Go back

This project claims that Bitcoin is useless: Breakdown of Pearl Research Labs

from More Than Speculation | The Chill Crypto Convo

16m 56s

This project claims that Bitcoin is useless: Breakdown of Pearl Research Labs

Pearl Research and its token PRL have drawn intense attention after a roughly 10X price move and heavy buzz at Token2049. The project is a Bitcoin-derived Layer 1 that swaps arbitrary proof-of-work puzzles for "proof of useful work": miners run noisy matrix multiplications (GEMM), the same math AI inference requires, and wrap proofs in ZK-Snarks so private inputs stay hidden. Its credibility rests on peer-reviewed research by founder Omri Weinstein, published code, a live mainnet, and a genuine partnership with Together AI, which offers discounted Gemma inference subsidized by PRL emissions. Its fair launch—no pre-sale, VCs, or airdrops—has strong community appeal. However, the central claim remains unproven at scale. Analyses suggest most mining involves dummy computations with no real AI customer, making PRL closer to AI-shaped proof of work than genuinely useful work. Daily emissions of about 1.2 million PRL (~$2.2 million) create constant sell pressure, and only ~16% of the 2.1 billion supply has been mined, leaving a large overhang against a ~$3.84 billion fully diluted valuation. Bullish voices like Brian Breslow see a next-generation Bitcoin, while skeptics note the single-partner dependency and lack of PRL-denominated demand. Key catalysts to watch are the FP8 hard fork, a PRL-priced inference marketplace with possible token burns, additional AI partners, and tier-one exchange listings. Ultimately, PRL is a bet that real AI demand will eventually validate useful work at scale—not a reflection of the present.

