Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)

May 17, 2026 Episode Page ↗
Overview

Caitlin Kalinowski, a hardware leader from Apple, Meta, and OpenAI, discusses VR's foundational tech for robotics, the coming memory price shock, hardware supply chain vulnerabilities, AI's impact on physical world engineering, and lessons from tech giants like Steve Jobs, Mark Zuckerberg, and Sam Altman.

At a Glance
10 Insights
1h 39m Duration
27 Topics
5 Concepts

Deep Dive Analysis

VR's Evolution and Impact on Robotics

Future of AR Glasses and Physical AI

Rise of Robotics and Hardware Popularity

Challenges of Hardware Development

Driving Trends in Robotics and Physical AI

Current State and Safety of Humanoid Robots

Supply Chain Dependencies for Robotics

Geopolitical Implications of Hardware Supply Chains

AI Safety Concerns with Physical Robots

Lessons from Apple's Hardware Excellence

Building Hardware Programs from Scratch at Meta

Critical Principles for Hardware Development

The MacBook Air Design Story

Customer Feedback in Hardware Innovation

Impending Memory Price Crisis for Hardware

Complexity of Robot Components and Supply Chain Verticalization

Balancing Off-the-Shelf vs. Custom Components

AI's Impact on Hardware Engineering

Humanoids vs. Dedicated Robots for Use Cases

Future of Robots Building Robots

Designing Robots for Human Connection

Robots in the Home and Future Outlook

Reasons for Leaving OpenAI

Hiring Exceptional Hardware Teams

Lessons from Steve Jobs, Mark Zuckerberg, and Sam Altman

Hardware Development Failure Story

Lightning Round

SLAM

SLAM (Simultaneous Localization and Mapping) is a technology that allows a device, like a VR headset or a robot, to understand its position in space relative to a simulated or real world by using cameras and depth sensors. This enables accurate positioning and movement tracking.

Actuator

An actuator is the motor in a robot that converts electrical power into motion, typically a rotating rotor with gearing that powers limbs, heads, or fingers. It's a foundational technology for robotics and drones, often relying on magnets.

EVT

EVT (Engineering Validation Test) is a critical stage in hardware development where the product is compiled for the first time using final components and materials, made on mass production tools. Its purpose is to validate the engineering design before proceeding to mass production.

AI Native

AI Native refers to individuals, typically younger engineers, who have grown up using AI tools so extensively that it's inherently integrated into their problem-solving and engineering processes. They approach tasks with an AI-first mindset, often leading to different and faster solutions.

Western Canon

The Western canon is a traditional list of influential works in Western culture, often including ancient Greek and Roman classics, that are considered essential for a broad understanding of literature and intellectual thought. It represents a distillation of foundational texts.

?
Why didn't VR take off despite advanced hardware?

VR's social aspect of having a headset covering the face was a significant barrier, but the underlying technologies developed for VR (like SLAM and depth sensing) became foundational for robotics and physical AI.

?
What are the surprising challenges in hardware development?

Unlike software, hardware can only be 'compiled' (redesigned and built) a few times, making it critical to be conservative, conduct thorough reliability checks, and account for high part variance across components.

?
Why are robots and hardware suddenly popular?

There's a growing realization that AI's capabilities behind a keyboard will eventually saturate, making the physical world (robotics, manufacturing, industrialization) the next frontier for AI application.

?
Are humanoid robots ready for mass deployment?

Humanoid robots are currently advanced prototypes, not yet ready for mass deployment due to safety concerns (especially with large, strong humanoids near people), manufacturing challenges, and complex supply chain dependencies for critical components like actuators.

?
How does Apple approach hardware excellence?

Apple's dedication to hardware excellence involves a methodical process where every design decision, even internal ones, is deeply considered to understand core purpose, leading to simple, high-quality outcomes.

?
How does AI impact hardware engineering today?

AI is currently impacting hardware engineering in high-level planning, strategic analysis, information gathering, and rapid spreadsheet creation, but it's still in early stages for core CAD design due to limitations in understanding physical properties like friction or contact pressure.

?
Why are humanoids not the answer for most robot use cases?

Humanoids are not ideal for most manufacturing or industrial tasks, which are better suited for dedicated, specialized robots designed for specific, repetitive functions (e.g., screwing parts) rather than generalist human-like forms.

?
What makes a robot feel human and connected?

To feel human and connected, robots need to exhibit complex nonverbal cues, acknowledge human presence, appear non-threatening and soft, be reactive, and clearly transmit their intent physically (e.g., looking before turning).

