Chokepoints maps the AI compute supply chain: ten layers from raw materials to applications, 17,500 companies, 2,376 dependencies, and the 95 points where the whole system narrows to a critical few. The map is free at chokepoints.ai. The newsletter examines it one company at a time.

What you get

Each issue takes one company from the map and works out what it controls, what it depends on, and whether the market has priced it. Specific numbers, named companies, and a clear view at the end. Short reads and full deep dives, straight to your inbox.

The thesis

The obvious AI winners were settled in private markets years before anyone outside could buy in. For public investors, the returns compounded one layer below the software. The whole market returned 5x since 2013; the single chipmaker at the chokepoint returned 684x. When memory became the constraint, SanDisk won. When electricity did, Bloom Energy won. Scarce, hard-to-replace positions are what markets price, and that is where we look.

Who writes it

Nitanshu and Romil. By day we run the AI team at a well known private equity fund in London; between us, machine learning research at one of the UK’s leading AI labs and years in the AI safety movement since Oxford. Chokepoints started as our own map of how AI plays out in the real economy.

Why we bother

The chokepoints are where the leverage is: the points where this technology can still be steered, slowed, or governed. They are also where the returns are, and that is no coincidence. We would rather you understood this infrastructure than looked away from it.

Subscribe and every issue lands in your inbox. Reply to any email; we read everything.