Skip to content
An iridescent pearl whose nacre is made of small square tiles

The currency
of AI.

The first conference dedicated to Pearl, a blockchain that GPUs can mine while they run AI models.

1 day in 2027. City and date to be announced.

Get first accessApply as sponsor

Pearl mainnet

Snapshot from prlscan.com, Oct 3, 12:07 UTC. Supply is computed from block height.

Block height
122,614each carries a proof under 60 KB
PRL mined
333.2M15.87% of the 2.1B cap
Hashrate
47.7 EH/s50-block estimate
Mainnet age
159 dayssince the Apr 27 fair launch
Last block
Oct 3, 11:49UTC, against a 194-second target

Proof of work, put to work.

Pearl miners compute matrix multiplications, the operation AI models spend most of their GPU time on. A GPU serving a model can mine PRL in the same pass.

PRLCON is the first conference built around that idea and the network running it: the research, the fleets, the capital and the open questions.

  1. 1992

    Cynthia Dwork and Moni Naor introduce proof of work to price junk mail, and ask whether the work could serve another purpose.

  2. 2009

    Bitcoin launches. Its proof of work secures the ledger and has no other use.

  3. 2019

    Vitalik Buterin rates useful proof of work “probably not feasible, with one exception.”

  4. 2025

    Ilan Komargodski and Omri Weinstein publish proofs of useful work from arbitrary matrix multiplication.

  5. 2026

    Pearl’s mainnet launches on April 27 with no premine and no VC allocation.

  6. 2027

    PRLCON.

1 pass, 2 outputs.

Pearl can mine inside the matrix multiplications a model already runs. When it does, the GPU returns the model’s answer and a stack of lottery tickets from the same work.

Construction: Komargodski and Weinstein, arXiv 2504.09971. Proof size: Pearl’s INT whitepaper.

What it costs

Mining while serving adds about 4 to 15% to a GPU’s work on today’s INT8 protocol, depending on the workload, according to Pearl’s July benchmarks and its CEO, Omri Weinstein. He puts the FP8 upgrade, which he says is on testnet, at about 0 to 5%.

What it doesn’t settle

Synthetic matrices can mine too. A GPU that already serves customers pays only the overhead, while a synthetic miner pays for the whole GPU-hour. Whether that edge decides who mines depends on inference demand relative to block rewards, Weinstein says.

Every layer since genesis.

A pearl keeps its history in its layers. These are Pearl’s first 159 days, drawn to scale by date.

Cross-section of Pearl's first 159 days, one ring per milestoneToday
Ring radius scales with days since genesis. The dashed ring is the next fork, which has no date yet. Tap any ring to see what it marks.
  1. Apr 27GenesisFair launch, with no premine and no insider or VC allocation. Certified Llama models and the open-source miner shipped the same day.
  2. May 15First provider integrationTogether AI launched a Pearl-powered Gemma 4 31B endpoint, which it said was discounted by over 25%.
  3. Jun 12First hard forkMixture-of-experts work could mine from block 71,935.
  4. Jul 23Dense-only soft forkAfter an exploit in MoE certificates, MoE mining switched off at block 91,630.
  5. Aug 6Rank-penalty soft forkFrom block 96,251, mining noise must have rank 128 or more.
  6. Aug 11Salted noise-seed hard forkCertificate V3 became mandatory at block 99,000.
  7. Aug 15Block 100,000Mined 110 days after genesis.
  8. Sep 14Proof of inferencePearl announced work with Attestable on privacy-preserving proof of inference and published its FP8 specification. Source
  9. By Sep 30Inference Platform livePearl’s own API serves 8 models on prepaid credits. Source
  10. NextFP8 Pearl-GEMMA hard fork to certificate V4 that adds Blackwell support to Pearl’s official miner and brings MoE mining back. Pearl’s CEO said on Sep 25 it’s on testnet and “expected to launch within weeks.” No mainnet height is set yet. Source

5 blocks, each built around an open question.

1 day, plus an invite-only dinner the night before. Every block pairs someone running production systems with someone who directs capital or research.

  1. 01

    Mining and fleet ops

    What does a GPU-hour earn when it serves a model and mines PRL at once?

    Fleet economics, Blackwell support in the FP8 branch, pool concentration, power contracts.

  2. 02

    Models and inference

    How cheap can inference get if block rewards cover part of the bill?

    Certified models, the Pearl Inference Platform, provider pricing, demand.

  3. 03

    Protocol and core

    How close to zero can mining overhead go?

    FP8 Pearl-GEMM, the road to FP4, proof of inference as an L2, post-quantum spending paths.

  4. 04

    Markets and liquidity

    How do allocators size a position while no major exchange lists PRL?

    Valuation frames, price discovery, OTC desks, treasuries holding PRL.

  5. 05

    Building on Pearl

    What do agents do when they hold the money they spend on inference?

    Agent payments, custody on a Taproot-only chain, wallets, paying for inference in PRL.

Who’s in the room.

  • Fleet operators

    Datacenter and GPU-cloud teams pricing mining as a second revenue line on hardware they already run.

  • Researchers

    Cryptographers, kernel engineers, economists and quantization specialists working on proof of useful work.

  • Capital

    Funds, family offices, corporate treasuries and OTC desks underwriting compute and the money it mints.

  • Builders

    Teams building agents, wallets, explorers and payment apps on Pearl.

17 speaking seats.

Names go up as speakers confirm.

Submit a talk
  1. Opening keynoteThe origin of useful work
  2. 01 Mining and fleet opsPool operatorGPU-cloud executivePower developer
  3. 02 Models and inferenceInference providerQuantization engineerNeocloud product lead
  4. 03 Protocol and coreProtocol researcherKernel engineerPolymath contributor (lightning)
  5. 04 Markets and liquidityOTC desk operatorFund partnerPoUW economist
  6. 05 Building on PearlAgent-payments builderCustody engineerVenture investor
  7. Closing keynoteTo be announced

Put your name on the first one.

3 tiers, allocated by application. Packages finalize with the venue and date, and early partners help shape them.

  • South Sea pearl, drawn at relative size

    South Sea

    1 available

    The largest cultured pearls, usually 8 to 18 mm.

    Title partner

    A single partner, named on the event and across the program.

  • Akoya pearl, drawn at relative size

    Akoya

    The classic round pearl, usually under 9 mm.

    Stage and floor

    Main-stage presence and a floor booth where rigs run live.

  • Keshi pearl, drawn at relative size

    Keshi

    Grown without a bead nucleus, as a byproduct of pearl culturing.

    Early room

    For teams that want to be in the room before it has walls.

Pearls drawn to relative size.

Apply as sponsorSubmit a talk instead

First Access.

You’ll get the city and date before they’re public, and an email when tickets open.

I’m interested in

Questions.

When and where is PRLCON?

It’s 1 day in 2027, with an invite-only dinner the night before. The city and date haven’t been announced yet.

Who is it for?

People who put capital or hardware to work in the Pearl economy: GPU-cloud and datacenter operators, protocol researchers, funds and treasuries, and teams building on the network.

Is PRLCON run by Pearl Research Labs?

No. PRLCON isn’t affiliated with Pearl Research Labs. It covers Pearl the way a Bitcoin conference covers Bitcoin.

Do I need to hold PRL to attend?

No. The program is built for people evaluating Pearl as well as those already mining or holding it.

How does sponsorship work?

3 tiers, South Sea, Akoya and Keshi, allocated by application. Apply through First Access and we’ll send the prospectus.

Can I submit a talk?

Yes, for any of the 5 blocks. Production data and primary research get priority over product presentations.

Is there press access?

A limited number of passes. Request one through First Access and name your outlet.