Key Moments
Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)
Lenny's PodcastKey Moments
Nearly indistinguishable virtual humans cost $1M to create, while advancements in AI are outpacing human comprehension in medical diagnosis.
Key Insights
The development of VR technology has been crucial for advancements in robotics, enabling a better understanding of spatial positioning and human perception of visual data.
The cost of memory, referred to as a 'meteor,' is expected to double for consumer hardware and robotics due to increased demand from AI and constrained supply chains.
Hardware development requires approximately four to five 'compiles' or major, often year-long, revision cycles, in stark contrast to software's daily iteration.
Humanoid robots are considered advanced prototypes, with significant safety considerations, such as lighter and softer designs, needed before widespread adoption, especially given concerns about their impact energy.
The supply chain for robotics is complex, with critical dependencies on raw materials like magnets and foundational technologies like actuators outsourced to countries like China, Japan, and Korea over the last 25 years.
AI is beginning to impact hardware design, with potential for rapid design in CAD and component selection for PCBs, but current models lack the understanding of physical properties like friction and contact needed for deep engineering.
Virtual reality's foundational role in physical AI
The extensive investment in virtual reality (VR) has yielded critical technological advancements that are now foundational for robotics and physical AI. Innovations such as SLAM (Simultaneous Localization and Mapping) for spatial positioning, depth sensor applications, and understanding human visual perception in space, though developed for VR, are directly transferable to robotics. These technologies are essential for robots to navigate their environments, understand distances, and interact with objects. Caitlyn Kalinowski, with her background at Meta, highlights that while VR gaming may remain a niche, the underlying learnings have accelerated progress in fields like robotics and autonomous vehicles, effectively bridging the gap between simulated and real-world interaction.
The promise and challenges of augmented reality glasses
Augmented reality (AR) glasses are seen as a key component of the future, offering a way to maintain social connections while accessing information without constantly looking down at a phone. Products like Meta's Orion, though perhaps ahead of their time due to immature microlleds and waveguide technology, demonstrated the potential of a 70-degree field-of-view binocular display that provides an immersive experience. Challenges remain, particularly in the input methods for AR glasses, focusing on quiet and discreet communication. However, the integration of AR displays, which can be turned on when needed, signals a continued evolution towards seamlessly blending digital information with the physical world.
The surprising difficulty of hardware development
Transitioning from software to hardware presents unique and often underestimated challenges. Unlike software, which can be compiled and debugged daily, hardware development undergoes only four to five major 'compiles' or revision cycles throughout its entire lifecycle. This means that design decisions, reliability checks, and testing must be exceptionally thorough from the outset. Furthermore, hardware engineering must account for 'part variance'—the natural variations in physical components that can affect assembly and performance. Solving for the last half-percent in precision ensures high yield, profitability, and minimal returns, a stark contrast to the more iterative nature of software.
The dawning realization: AI's next frontier is the physical world
A significant trend in the AI landscape is the growing consensus that digital advancements are approaching saturation, leading to a natural progression towards the physical world. Labs and startups are recognizing that as AI models become more capable in the digital realm, the next major frontier for innovation and impact lies in robotics, advanced manufacturing, and industrialization—essentially, the 'sensing layer' of the real world and the ability to manipulate it. This shift is driving a surge of interest and investment in hardware and robotics, fields previously considered less 'sexy' than software development, signaling a potential boom in this sector.
Humanoid robots: Advanced prototypes with safety hurdles
Humanoid robots are a particularly captivating area of development, appealing to our innate attraction to human-like forms. Companies like Tesla (Optimus), Figure, and Neo are at the forefront. However, Kalinowski emphasizes that current humanoid robots are largely advanced prototypes. Significant safety concerns exist regarding large, powerful humanoids operating near people. Designs that incorporate 'softer' materials and inward-pulling mass distribution, like those seen in 1x Neo, aim to mitigate risks by reducing impact energy and compliance. The path forward involves proving functionality, then focusing on cost reduction, manufacturability, higher yields, and, crucially, safety before they can be widely deployed.
Supply chain vulnerabilities: The critical reliance on outsourced components
The global supply chain for robotics is a complex web with significant vulnerabilities, particularly concerning critical components that have been outsourced over the past 25 years. Raw materials like magnets, essential for actuators (motors that convert electrical energy into motion), are often processed and integrated into actuators in countries like China, Japan, and Korea. This reliance creates a bottleneck; if these components, especially foundational ones like actuators and silicon chips, become unavailable, it can lead to catastrophic redesigns and production delays. The recent memory price surge, with costs potentially doubling, exemplifies this risk, highlighting the need for re-industrialization and greater domestic independence in manufacturing these essential parts.
