Key Takeaways
- • Grayscale's October 23 research note, authored by research head Zach Pandl, argues that crypto-and-AI investment should be separated into distinct business models: training-data supply ($DATA), GPU-backed lending (USD.AI), and AI computing rewards (Pearl).
- • USD.AI's documentation shows its loans are secured by GPUs and associated cash flows, with recourse generally limited to collateral and specified exceptions, which introduces standard credit underwriting questions.
- • Grayscale warns that a governance token, a claim on lending income and a service-payment token can carry very different rights, so holders must identify which exposure they actually receive.
- • The note treats the examples as Grayscale's selected expectations rather than independently proven outcomes, and says adoption, unit economics and enforceable claims will be the stronger evidence of durability.
Summary
Grayscale's October 23 research note examines the intersection of crypto and AI by focusing on who supplies training data, who finances hardware, and who receives payment when computing capacity is used. Research head Zach Pandl frames the analysis around three infrastructure roles, using $DATA, USD.AI and Pearl as illustrative examples. The note presents these as Grayscale's selected examples and expectations, not independently proven outcomes for the sector as a whole.
The report argues that grouping these models under a single AI-token label obscures meaningful differences. A data network depends on buyers willing to license its datasets, a financing protocol depends on borrowers who can service debt, and a computing network depends on useful workloads and a reliable way to measure delivery. Each model therefore has a different customer base and a different set of risks.
USD.AI's documentation makes the credit component explicit. It describes loans secured by GPUs and associated cash flows, with recourse generally limited to the collateral and specified exceptions, and a yield-bearing instrument that draws on loan payments and reserve-related income. This structure raises familiar underwriting questions: how quickly equipment may lose value, how concentrated the paying customers are, and how collateral could be recovered and sold if a borrower defaults. The note observes that blockchain records can improve visibility into a claim but do not resolve these issues on their own. The report also draws a parallel to Bitdeer's mining and AI disclosures, noting that machines, funded capacity and realized customer revenue are related but distinct measures.
Grayscale further cautions that infrastructure success does not automatically benefit every associated token. USD.AI's depositor guide separates the non-yield-bearing USDai instrument, the income-accruing sUSDai instrument and CHIP governance, illustrating that naming a protocol does not reveal which exposure a holder receives. For data networks, token incentives may attract initial supply, but durable demand depends on whether buyers find the data lawful, relevant and difficult to obtain elsewhere, and whether customers keep paying once incentives decline. The note concludes that the underlying business models merit examination, but that adoption, unit economics and enforceable claims will provide the stronger evidence of durable businesses.
The report argues that grouping these models under a single AI-token label obscures meaningful differences. A data network depends on buyers willing to license its datasets, a financing protocol depends on borrowers who can service debt, and a computing network depends on useful workloads and a reliable way to measure delivery. Each model therefore has a different customer base and a different set of risks.
USD.AI's documentation makes the credit component explicit. It describes loans secured by GPUs and associated cash flows, with recourse generally limited to the collateral and specified exceptions, and a yield-bearing instrument that draws on loan payments and reserve-related income. This structure raises familiar underwriting questions: how quickly equipment may lose value, how concentrated the paying customers are, and how collateral could be recovered and sold if a borrower defaults. The note observes that blockchain records can improve visibility into a claim but do not resolve these issues on their own. The report also draws a parallel to Bitdeer's mining and AI disclosures, noting that machines, funded capacity and realized customer revenue are related but distinct measures.
Grayscale further cautions that infrastructure success does not automatically benefit every associated token. USD.AI's depositor guide separates the non-yield-bearing USDai instrument, the income-accruing sUSDai instrument and CHIP governance, illustrating that naming a protocol does not reveal which exposure a holder receives. For data networks, token incentives may attract initial supply, but durable demand depends on whether buyers find the data lawful, relevant and difficult to obtain elsewhere, and whether customers keep paying once incentives decline. The note concludes that the underlying business models merit examination, but that adoption, unit economics and enforceable claims will provide the stronger evidence of durable businesses.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.