Grayscale's head of research, Zach Pandl, named DATA Network, USD.AI and Pearl in an October 23 note, saying the asset manager is keeping an eye on the three projects as blockchain technology takes on a growing role in the AI buildout. The three sit at different points of that buildout: DATA collects human training data, USD.AI lends against GPU hardware, and Pearl is designed to reward GPU operators whose compute doubles as proof-of-work mining.

DATA: 250 million registrations, but no disclosed licence economics

DATA collects human-contributed training material such as photos, video and voice recordings. According to Grayscale, applications on the network license those datasets to AI labs while rewarding the contributors who supplied them.

The Sui Foundation's October 12 announcement adds detail. DATA's Trace registry, launched in June, records where each contribution came from and the rights attached to it, then settles payment to the contributor. Trace has registered more than 250 million contributions and processes over 3 million receipts a day on average, the foundation said. Applications writing to it, including Kled, Numo and Oto, license data to frontier AI labs and robotics companies.

None of the published figures say how many contributions were actually bought, at what price, or how much contributors received.

The Sui Foundation is also investing $30 million in the DATA Network, a commitment that makes it the network's largest institutional holder, as DATA moves to Sui, a layer-1 blockchain, and Walrus, a decentralized storage network. DATA's own chain will be retired, and DATA tokens will carry over one-for-one, with holder economics and governance rights described as at least equivalent to today's. A transition plan is due before the upgrade. Neither announcement explains how licence revenue reaches DATA holders.

USD.AI: the yield and the ticker sit in different places

Grayscale lists USD.AI under the ticker CHIP and describes the project as arranging loans backed by GPU equipment, with on-chain pools that let participants earn yield. The borrowers are compute operators that need capital before their GPUs pay for themselves — the same squeeze behind the tougher funding market now facing miners moving into AI. In USD.AI's own documentation, however, the yield and the ticker sit in different places:

• USDai: a stablecoin backed one-to-one by PayPal's PYUSD. No yield and no exposure to GPU loans.
  sUSDai: earns the interest. Value tracks GPU loans and reserves and can rise or fall. Not a stablecoin.
• CHIP: Grayscale's ticker. Votes on upgrades and risk settings.

Under USD.AI's sUSDai page, the yield comes from the interest that GPU operators pay on loans, plus income on escrowed capital and reserves not yet lent out. Loans cover up to 80% of a GPU's value. Exits run a 30-day queue funded by loan repayments and reserves, and the token carries credit, liquidity and smart-contract risk.

CHIP's MiCA white paper, named for the European Union's Markets in Crypto-Assets framework, states that the token does not confer equity, dividend or profit rights and does not entitle anyone to payments or interest simply for holding it. Staking CHIP in insurance modules is a separate, optional activity. The same document lists a possible fee switch for tokenholders among future protocol decisions, and USD.AI says an updated white paper is being prepared.

Pearl: a 580% rally and the subsidy behind it

Grayscale says PEARL has risen 580% in the past three months, citing CoinGecko data as of October 22. Pearl wants GPUs to do double duty: run AI workloads and secure a proof-of-work blockchain through the matrix multiplication those workloads already perform, letting operators earn PRL rewards on top of their AI revenue. Grayscale labels the token PEARL while describing rewards in PRL, and Together AI writes it as ¶PRL.

Together AI, an AI cloud provider, launched its first Pearl-powered product, a discounted inference endpoint, on May 15. The discount of more than 25% is offset by the future value of crypto emissions, and the company said that as PRL's price rises, it expects to use more of that value to cut prices further. Customers pay less because newly issued coins cover part of the bill, and the size of the discount depends on what those coins are worth.

A June preprint by researcher Abhinaba Basu challenged the useful-work claim. Measuring a network of roughly 320,000 GPU-equivalents drawing an estimated 112 MW, it concluded that the network “produces zero useful AI computation.” The dominant mining software contained no inference code, and the paper's own miner had 44 shares accepted using random matrices. The study is single-author and has not been peer reviewed, and its measurements were taken in early June — before the three-month rally Grayscale cites.

Follow the money

Grayscale says the projects' long-term relevance will depend on adoption and competitive economics. Its disclaimer notes that the firm and its affiliates may have interests in the assets discussed, and that the projects named do not necessarily represent current or future holdings.

Three disclosures would show whether token holders get a share of what these businesses earn: DATA's transition plan, due before its Sui upgrade; USD.AI's updated CHIP white paper, including whether a fee switch ever reaches a vote; and evidence of how much of Pearl's rewarded work serves paying AI customers.

This article is for informational purposes only and does not constitute financial or investment advice. Project descriptions and economic arrangements can change; lending and digital assets carry risks of loss.