Compute that flowslike liquidity.
LooPIN pools idle GPUs from providers around the world into a PinFi liquidity pool. Builders draw compute at market-set prices, providers earn on every GPU-hour, and the loop keeps turning.
- 4,607 GPUs online
- 41 countries & regions
- 2,525 machines online
Running on the network today
- H100
- A100 80GB
- RTX 5090
- RTX 4090
- RTX A6000
- RTX 3090
- RTX A5000
- Tesla V100
- CUDA
- Docker
- Kubernetes
- Solana
The network, live
Read from the network at 23:52 UTC- GPUs online now
- 4,607
- Staked by providers and reporting in right now.
- Countries & regions
- 41
- Where providers have brought GPUs onto the network.
- Machines online
- 2,525
- Provider servers connected to the pool.
- PinFi liquidity pools
- 7
- On-chain markets where GPU-hours find their price.
Idle silicon in. Useful compute out.
Three moves, one market. Providers fill the pool, builders drain it, and fees flow back to the people who keep the GPUs running.
- 01
Provide
Stake idle GPUs into the pool. You receive stTokens that represent your machine's GPU-hours and earn as they're consumed.
- 02
Pool
The PinFi pool is dissipative: unused GPU-hours expire with time, so supply is always priced to be used, never hoarded.
- 03
Consume
Builders draw compute as servers, serverless inference or GPU-hour tokens, at the price the pool quotes in that moment.
One pool. Four ways in.
Rent a whole machine, call a model, trade GPU-hours or watch the network breathe. Every surface draws from the same liquidity.
Market-priced, not list-priced.
Traditional clouds publish a rate card and hold it whether their racks are full or empty. The PinFi pool quotes from live supply and demand: when GPUs sit idle the price falls until someone uses them, and when the pool runs hot it rises to pull more hardware in.
“This approach significantly reduces the costs of accessing computing power, potentially to as low as 1% compared to existing services.”
- Supply-aware. Price responds to how much of the pool is in use.
- Expiring, not hoarded. GPU-hours dissipate, so idle capacity is priced to clear.
- Settled on Solana. Pool state and trades live on-chain.
Built for every side of the market.
A market only works when both sides show up. LooPIN gives builders, GPU owners and liquidity providers their own way in.
Train, fine-tune and serve without a reserved-instance contract.
- Launch GPU servers from the dashboard: pick the card, the image and the hours.
- Call open models through the serverless API with one key and prepaid credits.
- Buy GPU-hours from the pool when you need them, at the price it quotes then.
pytorch/pytorch:latest24 hTurn idle cards into a yield-bearing position in the pool.
- Stake an idle machine and receive stTokens that track its GPU-hours.
- Watch status, utilisation and earnings per machine in one place.
- Take your GPU back out of the pool whenever you need it.
A100-SXM4-40GB × 8stToken · stakedBack the market that prices compute, and earn from its flow.
- Add liquidity to a PinFi pool and earn from the trades it routes.
- Swap GPU-hour tokens at the live pool price on the Swap page.
- Block-reward bounds for honest LPs are studied in our published research.
gpuTokenPool collateralPublished, not promised.
The mechanism behind LooPIN is published in the open, so anyone can check the maths before they trust the market.
Read the protocol paper on arXivWhat is LooPIN?
LooPIN is a decentralized GPU computing network built on a PinFi (Physical Infrastructure Finance) protocol. Providers stake idle GPUs into a liquidity pool; builders rent servers, call a serverless inference API or trade GPU-hours from that pool at market-based prices.
Why can't I sign up?
Registration is invite-only while we scale the network. If you already have a LooPIN account you can sign in as usual. Everyone else can join the waitlist and we'll send an invite as capacity opens up.
What makes a PinFi pool “dissipative”?
Unlike a DeFi pool of tokens that keep their value, GPU time is perishable: an hour that goes unused is gone. The PinFi pool prices that decay in, so idle capacity gets cheaper until it is consumed, and scarce capacity gets more expensive until new providers join.
How do GPU providers earn?
When you stake a machine you receive stTokens representing its GPU-hours in the pool. You earn as those hours are bought and consumed by builders, and you can remove the machine from the pool when you need it back.
What can I run on LooPIN?
Full GPU servers you deploy and control from the dashboard, and serverless inference against open models in the model zoo through an API key and prepaid credits. You can also buy and sell GPU-hours directly on the Swap page.
How is the price set?
By the pool, not a rate card. The PinFi mechanism quotes prices from current supply and demand. The research paper argues this can bring the cost of compute potentially as low as 1% of existing services.
The loop is opening.
We're inviting builders, GPU owners and liquidity providers in waves. Leave your email and we'll let you in as capacity grows.
Already a member? Sign in