AI's burgeoning role in accelerating hardware design
While AI's impact on software is well-established, its influence on hardware engineering is just beginning. Current AI tools can assist with tasks like generating surfaces or point clouds, and are showing promise in routing printed circuit boards (PCBs) and basic component selection. However, true CAD (Computer-Aided Design) capabilities, which require understanding complex physical properties like friction, weight, and contact pressure, are still emerging. The development of 'world models' and specialized engineering codecs is anticipated to revolutionize hardware design, enabling more rapid prototyping and iteration. The significant challenge lies in acquiring proprietary CAD data for training these advanced AI models, suggesting that open-source communities and hobbyists may lead initial breakthroughs.
Focusing on core goals and human connection in hardware development
Successful hardware development hinges on clearly defined goals and an unwavering focus on what truly matters to the user. This includes prioritizing cost, size, weight, and key performance metrics like display resolution in VR. Lessons from Apple's product philosophy, such as the 'back of the cabinet' principle, emphasize meticulous attention to detail even in unseen components, forcing a deep understanding of the product's purpose. For consumer robotics, creating a sense of human connection requires non-threatening, soft, and reactive designs that acknowledge human presence and intent, drawing parallels with Pixar's character animation. The ultimate goal is not just to build functional devices, but to create products that seamlessly integrate into human lives, enhancing experiences rather than just performing tasks.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●Books
●Concepts
●People Referenced
Hardware Development Best Practices
Practical takeaways from this episode
Do This
Avoid This
Common Questions
The speaker believes that the capabilities of AI in the digital world, like behind a keyboard, will eventually saturate due to their rapid acceleration. The next frontier for AI's growth is therefore in the physical world, driving innovation in robotics, manufacturing, and industrialization.
Topics
Mentioned in this video
Mentioned as a source of 100,000 drones and a country to be wary of regarding potential military threats and supply chain dependencies.
Mentioned as a current example where military technology, particularly drones, is changing incredibly fast due to rapid innovation and 3D printing in warfare.
Mentioned in the context of the war in Ukraine and the rapid changes in military technology and drone usage.
Described as a legendary builder who held an unwavering bar for technical talent and excellence at Apple, influencing design decisions like the 'back of the cabinet' philosophy.
Discussed as a successful leader who ran Meta very well, particularly in how the technical side and hardware organization operated with clear objectives and decentralized decision-making.
Described as a leader who consistently asks 'why not more?' pushing for 100x or 10,000x thinking, encouraging larger ambitions in projects.
The guest on the podcast, a highly accomplished hardware leader who worked at Apple, Meta, and OpenAI, specializing in AI hardware and robotics.
Mentioned as someone who envisioned a threat of 100,000 drones from China, highlighting a need for preparedness.
Co-founder of Oculus, who Caitlin considers a friend and agrees with on the need for increased drone investment over aircraft carriers in military strategy.
Presented an interesting talk at Stripe Sessions about OpenClaw's ability to influence real-world actions like changing a self-driving car's destination.
An Apple executive who discussed learning the 'back of the cabinet' philosophy from Steve Jobs, emphasizing detailed design decisions.
Mentioned as an Apple colleague who taught Caitlin the 'ruthless efficiency' of acting immediately on necessary tasks in hardware development.
The author of 'Mrs. Dalloway,' a book Caitlin enjoys and recommends.
Praised for his ability to define values for engineering trade-offs (e.g., gram of weight vs. cost) and for verticalizing Tesla's supply chain, allowing for better adaptation to shocks.
The CEO of Madic Robotics; asked a question about memory prices.
Mentioned as a brilliant roboticist who might be working on world lab companies, indicating the direction of solving complex physical understanding challenges.
Caitlin's former colleague and friend, who praised Caitlin's brilliance in hiring exceptional teams, particularly for zero-to-one opportunities.
Meta's CTO, who was involved in technical decisions and understood complex trade-offs, making the hardware organization run efficiently.
Mentioned as an Apple colleague who taught Caitlin the 'ruthless efficiency' of acting immediately on necessary tasks in hardware development.
A poet who created a challenging reading list of the Western canon, which inspired Caitlin to delve deeper into ancient texts.
Identified as world's best at design work involving emotion, intent, approachability, and engagement, which is crucial for making robots feel less creepy and more connected to humans.
Identified as world's best at design work involving emotion, intent, approachability, and engagement, which is crucial for making robots feel less creepy and more connected to humans.
Caitlin's former employer, where she was part of the original unibody MacBook Pro teams and technical lead on MacBook Air and Mac Pro; known for its best-in-class hardware development and first-tier hardware citizen approach.
Caitlin led the AR glasses and VR hardware teams at Meta, including products like Oculus Rift and Quest; the company renamed itself to lean into VR.
Caitlin's most recent employer, where she helped build their robotics and hardware division from scratch, but ultimately left due to disagreements over governance and the Department of War deal.
Mentioned as a comparison for Work OS, described as 'Stripe for enterprise features'.
A company recognized for being ahead in humanoid robot development, alongside Tesla and 1X Technologies.
An AI system that exhibits unpredictable and sometimes problematic behavior, such as sharing personal email addresses or changing Tesla destinations, highlighting AI safety concerns in a physical context.
