Pearl Research

@devlordone @brandonjcarl 🤷‍♂️

@brandonjcarl wouldn't be surprised if this is all due to @prlnet

Crazy price action in H200 cloud pricing – up 56% in 3 days. What is unusual is that the H200 is suddenly trading higher than the B200, a superior GPU. It’s not crazy to think that a fund could bid up supply in an illiquid tight market at a cost of $50K a day to engineer a short-term move in much more liquid stocks.

@Eli5defi Thank you for the thorough review. We're hosting AMA session on Discord tomorrow, you are welcome to join if you any any questions.

If your favourite CT says there’s nothing interesting to talk about because the market is dead, hit that unfollow button. They’re washed and can’t do proper research. There are a ton of hidden gems in this space, for example, @prlnet, which is flying under everyone’s radar. Here’s my thesis breakdown in under 30s: - ➠ What Pearl Is Pearl is a Layer 1 blockchain that forks Bitcoin and replaces SHA-256 with arbitrary matrix multiplication as its proof-of-work mechanism. In other words: it's a Bitcoin clone where miners run AI math instead of useless hashing puzzles. The same GPU that answers a ChatGPT query can also mint the token and secure the network at the same time. - ➠ The Problem It Solves Right now, AI runs on a winner-takes-all stack. You pay OpenAI, Anthropic, or a hyperscaler for compute. They own the datacenter. They keep the margin. The electricity gets burned, the model gets smarter, and the upside belongs to whoever owns the rack. The rest of the world has no way to own a piece of "AI compute" as a category. There's no neutral asset that tracks the cost of GPU work the way gold tracks the cost of pulling metal out of the ground. Meanwhile, Bitcoin has the opposite problem. Bitcoin's proof-of-work is brilliant security but the actual computation (SHA-256 hashing) produces nothing. Miners burn billions of dollars of electricity solving puzzles whose only purpose is to be hard to solve. Pearl's bet: what if the puzzle that secures the chain was also the puzzle the rest of the economy already wants solved? - ➠ How it Works Every modern AI workload (inference, training, fine-tuning, embeddings) boils down to one operation repeated billions of times: matrix multiplication (MatMul). Two grids of numbers multiplied together. That's what GPUs are built for. Pearl replaces Bitcoin's SHA-256 with arbitrary matrix multiplication as the proof-of-work. The math paper from Komargodski (Hebrew University) and Weinstein (Hebrew University / ex-Columbia, ex-NVIDIA, ex-VAST Data) proves you can make any MatMul cryptographically verifiable with overhead of 1 + o(1) - math notation for "negligible." Concretely: if running an inference normally costs you 100 GPU-cycles, running the same inference in a way that also produces a valid Pearl block costs you maybe 100.5 cycles. The security tax is rounding error. So the same kilowatt-hour does three jobs at once: ❶ Answers the customer's AI query (the useful work) ❷ Earns the miner a PRL block reward (the monetary incentive) ❸ Secures the Pearl blockchain (the network effect) This is what "proof-of-useful-work" (PoUW) means. Bitcoin's PoW is wasteful by design. Pearl's PoW is productive by design. - ➠ Why The Token Has A Price Floor The whitepaper boils down to the rule miners already follow: cost to mine = value of reward If PRL trades below production cost, miners shut off and difficulty falls. If it trades above, miners pile in and difficulty rises. Standard PoW. What’s different is the “cost.” Bitcoin’s is electricity. Pearl’s is useful GPU compute, the scarcest resource right now. So PRL’s price floor tracks the global AI compute cost curve. As long as H100/H200 supply is tight and capex keeps rising, producing PRL gets more expensive. - ➠ Tokenomics ▸ Total supply: 2.1B PRL forever ▸ Premine: nothing ▸ Team allocation: nothing ▸ Public sale: never happened ▸ Emission curve: smooth daily decline (not Bitcoin's halving cliffs) Bitcoin halvings hit miners hard every four years: revenue drops 50% overnight, inefficient miners wash out, and the network reshuffles. Pearl smooths this into a daily decline. Revenue fades gradually, making capacity planning far more predictable. Roughly half the supply is issued in the first four years, then issuance slows toward the cap. - ➠ The Current State ▸ Mainnet launched April 27 ▸ Hashrate 3.56 EH/s (~0.4% of Bitcoin) ▸ ~100s blocks (difficulty lagging miner growth) ▸ H100+ only: no gaming GPUs, no hobbyists ▸ OTC: $0.30 → $0.70 in 3 days ▸ Cost to produce 1 PRL: $0.30 → $0.51 ▸ Subsidized compute pool: $0.004 → $0.0052 per TH-hr (5 days) No CEX or DEX, no CoinGecko/CMC. token is live, but there’s basically no chart for retail to trade. - ➠ Future Lookout Right now people buy $PRL because mining costs are rising while supply is fixed. The real pivot is the planned compute marketplace (not live yet): - Pay for inference in PRL, not dollars (vs AWS) - Reserve/settle GPU contracts on-chain in PRL - Verify compute cryptographically, not via cloud billing If it ships and works, PRL shifts from “token tracking compute costs” to “unit of account for AI compute.” USDC did this for DeFi, one standard token replaced messy, fragmented settlement. PRL is aiming to be the same for compute. FYI, Render, Akash, and even Bittensor are pursuing similar ideas, but no one has won yet. - ➠ Institutional Locked-In ▸ a16z crypto hosted a talk on the Pearl paper. The page got pulled (now 404s) and archived. Pulled pages usually signal private conversations starting, not interest dying. ▸ Stanford Crypto Seminar hosted Weinstein. Same room that hosted the foundational talks for Bitcoin and Filecoin. ▸ Rob Hadick at Dragonfly (the GP who led Polymarket, Ethena, and Rain) publicly follows the account. ▸ Yonatan Sompolinsky (Kaspa founder, the most important PoW theorist outside Bitcoin Core) is in the paper's acknowledgements. That's not a polite footnote. That's the closest thing to a blessing the PoW research world produces. No VC has publicly invested. Given the no-premine / no-team-allocation structure, there's nothing for VCs to buy at a discount. They'd have to mine like everyone else. - ➠ Bear Scenario for PRL → The price jumped 133% in over-the-counter trading, but there weren’t many real trades. because trading was thin, we don’t know if the price would hold up when lots of buyers and sellers show up. → The “compute marketplace” is mostly a plan right now. if they don’t actually build it or it doesn’t work, then the project is basically just another bitcoin-like coin with a different math story. → They say there was “no premine” (no big stash made for insiders at launch), but no independent audit has confirmed it. for now you’re taking the team at their word until someone analyzes the distribution. → The low mining/pool rates are low because they’re being subsidized. once the subsidies end, the costs and incentives might change and the current economics might not hold. → The core crypto claim (verifying large matrix multiplications securely, even if someone is trying to cheat) is genuinely novel, but the efficiency claim is proven in theory. it hasn’t been proven in the real world under heavy load and active attacks. - ➠ Final Take Pearl makes AI math do double duty as money math. If that idea works at scale, the token becomes the most economically grounded crypto asset ever shipped, because its production cost is tied to the most demanded resource in the world. If it doesn't work, well, you get a well-designed Bitcoin clone that nobody uses.

