Astrid Wilde · founder of Inheritance HQ / Inheritance AI (robot-training data for physical AI) and a public-markets investor who posts holdings on X. A founder, not a professional analyst — robotics companies going public are her customers (a conflict she states), and her company profits from AI data demand, so read the physical-AI views as informed but interested.
Held — her memory position, front-run "before everyone figured out that we were sold out for the next 4 years"; sized larger because Nintendo offsets it.
Software isn't dead: big customers pay for maintenance and liability, not code, so when the software cost is immaterial "these customers are not going anywhere." No position stated.
Inheritance AI / Inheritance HQ (private — her company)
Her own private company (founder): packages human task footage ("human priors") into robot-training data for AI labs. Talks her book on "demand for data continues to be underestimated."
Her once-in-a-lifetime case study: a future top US cloud provider dumped by indifferent ex-Russian-exposure holders — "know who your counterparty is" plus a durable data-centre trend.
Held (Sep 2026). Every hyperscaler C-suite sees "no cap to the demand" for AI compute, so "you can comfortably buy NVIDIA" — a bet on buyer psychology holding.
Cited with Google: ~20% modeled data-centre returns crowd out other borrowers ("Why would you buy bonds?"); AI-generated ads upend advertising. No view on the stock.
Private. She'd wait for the IPO rather than own proxies; asks whether the marginal value of more intelligence justifies future compute spend ("probably not" malinvestment).
In one line: Wilde runs a two-rule process — front-run a large, durable change before consensus catches up, then do nothing — and applies it to AI as a chain of bottlenecks (crypto miners → compute, power, memory), with the next one she sees being physical manufacturing capacity, down to sensors. She holds NVIDIA and SK Hynix with a Nintendo hedge and a lot of cash, but declines to build a public robotics portfolio because the direct bets are still private (grounded in the 2026-SEP-11 Value Hive appearance).
Two rules. "The thing that I've done for the last six years is front run earnings" — not a $0.10 beat, but "a very large change… that's going to last for a long time… before consensus has caught up to it." "Thing 2 is just sit around and wait for thing one to happen."
AI as a sequence of bottlenecks. Crypto miners "before they converted to high-powered compute," power "before everybody figured out that that was the bottleneck," memory "before everyone figured out that we were sold out for the next 4 years." Next: "the biggest bottleneck is manufacturing capacity" — sensor vendors now demand 100–1000-unit orders with 3–9-month lead times, and "this hasn't made its way yet to public markets."
Less research, same returns. From 6–10 hours a day to "maybe 15 minutes a day" with "the same kind of historical CAGR." Ideas come pre-filtered from a network built by being publicly online; she passes on almost all of them.
Know your counterparty. Nebius — a future top US cloud provider dumped by holders of frozen Russian exposure — is her defining trade: "whenever… you know who your counterparty is, you're probably going to make a lot of money."
Bet directly or wait. Derivative vehicles add management and operating risk; she'd wait for an OpenAI IPO rather than own SoftBank or Oracle, and says the interesting robotics is private (Fanuc and ABB are the only real public robot makers).
Non-humanoid physical AI. The future is narrow robots that "do one job… really well" (Matic, the dishwasher), trained not by solving problems but by supplying enough varied data — "the bitter lesson." Demand for that data "continues to be underestimated."
Software survives; niche AI thrives. "Money moves at the speed of trust": large customers pay Salesforce for maintenance and liability, and vertical AI firms (Cognition, Harvey) owning one job will "do exceptionally well."
AI capex crowds out other borrowers. The labs model at least ~20% near-contracted returns on data centres, so spending continues and "the price of lending for other use cases is going to continue to go up… Why would you buy bonds?"
Commodities: regime change, but short holds. Western resource demand has "completely reverse[d] course" and isn't priced in, yet she avoids long commodity exposure because high prices summon new supply and automation will cut extraction costs — "never… bet against human ingenuity."
Transcripts
One dated page per appearance — each has its stock table, talking points, and the saved transcript. Newest first.