Grayscale's October 23 research note examines specific infrastructure roles within crypto and artificial intelligence rather than treating AI tokens as a monolithic category. The analysis focuses on three distinct functions: who supplies training data, who finances hardware, and who receives payment for computing capacity.
Research head Zach Pandl identifies DATA, USD.AI, and Pearl as examples of different infrastructure roles. DATA works on contributed training data, USD.AI focuses on GPU-backed lending, and Pearl attempts to combine AI computing with blockchain rewards. These represent selected examples of infrastructure models rather than independently proven outcomes across the sector.
Different Models Serve Different Customers
Each business model addresses a distinct market need. Data networks require customers willing to license datasets. Financing protocols need borrowers capable of servicing debt. Computing networks require useful workloads and measurable delivery mechanisms. Grouping these under a single AI-token label can obscure their fundamental differences.
USD.AI's documentation makes the credit element explicit, describing loans secured by GPUs and associated cash flows. This creates traditional underwriting questions in a blockchain distribution format, including equipment depreciation risk, customer concentration, and collateral recovery processes. While a blockchain record can improve claim visibility, it does not automatically resolve these financial questions.
Infrastructure Operations Versus Financing
The distinction between operating activity and financing arrangements also appears in mining and AI disclosures from companies like Bitdeer. Machines, funded capacity, and realized customer revenue are related but not interchangeable measures.
Token Rights and Underlying Activity
Successful infrastructure does not automatically deliver equal benefit to associated tokens. USD.AI's system separates the non-yield-bearing USDai instrument, the income-accruing sUSDai instrument, and CHIP governance—each with different rights and exposures. Naming a protocol alone does not identify which specific exposure a token holder receives.
Data networks illustrate this distinction. While token incentives can fund initial contributor activity, sustainable demand depends on whether buyers find data lawful, relevant, and unavailable elsewhere. Token incentives may finance early adoption without proving customers will continue paying once incentives decline.
Similarly, tokenized-asset growth requires examination of the specific rights attached to each token. Consumer applications built on AI infrastructure still depend on users returning because the product itself is useful.
Adoption metrics, unit economics, and enforceable claims will ultimately provide stronger evidence that AI infrastructure models are becoming durable businesses.
