Thesis

Anduril was the spearhead. They beat the legacy primes in an open market on price and speed without a government carve-out, and that's the piece you can actually replicate. Ukraine did the other half. When a $500 FPV drone can take out a $5M platform, the cost curve of warfare has flipped, and the DoD can't outspend that asymmetry. They have to out-build it. The culture moved last. Defense went from career-ending to fundable almost overnight, and the best engineers are actually in the room now. I spend most of my time on two things: autonomous systems, where hardware and software are really one problem, and the mission software that turns raw sensor feeds into decisions while they still matter.

Longevity has spent the last decade as a fringe topic for centenarians and biohackers, and the real product category — proactive medicine for healthy people — barely exists. GLP-1s cracked it open. For the first time the public watched a drug move the aging curve instead of just treating a symptom, and the cultural permission to spend on staying biologically younger went from vanity to legitimate almost overnight. Continuous biomarker tracking is getting cheap, AI is compressing the discovery cycle for compounds that target aging mechanisms directly, and the consumer is finally willing to pay for healthspan instead of waiting to be sick. Pharma's century-old loop — wait for disease, sell treatment — is the wrong shape for this market. I spend most of my time on two pieces: the data layer, where continuous biomarkers and AI interpretation turn a single user into a longitudinal trial, and the consumer health companies that own the patient relationship before the legacy system knows the patient is there. The TAM isn't a vertical, it's every human who ages — that's the trillion-dollar piece nobody has priced in yet.

The AI conversation is stuck on the model layer, on the assumption that whoever trains the biggest model wins. That assumption breaks. Foundation model performance is converging, and the cost of serving a token drops every quarter. Capital floods the middle. Margins compress. The interesting positions are at the edges. Below the model, the constraint is physics. I've been tracking silicon photonics for years, well before the AI cycle pulled focus to it — at 800G interconnect speeds, the rack becomes the bottleneck before the GPU does. Above the model, the constraint is integration. A probabilistic system needs an ontology and an evaluation layer that takes a decade to build, and the companies that have it sit in the wrong analyst bucket today. I spend most of my time on two pieces: the physical-layer companies solving interconnect and packaging, and the system-layer companies that own the data on top of whatever foundation model wins. The model is the visible piece. The business is everything around it.

The fintech infrastructure cycle was waiting on two things: regulatory clarity and a payments primitive built for software instead of humans. Both arrived in 2025. The GENIUS Act gave payment stablecoins a federal framework, and the CLARITY Act finally drew the line between what's a security and what isn't — the ambiguity that kept institutional capital on the sidelines for the better part of a decade. Stablecoins are no longer a workaround. They're a sanctioned settlement layer that clears in seconds and costs basis points instead of percentage points. The other half is what gets built on top. Coinbase's x402 protocol revives the dormant HTTP 402 status code so an AI agent can pay for an API call, a piece of data, or a service the same way it makes any other request. That's the missing primitive for agentic commerce. Card networks and ACH were designed around a human pressing a button — they break the moment the buyer is software making thousands of micro-decisions a second. I spend most of my time on two pieces: the issuance and orchestration layer for stablecoins moving between enterprises, and the protocols that let agents transact natively without a human in the loop. The first fintech wave digitized banks. This one builds the rails for a buyer that doesn't have hands.