The new clearinghouse risk sits in the AI stack

Concentration risk in finance used to mean a handful of clearinghouses or payment rails. The next version of that problem sits one layer up: the foundation models and cloud providers that almost every bank and insurer now route their AI work through.

The credit rating agency Moody's, which rates the creditworthiness of banks and insurers, frames this as "systemic dependency." That is a deliberate category break from generic AI risk. A model hallucinating a loan decision is a single-firm problem. A shared dependency on a few model and cloud vendors is a market-structure problem, the same shape as a clearinghouse outage or a single cloud region going dark. The failure modes are not new in kind, only in the layer they sit on: an outage at one provider could cascade across firms and sectors, a small vendor set could extract rents over time, and a panic in one product line could become deposit flight across many.

The offsetting levers matter and are real. Banks hold proprietary data, have decades of contract negotiating muscle, and are testing open-source models and partnerships. None of that dissolves the structural read. The AI risk in banking is not the models. It is that everyone is running on the same ones.