Transcription

2961 Words, 17300 Characters

English
Speaker 1A token pumped roughly the 10X from its bottom in just a few months. It's been one of the most talked about projects at token 2049. And up until very recently, you could only trade it on an exchange most people have never heard of, called Safe Trade. No token sale, no pre-sale, no VC allocations, no air drop farming. The only way to get this token from day one was to mine it with a GPU. So what is this project? Why is it generated this much buzz, and does the technology actually back up the hype? Today we're breaking down Pearl Research and its token, PRL, in plain language, so you can decide for yourself. Let's start from the most basic question. What problem is Pearl trying to solve? You probably know that Bitcoin uses something called proof of work. Miners around the world run computers that solve a mathematical puzzle over and over again. The first one to solve it gets the reward. That puzzle is completely arbitrary, it has no value to anyone except proving that the computer did the work. The only point is to make cheating expensive. Critics have been saying for years, "This is a massive waste of electricity. You're consuming the same amount of energy as a small country, and all you're producing is a number that nobody needs." Think of it like hiring a security guard for a bank, except the guard's job is to stand outside and solve Sudoku puzzles all day. The Sudoku proves they showed up and worked, but the Sudoku itself is completely useless. Pearl is asking, "What if we gave that guard something useful to do while they're standing there?" Specifically, what if instead of solving a meaningless puzzle, miners ran AI computations? Because AI models need an enormous amount of a specific type of math called matrix multiplication. And matrix multiplication is exactly what a GPU, a graphics card, is designed to do. The idea behind Pearl is called proof of useful work. Instead of wasting GPU power on arbitrary puzzles, miners run real AI math. That same computation that secures the blockchain could also be doing something an AI customer actually needed done. Before we go deeper into how Pearl works, you first need to understand AI inference. So, what is AI inference? You've used chat GBT, or maybe Claude, or Gemini. Every time you type a question and the AI generates an answer, that process of generating the response is called inference. The model is inferring, the best answer from everything it was trained on. Now, that inference doesn't happen by magic. It requires real computing power. Every single time you send a message to an AI somewhere in a data center, a GPU is doing millions of mathematical calculations to generate your response. That's inference. And inference is expensive. Companies like OpenAI, Google, and Anthropics spend enormous amounts of money running the GPUs that power their AI products. That cost is passed on to users and developers through API pricing. Pearl's pitch is, what if the same GPU during that expensive inference work could also be mining a cryptocurrency at the same time? The inference work that the GPU was already doing would produce a valid mining proof and the miner gets rewarded in PRL. In theory, AI gets cheaper, miners get paid, everyone wins. Let's get slightly more technical, but we'll keep it simple. Pearl is built as a layer 1 blockchain, its own independent chain, similar to Bitcoin, with its own coin and its own mining rules. It actually forked from Bitcoin's code. It uses the same model, longest chain wins, block rewards, halve over time, 2.1 billion total supply. The same philosophy as Bitcoin, just 100 times the number of coins. The key difference is the mining function. Instead of SHA, 256 hashing, Bitcoin's meaningless puzzle, Pearl miners run what's called a noisy gem operation. GMM stands for General Matrix Multiplication. Noisy means Pearl adds random noise to the input matrices before multiplying them. This is critical because it prevents miners from cheating by pre-computing shortcut answers. Now here's a clever problem Pearl had to solve. If a miner submits their matrix multiplication as proof, they'd be revealing the actual data they computed, which might be a customer's private AI request. So Pearl wraps each valid proof in something called a ZK Snark. A ZK Snark is a type of cryptographic proof that lets you prove you did a computation correctly without revealing what the inputs were. Think of it like a sealed envelope with a signature. The signature proves the letter is genuine, but you don't have to open the envelope to verify it. In September, 2026, Pearl also published a new specification to move from integer-based mining toward FP8, a floating point format used by modern production AI models. This matters because until now, there was a mismatch between how Pearl mined and how real AI models actually compute. FP8 closes that gap. The person behind Pearl Research Labs is Omri Weinstein, a theoretical computer scientist and academic who has been working specifically on the problem of making proof of work useful. He is not a crypto-influencer or a VC-backed founder, he is a researcher. The Mathematical Foundation for Pearl was published in a 2025 academic paper by Weinstein and Elon Komagotsky. The paper argues that it's possible to extract a valid mining proof from a matrix multiplication almost for free, meaning the proof adds almost no extra cost on top of the computation the GPU was already doing. That's the theoretical basis that makes the whole thing potentially viable. This is actually one of the things that separates Pearl from a long list of previous useful work crypto projects that either never shipped or turned out to be fraudulent. Pearl has a live