Astrid Wilde: A Bright AI & Robotics Future

Astrid Wilde runs Inheritance AI, a company focused on capturing human behavioral data and converting it into training data for physical AI and robotics systems. Rather than solving complex tasks directly, the company leverages deep learning's ability to extract patterns from high-quality, varied data. She explains that many tasks can be broken down into simple primitives (like standing, swinging, tracking) that are easier to capture than full end-to-end tasks, though human messiness in raw data requires significant downstream processing to extract meaningful segments.

Wilde distinguishes between two types of robots: deterministic machines like dishwashers that perform singular functions well, and more complex intelligent systems that must handle edge cases and varied circumstances. She's most excited about near-term automation of repetitive menial labor tasks—like robotic vacuums, construction work, and factory assembly—rather than humanoid home robots, which she sees as further away. She believes the real opportunity lies in thousands of niche robotics companies solving specific domain problems rather than a concentration around two or three players.

On the bottleneck question, Wilde identifies manufacturing capacity and sensor availability as the actual constraint, not compute. Eighteen months ago, sensor manufacturers would respond to small purchase inquiries, but now they require orders of hundreds or thousands of units and have multi-month to multi-year backlogs. This creates advantages for well-funded companies but also suggests the physical AI market is genuinely heating up.

Regarding AI consolidation, Wilde is less concerned about concentration of large language models because companies like Anthropic and OpenAI are building broadly applicable tools that enable smaller teams to solve narrow, specific problems at drastically lower costs than previously possible. She argues the leverage available to individuals to affect change continues to increase even as capital concentrates, and software itself isn't dead—it's evolving into specialized domain solutions backed by human trust and liability assumption.

On public market investing, Wilde employs a two-part strategy: front-running material business inflections driven by large, durable trends before consensus catches up, and waiting patiently for opportunities to materialize. She spends minimal time—about 15 minutes daily—analyzing markets, noting that additional hours of research don't improve returns. Her portfolio includes positions in memory stocks (SK Hynix, Nintendo), data center infrastructure plays, and occasional special situations like Nebius, which traded at disconnected valuations due to Russian market closure dynamics. She's interested in variant perception opportunities in struggling consumer companies like GameStop and Peloton that trade at extremely depressed multiples if their business trajectories shift positively.

Wilde emphasizes that being terminally online since age 12 has created a powerful network effect where interesting investment ideas are pre-screened and brought to her by knowledgeable contacts. She's a self-described info vore who reads constantly but through people networks rather than traditional media.

On commodities and physical demand, Wilde agrees that AI and robotics will drive massive increases in resource extraction demand, reversing decades of flat-to-declining resource demand in developed economies. However, she's cautious about direct commodity exposure because historical patterns show that rising commodity prices incentivize innovation in extraction efficiency, bringing prices back down. She notes that in mining specifically, many startups have been funded in the last two years to automate resource discovery, extraction, and transportation.

The conversation concludes with discussion of the marginal value of intelligence—whether OpenAI and Anthropic can continue extracting rents from marginal improvements, or whether open-source models and on-device inference eventually make frontier models less economically valuable. Wilde believes national security competition between the US and China, combined with hyperscaler C-suites' conviction that there's no upper limit to AI demand, will continue justifying massive compute investments. She notes that Claude Hopkins, the early 20th-century advertising pioneer, exemplifies how communication and persuasion reshape society, and she's skeptical that current AI companies are particularly effective at advertising their capabilities.