I invest in deep tech, and I write about how to act when you cannot know.
Those two things are the same job.
How I think
Most of my work reduces to a single question: how do you commit to something you cannot verify?
A founder tells you his margin will hold for three years. A price chart shows you three cycles and invites you to call it a law. A research paper reports a finding from a sample too small to bear it. In every case, the evidence runs out well before the decision does — and the decision still has to be made.
I’ve come to think the answer is not better prediction. It is structure.
Concretely, that means a few habits I apply everywhere — to due diligence, to my own portfolio, to my writing:
Separate fact from assumption from judgment. Most bad reasoning is three different kinds of claim wearing the same coat. Pull them apart and half the errors become visible immediately.
Read the parameters, not the conclusion. When someone hands me a model, I’m not looking at the output — I’m looking at the settings. A discount rate is a private theory of risk. A growth assumption is a founder’s real opinion of his own moat. People edit their conclusions. They rarely edit their parameters.
Audit your own model the way you’d audit someone else’s. This is harder than it sounds, and I was slow to learn it. The auditor reads parameters; the designer writes them. In the act of writing, you cannot see what you are writing.
Keep a reserve. There is a region of any problem you cannot see into — not temporarily, but structurally. The only honest response is not an argument about unknowability. It is setting something aside for the case where your entire framework is wrong. A belief you have paid for is the only belief you actually hold.
What I do
I work at a venture capital firm, investing across both primary and secondary markets — early-stage deep tech on one side, public equities and crypto on the other.
Focus areas: semiconductors and advanced packaging, AI infrastructure, hard tech, and the industrial supply chains underneath them. I hold a BEng in electronic engineering, which is less a credential than a set of intuitions — I find it hard to invest in something whose physics I can’t reason about.
How I run diligence: I care more about the shape of the evidence than its volume. A number that appears in a founder’s pitch, in his shareholder report, and in a lead investor’s independent account is worth more than ten numbers that appear once. Where the sources disagree is where the information actually is. I’d rather have four contradictory accounts than one clean deck.
What I’m skeptical of: single-point dependency of any kind. Customer concentration. A thesis with one supporting argument. A model with one scenario. I don’t trust anything that only works if one thing stays true — including my own judgment.
Research
I hold an MSc in strategic marketing, and I’m preparing to begin a PhD in management.
My research concerns strategic redirection in Chinese early-stage deep-tech ventures — how and why these companies change direction, and who actually drives that change. My working proposition is that redirection is not a founder’s decision but an outcome of negotiated belief revision among founders, investors, technical teams, and lead customers — and that the locus of influence shifts depending on what kind of uncertainty the venture is facing.
The path here has run from devices, to markets, to organizations — and the question sits at the intersection of all three. I arrived at it from the investor’s seat rather than the library, which I think is both the strength and the risk of it.
Writing
I write essays, mostly about epistemology and the practice of decision-making. They tend to start from something concrete — a model I built, an argument I lost — and work outward.
The recurring theme is the one at the top of this page: what to do at the edge of what can be known. Not as a philosophical puzzle, but as an operational problem, because I have to solve it before Monday.
I’m not trying to be right in these. I’m trying to be corrected. If you think something I’ve written is wrong, that is the most useful thing you can send me.
Elsewhere
I ride — road cycling, seriously enough that the recovery data matters. I read across a wider surface than my work requires, and I think that’s load-bearing rather than recreational: the ideas that have changed how I invest mostly came from outside investing.
I work in Chinese and English, and I think slightly differently in each.
A note on how I engage
I’d rather be argued with than agreed with, and I mean that structurally, not as a pose.
Dialogue can only test whether an idea is coherent. It cannot test whether it corresponds to the world — a theory that repels every objection can still be entirely false. So agreement tells me very little, and the people who have been most valuable to me are the ones who did not adapt to me.
If you disagree with something here, that’s the beginning of a useful conversation.