The AiFi Thesisis a research report I wrote with co-authors from Peri Labs and GAIB, published in February 2025. AiFi is shorthand for the collision of AI and finance, and the report's claim was specific: AI compute stops being just another cloud line item and becomes an asset class that capital markets finance directly, the way they finance power plants, ships, and real estate.
Three legs carried the argument:
- Compute as collateral. GPU capacity has measurable utilisation, contracted revenue, and depreciation schedules: the raw material of structured finance. Once underwriting standardises, dedicated debt and yield products follow.
- Capacity demand outruns balance sheets.Hyperscaler capex alone cannot absorb AI's compute demand curve; external capital must be invited in through instruments investors already understand.
- Agents need rails. If software agents transact with each other, they need payment infrastructure, spend controls, and clean records; whoever provides those rails owns the toll booth of the agent economy.
The receipts
The report was published on 28 February 2025. What followed:
- 24 September 2025: Oracle moved to raise billions in corporate bonds explicitly to fund AI cloud infrastructure: compute capacity financed by the debt markets.
- 10 April 2026: Blackstone filed a public registration for a digital infrastructure trust, a retail-accessible vehicle for newly built data centres.
- 7 May 2026: AWS launched AgentCore Payments with Coinbase and Stripe, giving agents first-class rails to pay for APIs, MCP servers, web content, and each other.
Read the report
Download The AiFi Thesis (PDF) ↗ The report is indexed on Google Scholar.
How it connects
The AiFi Thesis is the financing leg of the same worldview behind The State of Edge AI (where the compute lives) and The Economics of Wash Trading (what happens to market integrity when trading meets programmable incentives). The full set of research pages lives in the research hub.