Pearl is a Layer 1 blockchain where mining is a by-product of useful AI computation. The same GPU work that runs AI inference also secures the network — one operation, two outputs.
The problem Pearl solves
Traditional proof-of-work chains like Bitcoin secure themselves by burning enormous amounts of energy on calculations that have no purpose beyond being hard. The security is real, but the computation is discarded.
Meanwhile, hundreds of billions of dollars have gone into GPUs and data centres to build AI — and the overwhelming majority of that compute is spent on one operation repeated at vast scale: matrix multiplication.
Those two efforts ran side by side for years with nothing in common. It was widely believed to be mathematically impossible to make proof of work into proof of useful work.
What Pearl does differently
Pearl's core result is that the same matrix multiplication powering AI inference can, at the same moment, secure a monetary network. Every GPU cycle does double duty.
In practice that means a GPU running AI inference can mine PRL from work it was doing anyway. The mining isn't an additional cost — it's extracted from computation that already has value.
For the mechanism, see What is Proof of Useful Work?
The essentials
| Type | Layer 1 proof-of-work blockchain |
| Consensus | Proof of Useful Work (PoUW) |
| Token | PRL |
| Total supply | 2.1 billion PRL, capped |
| Smallest unit | A grain — one hundred-millionth of a PRL |
| Block time | 194 seconds (3 minutes 14 seconds) |
| Accounting model | UTXO, like Bitcoin |
| Addresses | Taproot only, beginning prl1p |
| Mining hardware | GPUs, via the Pearl vLLM miner |
| Licence | Open source, ISC |
What you can do with Pearl today
- Hold and transfer PRL using the Pearl wallet
- Mine by running a node and the vLLM miner on your own GPUs — see How mining on Pearl works
- Buy a compute package on the Pearl Compute Platform if you'd rather not run hardware yourself
- Run a full node and verify the chain independently
- Read the code. Everything is public at github.com/pearl-research-labs/pearl
The Pearl Inference Platform is in closed beta — you can join the waitlist at pearlresearch.ai/waitlist.
Who builds Pearl
Pearl Research Labs, a member of the NVIDIA Inception Program. The project came out of research and stays close to it — the whitepaper, published papers, and open problems are all public.
Read more
- The Pearl whitepaper — the full protocol specification
- The PoUW paper on arXiv
- Our vision
- Open problems in mathematics, cryptography, and consensus