mainnet. It has published code. It has peer-reviewed research behind its core claim. And the founder is a real computer scientist who has staked his academic reputation on it. That doesn't mean everything works as advertised. We'll get to that. But it does mean this is not a typical crypto launch. Pearl has one named production partner, Together AI. Together AI is a legitimate, well-funded AI infrastructure company that lets developers access large AI models through an API. In May, 2026, they announced an exclusive endpoint powered by Pearl. A version of Google's Gemma 431b model offered at a discount of more than 25% compared to standard inference rates. The discount is covered by PRL emissions that together AI earns from running Pearl compatible mining workloads. In plain terms, together AI runs a GPU data center. Those GPUs do AI inference for customers. Pearl's system lets those same GPUs also mine PRL while doing the inference. Together AI collects the PRL rewards, uses them to subsidize the inference price and charges customers less as a result. This is a real working integration. It's the clearest proof point Pearl has that the concept actually functions in production. But it is also the only one. And that matters, which we'll talk about next. Here's where we have to be honest with you, because this is the question that actually determines whether Pearl's valuation makes sense. The pitch is, every PRL mind represents real AI work done. GPU cycles that would have been wasted on meaningless puzzles are instead doing useful inference or training. That makes PRL a kind of compute back asset tied to real AI demand. The reality right now is more complicated. Hash rate index, a publication that specializes in crypto mining research, published a report in June, 2026, arguing that most of Pearl's mining is not producing any real AI output. Minors are running the matrix multiplication. That part is happening. But many of them are doing it without any actual AI customer on the other side. They're running dummy computations just to mine PRL. The matrices they're multiplying are not serving any real inference request. They're just random numbers chosen to satisfy the mining proof. If most of the network is doing this speculative mining without real AI customers, then Pearl is not really proof of useful work at scale. It's more like AI-shaped proof of work. The math is the same as AI math, but no one actually needed the output. There's also a second concern, emissions. Pearl is still in its early supply curve, block rewards are significant. That means a large amount of new PRL is entering circulation every day, currently around 1.2 million PRL per day. At today's price, that's roughly $2.2 million in new supply hitting the market every single day. For the price to hold or go higher, that daily cell pressure needs to be absorbed by new demand. And there's a third issue that is easy to overlook, the FDV. At today's price, Pearl's fully diluted valuation, the market cap, if all 2.1 billion tokens were already in circulation, is roughly $3.84 billion. But only about 16% of all PRL has been mined so far. The other 84% is still to come. That's a lot of future supply that the market will need to absorb over time. None of this means the project is fraudulent or broken. The research is real. Together, AI is a real partner. The main net is live. But investors who are buying PRL today are largely buying the thesis, the bet that real AI customers will eventually come in and make the useful part of proof of useful work actually true at scale. So why has PRL pumped this hard? There are a few genuine reasons and a few things worth being cautious about. First, the distribution model is unusually fair by crypto standards. No token sale, no VC allocation, no team reserve, no pre-sale. Every single PRL in existence was mined by someone running a GPU. That's the same model Bitcoin launched with in 2009. In a market where most new tokens give 20% to 40% to insiders before anyone else can participate, this stands out. It creates a genuine community of miners who have skin in the game and strong incentives to promote the project. Second, the AI narrative. Pearl sits at the intersection of two of the biggest themes in crypto right now. AI and proof of work alternatives. AI tokens have had massive runs throughout 2025 and 2026. PRL is one of the very few projects with actual research credentials and a live working main net in this space. Third, token 2049. The project got significant attention at one of the biggest crypto conferences of the year. Word of mouth at events like token 2049 can move markets on it. on. Fourth, pub decks listings. Entropy acquired the PRL perpetuals code for 500 hype e and launched perp trading, bringing leverage speculation into the mix and significantly increasing volume and price discovery. But there is a fifth factor worth naming honestly. Margin trade, a smaller exchange, has been running aggressive promotional campaigns, including paying PRL rewards for social media posts promoting the token. That venue has a direct financial interest in PRL going up, it earns fees on every trade. Some of the narrative momentum is organic. Some of it has been manufactured. PRL has been one of the most debated projects at token 2049 and across crypto social media this October. The conversation breaks cleanly into two camps and both sides have credible voices. Brian Breslow, whose fund No Limit Holdings sponsored the PRL Connect event at token 2049, published a lengthy research piece calling PRL a potential next generation Bitcoin and setting a long-term price target above $30. He calls the thesis Heresy, deliberately. His core argument, AI companies are already spending hundreds of billions on