Grayscale Looks Beyond AI Tokens To The Data And Credit Behind Computing

Reason to trust

Strict editorial policy that focuses on accuracy, relevance, and impartiality

Created by industry experts and meticulously reviewed

The highest standards in reporting and publishing

How Our News is Made

Strict editorial policy that focuses on accuracy, relevance, and impartiality

Ad discliamer

Morbi pretium leo et nisl aliquam mollis. Quisque arcu lorem, ultricies quis pellentesque nec, ullamcorper eu odio.

Grayscale’s October 23 research note looks at a more specific crypto-and-AI proposition than a general bet on smarter software: who supplies training data, who finances hardware and who receives payment when computing capacity is used.

Three business models sit behind one broad AI label

In the note, research head Zach Pandl describes DATA’s work on contributed training data, USD.AI’s GPU-backed lending and Pearl’s attempt to combine AI computing with blockchain rewards. These are Grayscale’s selected examples and expectations, not independently proven outcomes for the entire sector.

Each model has a different customer. A data network needs someone willing to license its datasets. A financing protocol needs borrowers who can service debt. A computing network needs useful workloads and a way to measure delivery. Combining them under an AI-token heading can hide more than it reveals.

USD.AI’s documentation makes the credit element explicit. It describes loans secured by GPUs and associated cash flows, with recourse generally limited to the collateral and specified exceptions. Its yield-bearing instrument draws on loan payments and reserve-related income.

That creates familiar underwriting questions in a new distribution format. What happens if the equipment loses value faster than expected? How concentrated are the customers paying to use it? Who can recover and sell collateral if a borrower fails? A blockchain record can improve visibility into a claim without answering all three.

The operating-versus-financing distinction also runs through Bitdeer’s mining and AI disclosures. Machines, funded capacity and realized customer revenue are related measures, but they are not interchangeable.

The token must connect to the underlying activity

Even successful infrastructure does not automatically deliver the same benefit to every associated token. A governance token, a claim on lending income and a token used to pay for a service can have very different rights.

USD.AI’s depositor guide, for example, separates the non-yield-bearing USDai instrument, the income-accruing sUSDai instrument and CHIP governance. Naming the protocol alone does not identify which exposure a holder receives.

The same discipline applies to data. Paying contributors can encourage useful supply, but sustainable demand depends on whether buyers find that data lawful, relevant and difficult to obtain elsewhere. Token incentives can finance initial activity without proving that customers will keep paying once incentives shrink.

Tokenized-asset growth similarly needs to be read through the rights attached to the token. And AI-facing consumer applications still need users who return because a product is useful.

Grayscale’s report provides a set of business models to examine. Adoption, unit economics and enforceable claims will provide the stronger evidence that those models are becoming durable businesses.

Written by the News Desk and edited by Samuel Rae.