My first years working were spent looking at companies rather than building them. DCFs for pre-IPO names, scouting lists for a deeptech fund, fundraising decks for portfolio companies, the strange arithmetic of a close. It is a good education. It is also an education in a specific kind of blindness.
A spreadsheet will tell you whether a business works at the numbers someone typed into it. It will not tell you whether the thing can be built.
What finance gave me
Three things, and I would not trade them.
A prior on where value accrues. Most companies describe themselves as platforms. Very few are. Underwriting enough of them gives you a nose for the difference between a business with margin and a business with a story about margin.
Fluency in the language of the other side of the table. When a founder talks to an investor, the investor is running a model in their head: market, margin, moat, timing. Having built those models myself means I can help a founder answer the question that is actually being asked, instead of the one they prepared.
Respect for the boring constraint. Unit economics are not a distraction from product. They are a description of what the product is allowed to be.
What it could not give me
Judgment about physical feasibility.
A DCF has no term for yield. There is no cell for “the enclosure will not release from the tool.” When I started running cross-border electronics operations out of Shenzhen — moving six figures of units a month through marketplaces, managing storefronts and the people behind them — the abstraction broke in a useful way. The questions became concrete and immediate: what is the lead time on this part, what happens if the batch is wrong, who do I call at 2 a.m.
Hardware is unforgiving in a way that finance is not. A model can be wrong for a year and still be revised. A tool cannot.
Why AI hardware, specifically
Because it is the part of the AI build-out where the physical constraint is still underexploited. The model layer is crowded and well-capitalised. The layer underneath it — actuators, sensors, thermal paths, power, enclosures, the factories that put them together — is where the next decade of companies will be won or lost, and it is much less legible to capital.
That asymmetry is interesting to me. It is also where I can be useful: I can read the model and I can read the line.
What I actually do now
I host Future Factory, which puts twenty founders inside Shenzhen’s factories for six days at a time. And I founded Illuminious, an NPI engineering firm that sits with the factory rather than across from it.
Both are attempts to shorten the distance between a good idea and a physical product. It turns out that distance is mostly made of information — who to ask, what to ask, and when.
I still think in models. I just don’t confuse them with the thing.