Angel investing is a lens into the future. After enough checks, patterns stop being abstract.
My book keeps clustering in a few places: AI infra and compute, physical AI and robotics, aerospace and dual-use hardware, and vertical software that actually collects revenue. Different surfaces. Same through-line.
1. Intelligence is becoming infrastructure
The winners aren’t only better models. They’re the picks and shovels — training data, inference, chips, edge compute — that make intelligence cheap, reliable, and close to the work. When AI stops being a feature and starts being capacity, the economics look more like energy than like apps.
2. The hard problems moved back into the physical world
Software ate a lot. What’s left is factories, yards, roads, airframes, warehouses. Robots and industrial automation aren’t “AI demos.” They’re margin, throughput, and labor where the spreadsheet ends. If it doesn’t work in the real world, it doesn’t work.
3. Space and defense are becoming normal markets
Launch, sensing, autonomy, counter-drone — still hard, less exotic. Capital is following capability that can be manufactured, contracted, and reused. Sovereign demand isn’t a narrative. It’s a buyer.
4. Vertical AI without retention is just a story
I’ve gotten more allergic to “AI for X” without a clear buyer, usage, and unit economics. The bar isn’t novelty. It’s whether customers keep paying after the demo.
The meta-lesson for me: the next decade of returns won’t come from smarter chat alone. It’ll come from companies that put intelligence into systems that move atoms, secure borders, and run businesses.
Still optimistic. Just more selective about where the optimism goes.