.

Our read: if you already own the card and your power is $0.10 or less, mining Pearl while it lasts — and selling the coins as they land — is the best daily cash flow; renting out is a close second and pays in money instead of a six-month-old token. At $0.15 power, the classics are all dead and Pearl and Quantus thin out fast, while the used market offers you two years of optimistic income up front. For most people holding a 4090 "for mining," the exit is the trade. None of these numbers include depreciation, either — a card run hard for two years will not sell for today's price.

Get posts like this by email: Subscribe to BogusAPY

Follow BogusAPY on X: @BogusAPY

Check it yourself: links and sources

• GPU profitability tables and calculator snapshots, checked October 31, 2026: WhatToMine — GPUs · hashrate.no GPU compare
• Prices, market caps, supplies and volumes, checked October 31, 2026: CoinGecko — Kaspa · Ethereum Classic · Conflux · Ergo · Ravencoin · Zano · Firo · Vertcoin · Quantus
• Quantus — independent guide, tokenomics and the September 2027 unlock: whatisquantus.com
• Pearl — launch coverage and profitability reporting: KuCoin and Odaily Pearl coverage; Basu, "The Usefulness Gap in Proof-of-Useful-Work" (preprint, June 3, 2026); Hashrate Index data via Odaily.
• Rental rates and host economics: Vast.ai · Clore.ai 4090 rental snapshot · Salad pricing — renter prices and host rates kept separate in the text.
• Used-card prices: GetPCParts and InsiderLLM eBay sold-listing trackers, October 2026.
• Dead-chain records: Clore official tokenomics docs (PoW chain "permanently unsupported") · RunOnFlux PoUW v2 notes (mining ended at block 2,020,000, October 25, 2025) · 2Miners December 2024 report (Pyrin) · Iron Fish shutdown reporting, October 31, 2026.
• Project docs for the classics: ErgoDocs (EIP-27), ECIP-1017 / ECIP-1111 (Ethereum Classic), Zano docs (ProgPowZ, Zarcanum), Firo fork notices.

Figures checked October 31, 2026.

Not financial advice.