Writing
Three books, no sum.
The thesis
Nobody adds the three together.
AI now sits on all three books of a regulated financial firm at once: the assets that fund data centres, the providers the firm depends on, the risks it insures. Each book is counted by a different team and reported under a different framework.
Book one
Assets
A bond, a loan or a securitisation note that funds the campus.
Held by the investment office. Counted under the name of the issuer.
Book two
Operations
The cloud and the models the firm itself depends on, often served from that same campus.
Held in the register of information. Counted under the name of the provider.
Book three
Underwriting
Construction, property, power, cyber, liability. Five lines of business, one site.
Held by each line. Counted under its own name, for the same site.
One name, three books
Oracle, as a regulated firm saw it in September 2026.
In the asset book
The issuer S&P cut to one notch above junk in July 2026, on negative outlook at Moody’s, with credit default swaps at a record high in September.
In the register of information
A critical third-party provider on the first European list under DORA, and a designated critical third party in the United Kingdom since July 2026.
In the underwriting book
The tenant of the Stargate campus in New Mexico, whose vehicle owns the racks and must insure them for the maximum foreseeable loss under its lease with the county, on a site whose $18 billion construction loan was quoted between 89 and 91 cents on the dollar in September.
Three systems see the same company. None of them tells the other two.
The articles
Three articles on AI and concentration risk in regulated finance, published on LinkedIn between July and August 2026 and kept here in full.