GPUs. Pearl lets some of that spendingman a hard-capped coin as a by-product. The fund is long and public about it. Jeremiah Rogers calls PRL an industrial by-product. Data centers will mine it as a side effect of running AI like a factory selling its scrap. He is bullish, but his entry point is specific. He wants to buy at $1.50 lower after FPH chips. His logic, the upgrade closes the gap between mining and real AI workloads and that's the moment the useful work claim becomes most defensible. Today's price, he says, is already above his buy zone. One widely shared analysis compared the total value of all GPUs currently securing PRL, the hardware backing the network, to the coin's market cap. The argument, PRL trades at about twice its hardware backing while Bitcoin by the same logic trades at 77 times. That framing makes PRL look undervalued relative to Bitcoin's own hardware premium. During a September AMA, CEO Omri Weinstein added fuel. He said that if just 15% of GPUs use their AI profits to buy PRL, that alone would soak up all new supply at a valuation 10 to 30 times higher than today. Matthew Siegel, head of digital assets research at Vanneck, pointed out the technical crack at the center of the useful work claim. The original algorithm did not stop miners from running dummy matrices and dumping the coin. Nothing in the protocol can distinguish a miner doing real AI inference from a miner generating random numbers. That is not a small objection. It is the core tension of the entire project. Hashrate Index was blunter. Most of the useful work is, in their words, inference nobody bought. Alia research added another layer. There is no market where anyone actually pays in PRL. The together AI discount is funded by rewards that shrink every block, and it all rests on one partner. An engineer named Vadim publicly listed seven problems with the protocol. Bresla responded to most of them. The one that didn't get a clean answer? AI demand is bursting. It spikes and drops. A gaming card mining a coin 24 hours a day doesn't care about that. So in the quiet hours, when no real inference is being requested, the junk miners keep going, and the useful work claim quietly evolves. And then there is the broader scoring of the project as it stands today. The near zero overhead useful work concept? Incredible. Real cryptographers, published research, and together AI shows it can run. Mining today being useful, not enforced, and the evidence says mostly not. Fair launch, no preming in the code, true. But 35% of all PRL was mined in the first six days by wallets nobody can identify, backed by a 16z or dragonfly, false. That rumor came from their partners following the account on X, and a 16z hosting a research talk, not an investment. And that brings us to the question neither side has fully answered. Who is buying two million dollars of new PRL every single day? At today's emissions rate, roughly 1.2 million new PRL enter circulation every day. At $1.83, that's approximately 2.2 million in new supply the market has to absorb. Every single day, just to hold the price flat. The bull case requires that answer to be. AI companies and inference buyers. The bear case says, for now, it is mostly speculators and miners selling to each other. So what's coming next that could move things? The first catalyst is the FP8 hard fork. PRL announced in September 2026 that it is planning a hard fork to upgrade mining from integer based computation to FP8, the floating point format that modern production AI models actually use. This is important because it closes the gap between pearl mining and real AI workloads. If successful, it makes the useful work claim more credible. The second is a potential inference marketplace. At an AMA in September 2026, founder Omri Weinstein discussed building a marketplace where AI inference could be purchased directly and priced in PRL. This is the missing piece that would truly connect mining activity to real AI demand. If PRL becomes the currency you spend to buy AI compute, the token has genuine utility beyond speculation. Weinstein also raised the idea of a token burn mechanism where PRL spent on inference is destroyed rather than recirculated, which would reduce the emissions pressure we talked about. Third, more AI partners. Together, AI is one company. The story changes significantly if pearl can sign a second, third, or fourth major AI infrastructure partner. Each new partner would mean more real inference workloads, more real demand for PRL, and more credibility for the useful work claim. Fourth, exchange listings. The token currently trades on safe trade, big on, and perp dexes. A listing on a tier one exchange, finance, coinbase, OKX, buybit, would bring in an entirely new wave of liquidity and visibility. Pearl is genuinely one of the more interesting projects to come out of 2026. The idea is real. The founder is credible. And the no-pre-sale, pure mining distribution is the kind of thing that the crypto community has been asking for more of. But the honest question, the one that will determine whether dollar PRL at its current valuation makes sense is whether the useful work actually happens at scale. Right now, it's one partnership, a lot of speculative mining, and a roadmap of plans. That's a bet on the future, not a reflection of the present. Watch the FP8 hard fork. Watch whether a second major AI partner signs on. Watch whether the inference marketplace launches and whether real AI customers start paying PRL for compute. Those are the signals that will tell you whether this is a new category of crypto asset, or a very well designed AI-shaped mining token. We'll be watching it closely. Like and subscribe for more breakdowns like this. We'll see you in the next one.

