Pearl: Bitcoin for AI?

Key Takeaways

• We’re bullish on the first real shot at useful mining. Pearl keeps Bitcoin’s fixed supply but swaps useless hashes for the matrix math that runs AI, so one GPU could earn AI revenue and PEARL at once. The concept is still early, since the chain only proves that matrix math was done, letting fake AI math count just the same and dominate the network. We estimate <1% of blocks match a real AI model, and consumer-grade GPUs run ~82% of hash rate.
• The FP8 upgrade and Pearl Research Labs’ push are the catalysts. Currently, a GPU serving AI mines at most a third as fast as a dedicated miner and gives up valuable AI work to do it, so each coin’s opportunity cost is roughly 10x what a dedicated miner pays in power. FP8 would make that opportunity cost competitive with synthetic miners’ power costs, which Pearl’s whole use case depends on. All eyes are on the newly funded Pearl Research Labs, likely the next real test of whether mining while serving AI can work economically.
• Whoever runs the software decides whether a GPU mines. Mining lives inside the AI software, so most Bitcoin miners turned AI hosts can’t switch on Pearl mining, since their tenants run the GPUs. That makes serverless providers, which run both the GPUs and the software, the natural first adopters of Pearl, with GPU marketplaces and neoclouds next if their renters and tenants see a reason to mine.
• The bull case is reflexive. If AI providers mine PEARL, they would quickly build the security necessary for a store of value (SoV) premium, and set Pearl apart as a SoV backed by useful work. Taken further, an inference provider that mines PEARL can use its rewards to cut its AI prices and therefore win more customers through routers like OpenRouter, forcing rivals to start mining PEARL to compete on price (more mining backs the SoV case, lifting PEARL’s price and funding even deeper cuts to AI prices).
• Dilution is the near-term test. The next year adds ~84% to supply (~$540M), and each day’s new coins are worth ~28% of daily trading. Although the hash rate has nearly tripled since late August, synthetic miners pay about a tenth of PEARL’s price in power for each coin, so their cost base gives the price little support. That makes PEARL a high-risk, high-reward hold. Until FP8 ships and works as designed, little sets PEARL apart from other GPU-mined coins, but at about 3% of Zcash’s market cap and 0.04% of Bitcoin’s, we think getting positioned ahead of it outweighs the dilution.

Introduction