Overview of Why we’re at the beginning of the AI hardware boom with Caitlin Kalinowski
In this episode, Lenny Rachitsky talks with Caitlin Kalinowski, a veteran hardware leader from Apple, Meta, and OpenAI, about why AI is pushing the next major technology wave into the physical world. Caitlin argues that as AI saturates keyboard-based digital work, the next frontier is robotics, drones, manufacturing, and AR glasses. The conversation covers why hardware is so hard to build, what’s still missing for mass-market robots, how supply chain constraints and memory shortages are shaping the industry, and what leaders need to do to build successful hardware teams.
Key Themes and Takeaways
AI is moving from software into the physical world
- Caitlin sees a “dawning realization” across labs and startups that AI’s digital acceleration will eventually saturate.
- Once that happens, the next frontier becomes:
- robotics
- manufacturing and industrialization
- drones and autonomy
- AR glasses and other physical interfaces
- She expects major change in warfare and defense technology sooner than in consumer electronics.
VR was a stepping stone, not the destination
- VR did not become mass-market, but it produced important breakthroughs:
- SLAM and spatial tracking
- depth sensing
- understanding how humans perceive 3D visual data
- These technologies are now directly useful in robotics, autonomous vehicles, and AR.
- Caitlin views VR as part of a longer technological arc leading toward robotics and physical AI.
AR glasses still look like part of the future
- She believes AR glasses will matter because people do not want to be glued to their phones.
- The main blockers today are:
- microLED and waveguide readiness
- yield and cost
- input methods for public/social use
- Orion, Meta’s advanced AR prototype, gave a glimpse of what immersive AR could feel like once the hardware matures.
Why Hardware Is So Hard
Hardware “compiles” only a few times
- In software, teams can compile and debug constantly.
- In hardware, teams may only get a handful of major iterations before mass production.
- Once a product ships, changes are slow, expensive, and sometimes impossible without a full redesign.
Tolerances and part variance make mass production difficult
- Hardware teams must account for manufacturing variance across thousands or millions of units.
- The challenge is not just making one working prototype, but making something reliable at scale with high yield and low returns.
One missing component can derail everything
- If a key part is unavailable, a whole product can stall.
- Caitlin emphasized the fragility of modern hardware supply chains, especially for:
- silicon
- RAM/memory
- actuators
- magnets
- A missing component can force a redesign of the board, enclosure, or even the whole system.
Robotics, Humanoids, and Safety
Humanoids are exciting, but not yet ready for scale
- Caitlin views humanoid robots as advanced prototypes, not finished mass-market systems.
- Her biggest concern is safety, especially with:
- large, strong robots operating near people
- high-impact limbs and actuators
- unpredictable behavior in human environments
Softer, lighter robots are safer
- She explained that impact risk depends on both mass and compliance.
- Robots with softer, lighter arms and inward-shifted mass are safer around humans.
- She cited 1X’s Neo as an example of a more safety-conscious design approach.
Specialized robots may be more useful than general-purpose humanoids
- Caitlin does not think one humanoid form factor will solve everything.
- Her view:
- use dedicated robots for specific jobs
- reserve humanoids for long-tail tasks where human-shaped movement is actually useful
- Examples:
- factory assembly robots for repetitive tasks
- construction robots
- electrical work robots
- logistics robots
Supply Chain, Reindustrialization, and Geopolitics
The robotics supply chain is a strategic vulnerability
- The robotics stack depends on materials and components that are often outsourced:
- magnets and raw materials
- actuators and motors
- silicon and memory
- Caitlin stressed that the U.S. and other countries need more independence in these layers.
Reindustrialization is a national security issue
- She argued the U.S. should re-industrialize significantly to protect itself militarily.
- Her view is that geopolitical alliances can shift, so countries should not assume supply stability or permanent alignment.
Drones are changing warfare faster than traditional systems
- Caitlin believes military innovation is moving much faster than consumer electronics.
- She highlighted:
- cheap drones
- rapid iteration in Ukraine
- 3D printing and fast battlefield updates
- She echoed the idea that investing in drones may matter more than investing in aircraft carriers.
