Pearl’s Useful Work Test

Pearl’s pitch is Bitcoin without the waste. Bitcoin spends its security budget on hashes that are useless outside the chain, while Pearl’s miners do matrix multiplication that runs AI models, which in theory allows a GPU to earn an AI customer’s fee and a PEARL block reward at once.

Pearl runs cuPOW built on GPU matrix multiplication. Miners choose what to multiply, so the protocol proves only that the math was done, not that it served any AI requests, a dilemma at the core of the PEARL investment thesis.

Pearl is designed to be duplex mined (PEARL + AI work) on top of existing AI workloads at low overhead, but the network rewards the volume of multiplication and not what was multiplied. That favors synthetic miners that only multiply random numbers for the sake of mining PEARL. A real AI server mines at a fraction of a dedicated miner’s speed, as most of its AI work is the wrong shape to count towards Pearl mining. 

The upcoming FP8 hard fork aims to switch mining to a standard format AI companies already use on Nvidia’s Hopper and Blackwell data center chips. This would allow more recent open weight models like Llama and DeepSeek to be mineable without being rebuilt for Pearl, and would allow partial mineability of the latest generation of Mixture of Experts models. 

However, with few outside adopters and the upgrades not yet shipped, Pearl Research Labs is trying to make its own platform the best test of the thesis. On October 7, it was named one of six launch providers on Liquid Inference, a new marketplace where providers compete on price for every request. This is the clearest case study for Pearl trying to make useful mining work in practice.

Controversially, Pearl Research Labs is also raising money to scale that test. On October 25, it closed about $55M in equity and a $75M vehicle to build GPU capacity. I view this as a bullish sign for the thesis, since the lab, the team best placed to make the technology work, is now actively trying to prove it themselves and not waiting for outside providers.

There have been three key critiques in regards to the raise, but they are largely overblown in my view: 

1. The lab sells AI for dollars and keeps any PEARL it mines, so its investors can profit even if PEARL falls. But PEARL is meant to be a SoV like Bitcoin, so buybacks would start a classic fight over how much of the lab’s revenue goes to shareholders vs token holders. The lab raised its own money to build a business that earns separately. That is the separation the SoV narrative actually needs, and the raise allows the lab to not sell the PEARL it may have for funding. 

2. The lab has written every fork and therefore controls the network and would cater to itself. But a fork only sticks if the majority of hashrate runs it and the lab’s $75M fleet would add only 2% to 3%. More importantly, the network won’t survive unless someone actually proves useful mining works, so the lab pushing to prove it outweighs the risk of its control. 

3. The lab has never disclosed how much PEARL it holds and the whitepaper says the low starting difficulty was partly meant to fund the team behind it. But the raise was cash for shares and not a token sale, and while we can’t tell which addresses are the lab’s, 98% of the coins mined in launch week have never moved. Although we note more transparency around their current holdings would help the market give more confidence around the entity. 

The bull case is that a SoV’s premium follows its security, and that AI providers mining Pearl would add both security and credibility while issuance slows. That makes the FP8 upgrade the key catalyst, as if it succeeds, AI servers should start becoming a material percent of the network. 

The three catalysts to watch, in order: 

1. The upgrades ship successfully and allow more of an AI model’s math mine, while cutting the slowdown on a GPU serving real traffic to near 0%. 

2. The share of blocks doing real AI work rises to a material share of the network (from ~1% today), increasing hashrate and the SoUW narrative. 

3. An AI provider passes Pearl mining income on to customers as a discount allowing it to gain market share. This would be the single most bullish development for PRL, pushing rival AI providers to mine PEARL just to keep up.

Until then, Pearl stays a network of gaming cards doing math no customer actually uses (just another GPU mined coin), and its holders will have to absorb ~$610M of new coins over the next year (84% of supply growth). If the network stays dominated by gaming card miners, it would make this overhang difficult to digest, as their cost curve sits about a tenth of the PEARL price today, so they can keep selling profitably even if the price continues to fall. So without legitimate AI work to bootstrap its security and adoption, Pearl is a smaller, less secure and faster-inflating Bitcoin, with little reason for new buyers to step in. 

— Sam

Pearl’s network hash rate is around 73 EH/s, nearly triple its late-August level, and has spiked above 100 EH/s, solid growth for an early PoW network.

All of finance is coming onchain and the institutional heavyweights are headed to DAS London this November this 10-11th.

Grab your tickets for unmatched access to the leaders bringing institutional finance onchain.