Agentic AI Backbone The infrastructure agents run on

About

Own the backbone every AI agent runs on

The bet is on the layer underneath. Every agent that runs in production consumes silicon, memory, bandwidth, electricity, and floor space, and someone bills for all of it. That spending is contracted years ahead and is far easier to observe than guessing which model or application wins. The portfolio holds the suppliers of that capacity and the platforms that meter it.

How a name is judged

Each company gets two separate judgments. One is how essential it is to the thesis actually playing out, which changes slowly and only when the company's position genuinely changes. The other is whether now is a good moment to own it, which is allowed to move faster.

Scores are absolute. A name is judged against a fixed standard, never ranked against the others, so a good company is never marked down because a different one had a good week. Every change to a score carries a written reason and the evidence behind it.

How the book is run

Rules that do not move

Long only. No leverage, no derivatives. Limits on any single position and on any one sector. Nothing is removed by a formula.

Decisions a person makes

Every purchase, sale, and removal is proposed with a written case and decided by a person. No trades are placed automatically.

Honest limits

Whether these judgments are any good is measured rather than asserted, and the record is short. Where the evidence is not yet strong enough to claim skill, the honest answer is that it cannot be claimed, and that is the answer given. Published findings carry their sample size.