A brand that makes interesting clothes based on new material science, which Caitlin recommends following.
A podcast sponsor that helps companies automate compliance and risk management (SOC 2, ISO 27001, HIPAA) to earn and prove customer trust.
A company making robot vacuums (Maurizio Nardi's product) that the speaker owns and loves due to its effective and reliable operation.
Mentioned as a source for code data, with labs reportedly buying pre-2021 GitHub repos for AI training before AI significantly impacted human-written code.
Caitlin was part of the original unibody MacBook Pro teams and worked as the thermal lead on the first model.
Caitlin was the technical lead on the MacBook Air, specifically the wedge-shaped version that achieved higher volume sales after the initial Manila envelope prototype.
Caitlin was the technical lead on the cylindrical Mac Pro during her time at Apple.
A VR device that Caitlin helped design as part of the VR hardware team at Meta, with Quest 2 notably achieving success through price reduction via redesign.
Mentioned as an impressive piece of hardware that, despite its magical experience, has not caught on for mass adoption, indicating a broader struggle for VR/AR.
Believed to be part of the future of technology, allowing information access while maintaining social connections.
Mentioned as a product that taught the importance of the social aspect of wearables, particularly how having something on one's face can hinder social interaction.
A humanoid robot created by Tesla, mentioned as one of the few companies ahead in humanoid robot development.
A humanoid robot highlighted for its significant safety considerations like pulling mass inwards, making it softer and safer for human interaction.
A hacking hardware startup famously started by Palmer Luckey, later acquired by Meta, and known for developing VR devices like Rift and Quest.
An Oculus prototype made before Meta's acquisition, reflecting the startup's spirit of rapid iteration.
A VR headset whose price was significantly reduced through redesign to democratize VR, leading to it becoming the highest-selling VR headset of all time.
A common hardware component where AI is increasingly aiding in routing and basic component selection and layout.
A well-known past issue with Apple's keyboard design, acknowledged as something that needs to be 'gotten right' in hardware.
Used as an example of a fundamentally new product where traditional customer feedback wouldn't work, as customers wouldn't know what they wanted before seeing the innovation.
Processing memory, crucial for running programs, experiencing significant price increases and supply chain constraints due to AI demand, especially from data centers.
Mentioned as an example of vertical integration in manufacturing, where Elon Musk's companies famously produce many components in-house, leading to better adaptation to supply chain shocks.
A VR headset that experienced a significant hardware development failure at the EVT stage due to a misinterpretation of camera specifications, requiring a critical redesign.
Meta's most advanced AR product, on which Caitlin led the hardware team; noted as being ahead of its time due to waveguides and microLEDs not ready for mass production.
A sponsor of the podcast, providing APIs for B2B SaaS companies to integrate enterprise features like SSO, SCIM, and RBAC.
The primary tool for hardware design; discussed in terms of its current limitations for AI integration, though AI is expected to transform it in the future through rapid design and automation of less fun tasks like tolerance stacks.
A social media platform mentioned where OpenClaw was given access and subsequently posted private information.
An AI model capable of generating surfaces or point clouds, but not yet 'real CAD' with dense, solid entities.
Described as word generators/guessers, currently not sufficient for the physical understanding (friction, weight, contact pressure) needed for deep engineering work in CAD.
Google's AI model, mentioned in the context of world models and new model types for physical understanding.
An AI system that Caitlin desires for hardware engineering, highlighting its value in other fields but a current lack of equivalent for her domain.
The future concept that Meta renamed itself around, though many are now leaning out of it.
A technology used in AR glasses like Project Orion, noted for not being quite ready for mass production due to yield and cost challenges.
Display technology used in advanced AR glasses like Project Orion, currently not ready for mass production due to yield and cost issues.
The motors that convert electricity into motion in robots and drones; identified as a potential bottleneck in the supply chain due to dependency on foreign manufacturers and raw materials like magnets.
A technical concept mentioned as a potential long-term solution for safely hosting and training AI models on private company data within their own data centers.
Refers to the classic reveal of the first MacBook Air by Steve Jobs, where it was pulled from a Manila envelope to showcase its thinness, though it was a lower-volume proof-of-concept model.
Potentially new model types that could form the base for CAD and physical engineering by understanding real-world concepts like friction, weight, and contact.
A fiction book that Caitlin recommends, noted for being a classic.
A book by Virginia Woolf about transitions, which Caitlin recommends as an interesting post-war read.
Considered the first history book, which Caitlin finds incredible for its insights into a different era, despite factual inaccuracies.
More from Lenny's Podcast
View all 30 summaries
100 minHow Anthropic, Costco, and Patagonia all build incorruptible companies | Eric Ries
88 minAI era skills: Why cultivating agency matters more than job titles | Max Schoening (Notion)
71 minSnapchat CEO: Why distribution has become the most important moat | Evan Spiegel
86 minHow Anthropic’s product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free