Get paid for the GPU time you're wasting.
Mine a coin and you're exposed to its price. Rent the card for AI and it earns nothing when nobody books it. GPN removes the choice: every node is both a miner and an AI provider on the same GPU. It mines by default, and when a job arrives it pauses mining, frees the VRAM, runs the job, and resumes — one card, one job at a time.
Two ways to earn, and you have to pick one
Owning a GPU has always presented two, siloed markets for the same silicon. They were never connected — so the hardware can't move to where the money is.
① Mine a proof-of-work coin
You're paid whatever the coin is worth today. Coin dips and you mine at a loss — the card keeps hashing because it can't see anything else.
② Rent the GPU for AI work
You're paid whatever compute is worth today. List it and nobody books you, and it earns exactly nothing while it idles.
AI compute is expensive and concentrated: buyers pay hyperscaler prices, while consumer GPUs — the cheapest compute on earth — sit idle.
One node, two income streams, no mode to pick.
A GPN node is always a miner and an AI provider — one identity, one wallet, one GPU. Mining runs whenever the card is free. An incoming AI job pre-empts it: mining stops, VRAM is released, the job runs to completion, mining resumes. The chain handles coordination, escrow, and settlement; the heavy compute stays off-chain.
There is no provider-only or mine-only mode, and no profitability switch. The GPU is fully yielded per job — so a job never contends with the miner for VRAM. A separate Staking role is planned.
Mine, serve, route, settle
One install gives you a node that mines on its GPU and serves AI jobs when they arrive. A buyer hits one endpoint; the network does the rest.
Mining
GPU proof-of-work (Ethash-derived, memory-hard). The node that produces a block keeps that block's reward — Bitcoin-style. Eligibility is GPU-only, verified by work, not by a permissioned validator set.
Serving AI
LLM inference, image/video generation, ML inference, and fine-tuning — priced per token against real cost, with per-job escrow and an automatic surplus refund.
One GPU, one job
A job pre-empts mining: the miner stops, VRAM is freed before the model loads, and mining resumes the instant the job settles. No two jobs ever share a card — so it never over-subscribes compute or memory. Every job type is enabled by default.
Routing
Buyers hit a single access point. The scheduler filters by capability, requires an idle GPU, and picks the lowest-latency, least-loaded verified provider — discovered on-chain, not from a hand-maintained list.
Escrow & settlement
Funds are escrowed on-chain and released on delivery. Optional payment channels lock funds once and settle each job off-chain, so the AI path has no per-job block delay.
Privacy
A shielded pool (Sapling-style commitments + nullifiers) supports private transfers and withdrawals. The chain sees job specs and result hashes — not prompts, models, or outputs.
Three public nodes — one Linux bootnode and two Windows full nodes — are
reachable on the testnet. Builds roll out through rolling GitHub releases
(windows-latest, linux-latest, macos-latest). The repository
is still private; ask for access to download.
Both markets, one machine
No incumbent lets a GPU mine and serve with an automatic switch. Here's the factual positioning.
| Network | What it does | Where GPN differs |
|---|---|---|
| NiceHash | Hashpower marketplace — sells your GPU hashing to the highest bidder (among PoW coins). | GPN also serves AI compute on the same card — mining runs whenever no job is queued, and pauses while a job owns the GPU. |
| Vast.ai | Rental marketplace that lists GPUs for rent; owner picks a listing and a price. | GPN routes AI jobs to your card by capability + latency and runs mining on it the rest of the time — no listing step. |
| RunPod | Centralized GPU cloud for AI; rent capacity on demand. | GPN is a permissionless, trust-minimal network with on-chain escrow — not a hosted cloud. |
| Render | Decentralized GPU rendering network; job-based payouts to node operators. | GPN couples the marketplace with a native proof-of-work chain and mines whenever the GPU is idle. |
| Akash | Decentralized compute marketplace (containers/VMs) with a reverse auction. | GPN is GPU-native, AI-job-shaped (per-token LLM/image/video), with PoW rewards for idle time. |
| io.net | Aggregates distributed GPUs for AI compute clusters. | GPN targets single consumer GPUs and lets each one fall back to mining when unbooked. |
Positioning based on each project's public product framing; not a ranking or endorsement.
Built to be earned, not trusted
Bitcoin-style rewards
- Producer keeps the block reward — a coinbase model, no shared beneficiary.
- Per-node wallets — each miner's own account is its payout address.
- Self-tuning difficulty; 15-second blocks; a ~2-year halving.
GPU-only, verifiably
- Memory-hard PoW makes CPUs non-competitive by construction.
- Attestation tiers (Bronze/Silver/Gold) raise your reward multiplier.
- Stake-free path: producing a valid block is proof of real GPU work.
Trust-minimal
- On-chain escrow; the chain is the root of trust.
- Optional payment channels — no per-job block delay.
- Committee governance (13-of-25), timelocked; an immutable genesis.
Privacy
- Shielded pool: hidden notes, zero-knowledge spends, unlinkable nullifiers.
- Prompts, models, and outputs stay off-chain.
- On-chain values are salted/opaque.
GPN at a glance
A fair-launch, proof-of-work coin whose durable value derives from the AI compute it enables — not just mining inflation.
Fair launch
No ICO and no liquid premine — all GPN is minted per block and halves every 4,204,800 blocks (~2 years at 15 s). Halving is the sole emission mechanism; emission converges at ≈ 420 M GPN.
Disclosed, vested treasury
100 M GPN held by a keyless pallet account — 70 M unlocked at genesis and 30 M vesting over 10 years — spendable only by a 13-of-25 governance roster. A 5% marketplace fee splits 40% staking / 30% treasury / 20% burn / 10% oracle operators.