The AI Hardware Boom Is Just Starting — Here's What I'm Watching

I just listened to Lenny Rachitsky interview Caitlin Kalinowski, and I had to write about it. If you don't know her: she ran hardware at OpenAI, before that she led the Quest and Orion projects at Meta, and before that she built the MacBook Air and the unibody MacBook Pro at Apple. The lady has shipped the actual products in your house and on your face.

Her thesis on the show: "We're at the beginning of the AI hardware boom." Not the middle. Not the late stage. The beginning.

After the last two years of pure software-side AI hype, that statement woke me up. So let me break down what she said and where I think she's right.

Her Track Record Is Why This Matters

A lot of people are talking about AI hardware right now. Most of them have never shipped a physical product in their lives.

Kalinowski has. She's been inside the room when Apple decided what the MacBook Air was going to be. She was at Meta when the Quest 2 came out and at OpenAI building whatever's coming next. When she says memory prices are about to spike and the market hasn't priced it in, that's not a take — it's a data point pulled out of the supply chain she's lived in for fifteen years.

The Memory Price Shock Is the Underrated Story

The single most actionable thing she said on the show was the warning about memory pricing.

In her words, a memory price shock is coming, and her advice to startups was to pre-purchase inventory now before costs climb. That's the kind of operational detail that most AI commentators completely miss because they don't understand that AI hardware lives or dies on memory bandwidth.

If you're building anything with onboard inference — robots, glasses, edge devices — your bill of materials is about to get worse. The companies that lock in supply now win. The ones that wait get crushed.

VR Isn't Dead — It's Just Not the Consumer Story

Kalinowski's honesty about VR was refreshing.

Meta's Quest 2 sold gangbusters. Apple's Vision Pro launched and quietly underdelivered against expectations. Meta's Orion glasses are a prototype showing what the next paradigm might look like. But all of it together hasn't produced the consumer VR market we were told to expect five years ago.

What I love about her framing: she didn't pretend VR "won." She separated the engineering wins from the commercial outcome and was specific about what each platform represented.

The consumer VR market didn't materialize. The B2B and prosumer market is real but small. The real future is glasses, not goggles — and we are several manufacturing generations away from that being a product anyone actually wears.

The Robotics Take Is the Mature One

The sharpest thing she said about robotics: the AI is not the bottleneck.

Humanoid robots stay prototypes because the manufacturing and deployment economics aren't there yet — not because the models can't think. The model layer is moving faster than the supply chain can build chassis and actuators.

That's a critical reframe for anyone trying to start a robotics company in 2026. You are competing on supply chain, not algorithms. The companies that will win this category are the ones that already know how to make hardware at scale — Anduril, Tesla, the established robotics players — not the ones with the cleverest demo videos.

Where I Land on This as Somebody Who Actually Builds

I have ten-plus patents to my name, I've sat through enough product development cycles to know what shipping physical things really looks like, and I run Diamond IQ and a stack of AI-powered apps that I build with my own custom assistants. So I have skin in the game on both sides.

Here's my honest take on Kalinowski's thesis:

She's right that we're early. The consumer-facing AI hardware story today is glasses, robots, and ambient computing. None of those have a Killer Product yet. We're at the iPhone-in-2005 stage, not the iPhone-in-2010 stage.

She's right about supply chain being the moat. I've watched what happens when small companies try to compete with Apple on a hardware feature. They get out-purchased on every component they need. That has not changed and it isn't going to.

Where I'd push back: She talks about humanoid robots as the next wave. I'd argue specialized robots (kitchen, warehouse, agricultural, eldercare) ship before humanoids, because the regulatory and form-factor problems are smaller. "Generally capable" is a harder commercial sell than "does this one thing perfectly."

What I'd Actually Do With This Information

If you're trying to build in AI right now and Kalinowski's interview hit a nerve, here's what I'd do:

• Pick a vertical, not a paradigm. Don't build "AI for everything." Build AI for warehouse picking, or AI for school tutoring, or AI for legal contract review. Specialized wins this decade.
• Watch the supply chain. If your business depends on Nvidia chips, memory chips, or any specialty silicon, you're a price-taker on commodities that are about to get more expensive. Plan for it.
• Stop chasing models. Start chasing distribution. The models are commodities now. What's not a commodity is the audience who trusts you to put one in their workflow.

The Bottom Line

Kalinowski's calm, supply-chain-grounded view of the AI hardware market is the antidote to the X.com hype cycle. We're early. The winners aren't going to be the loudest founders — they're going to be the operators who understand that hardware companies live and die on logistics, manufacturing, and inventory.

If you're a software founder thinking about jumping into hardware: listen to that interview first. If you're an operator who's been quietly watching this space: this is the moment to start placing bets.