PRL public chain replaces traditional hash computation with AI computing power mining. As L2s shut down one after another, can the PoW narrative be revived through useful proof of work? The Pearl team's background and commercial deployment progress have sparked heated discussion, but its real revenue and token economics remain controversial.
Author: Bitcoin Orange Trader
L2s are lining up to shut down, yet PoW is starting to cluster: What exactly are the foreigners cooking up with $PRL, the hot topic at TOKEN2049? (Related reading: TOKEN2049 essay collection: Fewer Builders, more institutions, Crypto begins to move from virtual to real)
Recently I've seen several PoW projects emerge in succession, right as L2s like Blast and Abstract are shutting down one after another.
During TOKEN2049, one project the foreigners were discussing quite intensely was Pearl ($PRL). After researching it for a while, I found that its team, technology, and the funding behind it really do have substance, and it is trying to solve a problem that has existed for many years: can AI compute incidentally mine a currency similar to Bitcoin?
First, what exactly does PRL do?
It is an independent PoW public chain launching on April 27, 2026, using Proof of Useful Work (PoUW). Bitcoin miners perform SHA-256 hash computations to compete for block rewards, while Pearl replaces the mining work with the matrix multiplication heavily used by large AI models.
If an AI company already needs to use GPUs to do inference for customers, then the same batch of computation can also participate in Pearl mining and earn PRL rewards.
Ideally, a single GPU simultaneously earns money from AI customers and from mining; miners can earn more, and AI service providers can also use part of the revenue to lower prices.
The team background is one of PRL's biggest pluses.
CEO Omri Weinstein is a Princeton computer science PhD, an associate professor at Hebrew University, and has taught at Columbia University, with a resume involving NVIDIA and VAST Data.
Another co-founder, Ilan Komargodski, is an associate professor of cryptography at Hebrew University, a PhD from the Weizmann Institute of Science, and has conducted research at Cornell Tech and NTT Research.
The third co-founder, Idan Sugarman, has a business and investment background, and Cornell cryptography researcher Rafael Pass is also involved in economic security research.
The core team has long studied computational complexity, cryptography, and verifiable computation. They published the original PoUW paper back in 2025, so the technology was not hastily cobbled together to ride the AI hype.
There has also been substantive progress in commercial implementation.
On May 15 of this year, Together AI officially announced a partnership with Pearl Research Labs to launch the Gemma-4-31B-it-Pearl inference service, using mining subsidies to reduce the inference price of that model by more than 25%.
Pearl itself has also built an inference platform compatible with the OpenAI API, supporting open models such as GLM and DeepSeek, with USD top-ups and billing by Token consumption.
Customers, products, and a charging mechanism are all in place, but key data such as real paid revenue, average daily inference volume, and customer retention are still missing, so for now it cannot yet be proven that this business has already scaled up.
Next, who is supporting it?
No Limit Holdings (NLH) has publicly stated that it is actively accumulating PRL. Its investor Brian Breslow published a long research piece in September calling Pearl a potential next-generation Bitcoin and even giving a long-term target of more than $30.
NLH's core figure Gin Chao once served as Binance's Chief Strategy Officer and is also a BinanceUS board member. This shows that the funding circle bullish on PRL does have certain backing, but it should not be understood as Binance investing in or endorsing Pearl.
PRL has no traditional VC token rounds or pre-mine allocation. Early chips were mainly produced through GPU mining and then circulated through third-party OTC markets such as Pearl OTC and Black Pearl OTC.
The Pearl OTC website self-reports cumulative trading volume of about $31.5 million, and spot trading has now also appeared on exchanges such as BigONE and SafeTrade.
PRL has a total supply cap of 2.1 billion coins, with no conventional team pre-mine or VC unlock. Its issuance method is a smooth decline in block rewards, without Bitcoin-style sudden halvings every four years.
The problem is that early mining difficulty was very low. In the first roughly 5.9 days, 121.7 million coins were mined, accounting for about 36% of the currently circulating supply. How much the team and early miners each took is not transparent.
At a reference price of about $1.28 on October 8, the circulating market cap is about $430 million, FDV is already close to $2.7 billion, and theoretically about 1 million PRL are still being added every day. The potential selling pressure formed by miners' continuous output cannot be ignored.
There is also a major controversy on the technical side:
Pearl can prove that miners did indeed complete matrix computations, but it cannot prove that these computations necessarily came from real paying AI customers.
In June of this year, a study observed miner software and actual mining behavior and found that at the time a large amount of computation could be completed with random matrices, without anyone needing to purchase inference services at all.
The project team's logic is that in the future miners truly serving AI customers will have two sources of income, while pure miners will only have coin rewards. As competition increases, the latter's cost advantage will disappear.
This economic model makes sense, but it has not yet been proven by large-scale real commercial data.
The MoE mining vulnerability that appeared in July also reminds us that the consensus implementation is still at an early stage, and the FP8 upgrade supporting broader AI workloads still needs to be verified on the mainnet.
At present, Pearl's inference platform mainly collects USD, miners earn customer payments and PRL rewards, and the token has no confirmed large-scale revenue sharing, mandatory buyback and burn mechanism.
A larger business can improve miner economics, strengthen network security, and boost market confidence, but for AI demand to continuously translate into direct buying of PRL, it still depends on whether there will be real token payment, coin-denominated settlement, collateral, and reserve demand in the future. Otherwise, even if inference revenue grows, the coin price may still be diluted by continuous mining issuance.
The death wave of L2s shows that the market does not need so many redundantly built chains, while a project like PRL is trying to find a new way out for PoW. PoW+AI has the potential to become an important narrative going forward, but what ultimately determines how far PRL can go is whether AI companies are willing to pay for this compute over the long term.
Being able to let AI compute mine a coin is a technology. Being able to make this coin an asset the market truly needs to hold is the real big business.