?
Why did Caitlin Kalinowski leave OpenAI?

Caitlin left OpenAI due to disagreements with the speed of decision-making, governance, and the lack of defined guardrails around the announcement of a Department of War deal, despite having many friends and respecting the company.

?
What qualities are important when hiring hardware teams for new industries?

For new industries like AI and robotics, look for strong generalists who can adapt skills, a mix of experience in building new things and scaling existing ones, and 'AI-native' young talent who use AI from the ground up.

1. Pre-buy Memory for Hardware Projects

Advise startups to pre-buy memory to ride out anticipated price spikes and supply chain disruptions, as memory prices are expected to double due to AI demand and constrained supply.

2. Design Hardest Parts First

When building hardware, prioritize designing the most challenging or riskiest components first, rather than starting with familiar parts, to identify and mitigate potential failures early.

3. Focus Iteration on User-Facing Parts

Allocate more iteration and design effort to the parts of a hardware product that customers touch or interact with most, such as trackpads or keyboards, to ensure high quality and reliability.

4. Act Immediately on Known Tasks

In hardware development, execute known tasks and necessary actions immediately, even if more time seems available, because unexpected issues and surprises will inevitably arise, consuming future time.

5. Define Hardware Goals Early and Stick to Them

Establish clear, written goals (KPIs) for hardware products early in development and minimize changes, as hardware is less adaptable to mid-development shifts, impacting timing and feature sets.

6. Embrace AI for Engineering Strategy

Utilize AI tools for high-level planning, information gathering, and rapid spreadsheet creation in engineering, even if it’s not yet capable of core CAD design, to speed up strategic processes.

7. Cultivate AI-Native Talent

Actively seek out and learn from young, AI-native engineers (typically 20-21 years old) who approach problem-solving with AI from the ground up, as they can significantly accelerate engineering processes.

8. Verticalize Supply Chain for Resilience

Consider verticalizing the supply chain, like Tesla and Starlink, by bringing more component manufacturing in-house to better adapt to supply chain shocks and component shortages.

9. Use Off-the-Shelf Components for Prototyping

Leverage off-the-shelf components whenever possible during prototyping phases to quickly demonstrate functionality and iterate, while ensuring they can eventually fit the final industrial design.

10. Design Robots for Social Interaction

When designing robots that interact with humans, incorporate nonverbal cues, intent signaling (e.g., looking before turning), and a non-threatening, reactive appearance to build trust and avoid creepiness.

In hardware, we only get to compile our code, quote, unquote, like four or five times. Total.

Caitlin Kalinowski

If you think about that as a frontier, you can see the end of that tunnel. Now, I don't know when it's going to be again, but we can see that that's going to saturate at some point, or at least people think it will. And when that happens, the next frontier is hardware.

Caitlin Kalinowski

You really want me to be like, this isn't going to work and this isn't going to work and this isn't going to work and just like be, be like kind of worried about all these things going wrong.

Caitlin Kalinowski

Sam is really good at saying, why not more? Why not 100x or 10,000x? You're thinking too small.

Caitlin Kalinowski

For Steve, the bar he held for the company, for technical talent and for excellence, was not wavering.

Caitlin Kalinowski

If a robot looks before it turns and then goes, it's much less alarming.

Caitlin Kalinowski

I think there's probably more change in war than there is in consumer electronics in the next two years, for example.

Caitlin Kalinowski

Hardware Development Principles

Caitlin Kalinowski
  1. Define goals early and stick to them, minimizing changes throughout development.
  2. Design the hardest or riskiest parts of the product first to address potential failures.
  3. Prioritize and boost iteration on components that customers touch or interact with the most.
  4. Act immediately on known tasks and necessary actions, as unexpected issues will inevitably arise.
plus and minus three sigma or more
Part variance in hardware When two parts go together, you must account for the smallest version of one and the largest version of the other.
going to double, probably
Expected increase in memory prices Due to supply chain shock and AI demand, though the timeline is uncertain.
between 50 and 150 parts
Number of major parts in a robot (e.g., Matic) If counting major assemblies; can be thousands if counting all small components like caps off PCBs.
from 200 people to 10 now
Reduction in staff on advanced manufacturing lines In the most advanced manufacturing facilities, human labor has been significantly reduced.
7 million views
Views on Caitlin Kalinowski's tweet about leaving OpenAI Her tweet explaining her departure from OpenAI.
58,000 likes
Likes on Caitlin Kalinowski's tweet about leaving OpenAI Her tweet explaining her departure from OpenAI.