Podcast Summary

Key Points:

  1. Pearl Research's PRL token rose roughly 10X from its bottom in a few months and was heavily discussed at Token2049, initially trading only on the little-known exchange Safe Trade.
  2. Pearl is a Bitcoin-forked Layer 1 blockchain using "proof of useful work," replacing arbitrary SHA-256 puzzles with noisy General Matrix Multiplication (GEMM) tied to AI inference, with proofs wrapped in ZK-Snarks to protect input privacy.
  3. The project launched with no token sale, pre-sale, VC allocations, or airdrops; all PRL is GPU-mined, with a 2.1 billion supply and halving block rewards.
  4. Founder Omri Weinstein is a theoretical computer scientist whose 2025 peer-reviewed paper with Elon Komagotsky underpins the claim that mining proofs can be extracted from matrix multiplication at near-zero overhead.
  5. Together AI is the sole named production partner, offering a discounted Gemma 431B inference endpoint subsidized by PRL emissions, demonstrating a working integration.
  6. Critics, including Hashrate Index and VanEck's Matthew Siegel, argue most mining is "AI-shaped proof of work" using dummy matrices with no real AI customers, since the protocol cannot distinguish real inference from random computation.
  7. Daily emissions of roughly 1.2 million PRL (~$2.2 million) create heavy sell pressure, and the fully diluted valuation of ~$3.84 billion rests on only ~16% of supply being mined so far.
  8. Upcoming catalysts include an FP8 hard fork to align mining with production AI workloads, a potential PRL-priced inference marketplace with token burn, additional AI partners, and tier-one exchange listings.

Summary:

Pearl Research and its token PRL have drawn intense attention after a roughly 10X price move and heavy buzz at Token2049. The project is a Bitcoin-derived Layer 1 that swaps arbitrary proof-of-work puzzles for "proof of useful work": miners run noisy matrix multiplications (GEMM), the same math AI inference requires, and wrap proofs in ZK-Snarks so private inputs stay hidden. Its credibility rests on peer-reviewed research by founder Omri Weinstein, published code, a live mainnet, and a genuine partnership with Together AI, which offers discounted Gemma inference subsidized by PRL emissions. Its fair launch—no pre-sale, VCs, or airdrops—has strong community appeal.

However, the central claim remains unproven at scale. Analyses suggest most mining involves dummy computations with no real AI customer, making PRL closer to AI-shaped proof of work than genuinely useful work. Daily emissions of about 1.2 million PRL (~$2.2 million) create constant sell pressure, and only ~16% of the 2.1 billion supply has been mined, leaving a large overhang against a ~$3.84 billion fully diluted valuation. Bullish voices like Brian Breslow see a next-generation Bitcoin, while skeptics note the single-partner dependency and lack of PRL-denominated demand.

Key catalysts to watch are the FP8 hard fork, a PRL-priced inference marketplace with possible token burns, additional AI partners, and tier-one exchange listings. Ultimately, PRL is a bet that real AI demand will eventually validate useful work at scale—not a reflection of the present.

FAQs

Pearl aims to replace Bitcoin's arbitrary proof-of-work puzzles with 'proof of useful work,' where miners run real AI computations (matrix multiplication) instead of meaningless hashing.

Inference is the process of an AI model generating a response to a prompt, which requires a GPU to perform millions of calculations in a data center.

Pearl is a layer 1 blockchain forked from Bitcoin that uses noisy GeMM (General Matrix Multiplication) operations instead of SHA-256 hashing, with proofs wrapped in ZK-Snarks to protect private inputs.

Omri Weinstein, a theoretical computer scientist and academic, founded Pearl Research Labs and published the mathematical foundation in a 2025 academic paper.

Together AI is Pearl's only named production partner; it runs an exclusive endpoint offering a discounted Gemma model, with the discount subsidized by PRL mining rewards.

Critics like Hashrate Index argue most mining involves dummy computations with no real AI customer, making it 'AI-shaped proof of work' rather than genuinely useful work at scale.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.