Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)
Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)
Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)
Caitlin Kalinowski, a veteran hardware leader from Apple, Meta, and OpenAI, discusses the transition from digital AI to physical AI, focusing on robotics, manufacturing, and the geopolitical implications of hardware supply chains. She shares insights from her work on iconic products like the MacBook Air and Meta's Quest, and offers a critical look at the current state and future of humanoid robots, AR glasses, and the hardware industry.
Kalinowski explains that VR's failure to achieve mass adoption stems from social awkwardness and technological limits, but its underlying technologies like SLAM are now crucial for robotics. She warns of an impending memory price shock due to AI demand, advising startups to pre-buy components to survive. While humanoid robots are overhyped and still unsafe for mass deployment, specialized robots will dominate manufacturing. She emphasizes that hardware's next frontier is physical AI, but supply chain bottlenecks, especially for actuators and magnets, are critical hurdles. Drawing from her experience, she advocates for defining core goals early, prioritizing the hardest design parts, and ignoring customer feedback for truly novel products. Kalinowski also reveals she left OpenAI due to disagreements over decision-making speed and governance, and outlines a hiring strategy that blends generalists, experienced builders, and AI-native engineers.
00:00
00:00
Shift from AI behind keyboards to physical AI.
02:35
02:35
Social awkwardness of wearing a headset is a key barrier
04:55
04:55
Orion's 70-degree field of view is a glimpse of the future.
11:34
11:34
The next frontier is hardware, robotics, and the physical world.
14:30
14:30
Humanoids are not yet ready for mass deployment
16:14
16:14
Supply chain dependencies hinder scaling humanoid robots
17:34
17:34
Actuators and magnets are foundational bottlenecks; without them, robots cannot be built.
23:33
23:33
Cost asymmetry in defense will be solved by engineering culture.
24:50
24:50
Robots can cause more damage than chatbots
26:51
26:51
Apple treated hardware as a first-class citizen
30:10
30:10
Small details reveal larger operational discipline
31:39
31:39
Core goal drives all design decisions
38:23
38:23
Clear KPIs enable quick decisions.
40:03
40:03
First model proved CNC manufacturing possible
41:01
41:01
Praises modern MacBook keyboards.
41:43
41:43
Customers can't envision what doesn't exist yet.
44:46
44:46
Pre-buy memory to survive price spikes
49:31
49:31
A single unavailable component can cause catastrophic redesigns.
52:53
52:53
Off-the-shelf parts are preferred for prototyping, custom for mass production.
57:33
57:33
LLMs lack understanding of physical properties like friction and weight.
1:00:27
1:00:27
Humanoid robots are overhyped
1:03:05
1:03:05
Proprietary CAD data is valuable IP that companies won't share.
1:06:23
1:06:23
Non-verbal cues and intent foster connection
1:11:33
1:11:33
Autonomous cars lack human gestures, making them appear rude
1:14:35
1:14:35
War will drive significant robotics innovation.
1:15:38
1:15:38
I couldn't continue with the lack of guardrails.
1:20:50
1:20:50
Hire AI natives who think differently about AI.
1:23:42
1:23:42
Think bigger, 100x or 10,000x
1:27:27
1:27:27
A miscommunication about a 0.15 mm tolerance caused a critical issue.
1:32:34
1:32:34
Stay present, inspired by a branching paths image.
