Caitlin Kalinowski, ex-OpenAI, Meta and Apple, argues that AI saturating screens pushes the next frontier into robotics, manufacturing and industrialization.
Kalinowski says the United States should invest far more in drones than aircraft carriers and must reindustrialise to be safe militarily.
She calls humanoids still-advanced prototypes, saying scale and supply chains are the real problems, since the US lacks strong actuator companies and most robot supply chains went to China, Japan and Korea.
AI's next frontier is hardware, held back by supply chains
As AI saturates what can be done behind a keyboard, the next frontier is physical hardware, and supply chains set the limit. The US outsourced actuators, magnets and batteries over 25 years, so humanoid hype runs ahead of what can be built at scale. Hardware teams get only a few final design iterations, which makes the work slow and conservative.
Takeaways
Physical world is AI's next frontier. AI labs, big tech and startups are realising together that digital AI will saturate. Attention is shifting to robotics, manufacturing and industrialization. Kalinowski adds that the US must reindustrialise for military safety, and expects more change in war than in consumer electronics within two years.
Humanoids are hype and supply chains are the problem. Humanoids remain advanced prototypes. Scale and supply chain come first, and the US lacks strong actuator companies. Actuators, magnets and batteries are bottlenecks after 25 years of outsourcing to China, Japan and Korea. Dedicated manufacturing robots will do most jobs.
Hardware's few final iterations force conservative design. Teams can compile a design only four or five times because each release is final for mass production. They must test for part variance, plan two or three years ahead, and ladder toward a goal six years out.
Component shortages reward pre-buying and verticalisation. Kalinowski expects memory prices to double as data centres absorb supply, and has advised startups to pre-buy. Silicon and displays usually have the longest lead times. Tesla redesigned a circuit board fast when silicon ran short, which shows the value of verticalisation.
VR missed the mass market but fed robotics. Meta spent roughly $10bn on VR. The work taught camera-based positioning, depth sensing and how humans perceive space, which robotics now uses. AR glasses are the likely next step, but Orion's waveguides and MicroLED displays have low yields and high cost.
AI speeds engineering but cannot yet do CAD. AI can route circuit boards, help with planning and speed design work, but it cannot do real CAD. Kalinowski wants a Codex for hardware engineering, which may need new model types. Chinese assembly lines of 200 people may now have 10.
Home robots face a harder sell than cars. A home robot replaces nothing people use today, unlike a self-driving car. Robots must show intent to feel safe, and Pixar and Disney are probably the best at designing emotion and approachability in characters.
Career lessons and leaving OpenAI
She left OpenAI over the speed of decision-making and the lack of guardrails around the Department of War deal. She had hoped for a third path between going along and scorched earth criticism.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
She hires generalists, AI-native engineers and people aligned with the mission. Mission alignment unifies teams when AI researchers and hardware engineers come from different worlds.
Mark Zuckerberg ran Meta well by making decisions at the lowest possible level to keep speed.
A misread camera spec on Quest 2 forced a redesign at Meta, yet the build and shipping stayed on time.