AI Hardware: What’s Changing Right Now
Memory prices are becoming a major bottleneck
- Caitlin warned that memory prices are under severe pressure due to AI demand.
- Her advice to startups:
- pre-buy memory if possible
- prepare for supply shocks
- She expects continued price volatility and possible doubling or worse, depending on timing and demand.
AI is already helping hardware teams, but not replacing them
- AI is useful for:
- planning
- competitive research
- spreadsheets and analysis
- early prototyping support
- It is not yet doing real day-to-day CAD or electrical/mechanical engineering at a high level.
CAD is a major near-term frontier
- Caitlin wants “codecs for engineering” — AI tools that can truly help with hardware design.
- She believes current models are still weak at:
- geometry
- friction
- pressure
- material behavior
- tolerance stacks
- She expects future world models to become foundational for engineering AI.
Lessons for Building Hardware Teams
Start with clear goals and don’t change them constantly
- In hardware, changing specs late is costly.
- Teams should define priorities early, such as:
- cost
- weight
- size
- resolution
- speed
- Once those goals are set, stay disciplined.
Solve the hardest and riskiest problems first
- Great hardware architects identify the bottleneck early.
- Don’t start with the easy visible parts; start with the part most likely to fail.
Focus on the surfaces people touch most
- Hardware quality should be concentrated where users interact most:
- trackpads
- keyboards
- input controls
- comfort points
- These require more iteration than secondary components.
Move fast, even when it feels like you have time
- Her rule: do the necessary work now.
- Hardware always takes longer than expected, so delay is dangerous.
Hire for a mix of specialists, generalists, and AI-native talent
- Caitlin looks for:
- strong generalists
- specialists who know how to build and scale
- AI-native younger talent
- She believes “cracked” young builders who use AI fluently are valuable teachers for older teams.
Leadership Lessons from Apple, Meta, and OpenAI
From Apple: excellence, discipline, and first principles
- Apple taught her:
- hardware is a first-class product concern
- every detail matters, even invisible ones
- leaders should ask why a design exists, not just how it works
- She cited Steve Jobs’ insistence on high standards and meticulousness.
From Meta: speed plus structure
- She praised Meta for:
- clear decision-making
- strong technical reviews
- pushing decisions as low in the organization as possible
- She learned the importance of operating a hardware team with clarity and rigor.
From Sam Altman: think much bigger
- Caitlin said Sam Altman constantly pushed:
- “Why not 100x?”
- “Why not 10,000x?”
- The lesson: ambitious leaders should force teams to think at much larger scale than they initially would.
Notable Failure Story: Quest 1 Camera Spec Mismatch
What went wrong
- During Quest 1 development, a spec mismatch around camera placement caused a late-stage problem at EBT, right before mass production.
- The issue was a different interpretation of tolerance ranges.
How it was fixed
- The team redesigned part of the structure:
- two cameras were fixed relative to each other
- the other cameras were allowed to float
- They adjusted materials and preserved launch timing.
Why it mattered
- It showed how brittle hardware development can be.
- Even a small misunderstanding in millimeters can trigger a major redesign late in the process.
Caitlin’s View on the Future
What the next five years may look like
- More AI capability in knowledge work
- More visible robots and autonomous systems in public
- Faster advances in drones and defense technology
- Continued slow but real movement toward useful robots in homes and workplaces
What won’t happen as fast as people think
- Caitlin does not expect tens of millions of humanoid robots in five years.
- She believes supply chain, manufacturing, and reliability challenges will slow mass deployment.
Final Message
Caitlin’s core message is optimistic but grounded: AI is about to reshape the physical world, but the winners will be the companies and countries that can combine ambitious vision with rigorous hardware execution. She encourages builders to use AI tools daily, imagine the future deliberately, and help design the world they want rather than waiting for it to happen to them.
Recommended Mindset for Builders
- Be clear about the mission
- Solve the hardest problems first
- Design for scale and safety
- Treat supply chain as a strategic constraint
- Use AI aggressively, but realistically
- Build with the future in mind, not just the present