Working Paper 27
Visible Money, Invisible Counterparties: The Financing Ecosystem of the AI Infrastructure Buildout
October 2026
Abstract
This paper maps the financing of the current AI data center buildout from primary records: county land records in more than fifteen counties, SEC bulk datasets (22.0 million Form N-PORT holding rows and 216,921 Form D issuer rows), issuer filings, appraisal rolls and entity registries, assembled into a 1,495-document archive covering 29 campuses. Three results stand out. First, of $1,165.00B in uncommenced lease obligations across seven issuers, $38.85B (3.3 percent) can be attributed to a named counterparty from public records; the unattributed remainder is structural, because leases are not recorded instruments and none of the seven issuers files a data center lease as an exhibit. Second, the debt layer is documented in size where the lease layer is not: a $1,400M bank facility behind Galaxy Digital’s Helios campus is sized from the credit agreement filed as an SEC exhibit; QTS carries $4,646.2M of indexed deeds of trust in one Virginia county; hyperscalers bought land for cash (Amazon, about $3,080M of deed consideration across two Virginia counties) and record no mortgage debt, while operators carry the recorded debt. Third, holder visibility is regime-specific: SEC-registered funds report 36.5 percent of the largest project bond issue and 3.33 percent of the second, and the insurer layer sits behind a paywall this study reports but does not cross.
Keywords
data center finance, project finance, off-balance-sheet leases, Rule 144A, Form N-PORT, private credit, land records