Inference is becoming the largest compute market and energy consumer in AI. Pearl turns inference CapEx of hyperscalers into a profit center: every LLM token produced by GPUs can simultaneously generate ¶PRL in parallel. That means users get 2-for-1 economics: useful inference today, and Pearl coins that discount future compute. We’re excited to partner with Together AI to launch the first 2-for-1 inference endpoint offering SoTA LLMs at discounted price, and securing the Pearl network in parallel.

"The real limit to AI is not energy, it is actually cash"

Turning intelligence into currency.

Demand for Pearl is skyrocketing. @prlnet We’ve opened up a compute platform for the people — first come, first served, with limited supply. We’re working hard to expand capacity. Subscribe for updates: compute.pearlresearch.ai

👀

2-for-1 Inference on Gemma4 is now available on Pearl! huggingface.co/pearl-ai/Gemma…

Every era has a resource so fundamental it becomes money. Bitcoin turns energy into currency. Trustless, global and censorship-resistant. AI is doing something even more powerful: it turns energy into intelligence. Deployable anywhere and useful for almost everything. But intelligence has no native financial primitive. It isn't fungible. It can't move outside the dollar system. Pearl changes that. Pearl is the first asset natively produced by AI and natively secured by AI. Every GPU cycle producing LLM tokens can simultaneously mint Pearl tokens with marginal extra electricity, zero wasted compute and one unified primitive. 2-for-1. Sitting atop one of the largest capital expenditures in history, Pearl changes the unit economics of AI. This is what sets Pearl’s breakthrough apart. Previous attempts at useful-work blockchains captured a narrow slice of compute. Pearl's addressable market is every matmul computation on earth which, at current trends, will be the majority of all compute. Bitcoin’s security is competing with AI for energy. Pearl's security scales with AI adoption. Proof of work represents humanity's demand for energy, monetized. Pearl represents humanity's demand for intelligence, monetized. Pearl is now live. pearlresearch.ai

It is becoming increasingly clear that the future economy will be denominated in compute cycles more than in human labor. In a world where AI drives the majority of electricity consumption and GDP, compute is the natural collateral for money: an open, auditable, AI-native currency, produced directly through inference and training. Since the inception of Bitcoin, an outstanding open problem in distributed systems was whether it is possible to implement Proof-of-Work consensus on top of real-world computation, as opposed to useless random hashing. While long considered impossible, last year we answered this question affirmatively. Pearl’s mathematical breakthrough enables every GPU cycle powering AI systems to simultaneously produce a native digital currency: ¶PRL. What this means is that the hundreds-of-billions (and soon trillions) of dollars of compute being deployed for AI workloads will double--for effectively free--to secure Pearl's Proof-of-Work chain; All the properties of Bitcoin, but secured as the by-product of AI inference and training, i.e., by the native operation of GPUs: matrix-multiplication (GEMM). Pearl changes the unit economics of LLMs, which are are fundamentally non-fungible, and will shift a portion of the wealth generated by AI back to users – who drive production, model improvement and demand, yet currently capture none of the upside of the AI era. We’ve spent the last year turning this “2-for-1” breakthrough into a working infrastructure, building from the linear algebra down to the CUDA kernels, alongside world-class mathematicians and low-level engineers. Today, we’re excited to announce that the Pearl Network is ready, and will soon support state-of-the-art LLM serving, through vLLM and SGLang plugins. Running AI workloads on Pearl transforms AI compute from a sunk expense into an AI-native asset, anchored directly to the production of intelligence. If you’re interested/skeptic or ideally both – we’ve published our next tranche of open problems as a collaborative Polymath challenge – containing math, systems and economics questions we’re grappling with next. We invite you to tear it down, prove it or propose better implementations: pearlpolymath.com. #AIMoney #ProofOfInference