🤖 Mind the Gap

Good morning! Matt here, and if you can believe it, it’s Friday once again! Which means it’s time to break down the week and build up some infrastructure.

The world is physical, but it mostly runs on digital technology. And there’s a serious gap between the two.

So naturally, we looked to the smartest people building in that intersection to understand what that gap is and what they’re building to help close it. 

Quez from 375AI, Nils from Auki, and Richard Robinson from Intercognitive all happened to share the same powerful statistic that underpins their respective theses:

That 70% of all commerce happens IRL, in person. This presents a massive market for physical AI to optimize.

When the smartest people in a space independently converge on the same stat in the same week, something is up.

Let's take a look at the statistics driving the future of physical AI forward. 👓

🛰️ On the Radar

📡 Quez from 375AI on why machines need ground truth — and how you can sell it to them.

👁️ The DePIN sensor stack: eyes, ears, and coordinates for the physical AI era.

🤖 Figure scaled humanoid manufacturing 24x in 120 days. Proof of Scale entered the chat.

🧠 KAI Robotics ships a humanoid that learns without updates. The on-device era begins.

🏭 NVIDIA, Flex, and Teradyne are all racing to dominate physical AI. Incumbents see the gap, too.

Time to dive in. 🥽

📡 The Relay

This week, we sat down with Quez from 375AI to talk physical AI. We got into what it is, why it matters, and the insufficient-data problem that presents a massive opportunity.

Quez framed it cleanly: pushing bytes is easy, pushing atoms is hard. LLMs operate in a digital sandbox. But Physical AI doesn't get that luxury. A self-driving car that "thinks it saw something" causes accidents. A robot that misreads a room breaks things. The only way to effectively power physical AI is through the volume, freshness, and accuracy of real-world data. This is what Quez and others call the ground truth, what machines need to build a working world model.

For 375ai, they’re attacking this issue from three angles. 375 Edge sits on 40,000+ billboards capturing vehicular behavior at fixed points across LA, NY/NJ, and now South Florida. 375 Go turns the phone in your pocket into a retail intelligence node. In plain English, capture shelf data at Target, Kroger, or 7-Eleven and get paid in $EAT. One user hit 115 stores last week. That stacks. Just ask SuburbanCrypto, another power user.

Then there's 375 Street, an upcoming self-deployable node for foot traffic, malls, parks, and commercial strips. This is where vertical stacking gets real: if you already host Helium or GEODNET, you're sitting on prime Street locations.

The robots will get the headlines. The networks training them will get the returns.

🎙️ Listen to the full episode | 📱 Grab 375go on iOS or Android | 🌴 Catch us and Quez at Consensus Miami next week.

👁️ Eyes, Ears, and Coordinates

375 is just one piece of the stack. Physical AI needs the full sensory picture, and a handful of DePIN networks are quietly assembling it. To name but a few:

👁️ Eyes — Auki is building the spatial layer. Visual positioning, called posemesh, enables a robot to know exactly where it is in a room down to the centimeter. Co-founder Nils has been hammering the same 70% point Quez made. Most of the global economy lives offline, and machines can't operate there without spatial understanding.

👂 Ears — Silencio is collecting the audio layer. Noise data, voice data, and accented speech across languages. The kind of training data that makes a robot understand a command shouted across a noisy warehouse, not just one whispered in a quiet lab.

📍 Coordinates — GEODNET delivers the centimeter-accurate positioning that everything else depends on. We covered Mike Horton's full breakdown last week, but it bears repeating that without ground-truth GPS, none of the rest works.

We’ve built our physical world for humans, not robots. This, however, is what physical AI infrastructure looks like when you build it from the ground up. Human-based, but robot-centric.

🔊 Signal Boost

The pace is accelerating. Three stories from this week make that impossible to ignore:

🏭 NVIDIA + Flex/Teradyne — NVIDIA Omniverse is making serious strikes in physical AI in manufacturing with ABB, JLR, and Tulip. Flex and Teradyne just expanded their partnership to scale cobots and AMRs across 100+ facilities in 30 countries. The incumbents see the same gap we do, and they're moving fast to close it. The window for distributed networks to claim the data layer is open, but it won't stay open forever. 🔗 Read the article

🤖 Figure — Figure scaled humanoid manufacturing 24x in 120 days, going from 1 robot/day to 1 robot/hour. They're shipping 55 units this week alone, showing that proof of scale beats proof of concept every time. The rise of the robots never looked so good. 🔗 See the post

🧠 KAI Robotics — China's KAI Robotics launched a humanoid that learns new tasks without firmware updates. On-device learning means less reliance on centralized model pushes. It also means more demand for the real-world data that trains those on-device models. 🔗 Watch the demo

👋 Signing Off

The gap between bytes and atoms is the biggest opportunity in tech.

Mind it. The networks that fill it are the ones that get built into everything that comes next.

Quez left us with a challenge worth emphasizing: lean into curiosity over judgment. Ask the questions. Be the newb. The physical AI revolution is coming, whether you understand it yet or not. The folks who lean in early are the ones who solidify their position.

Share this with someone who's still sleeping on physical AI. Tune into The Relay live on X every Wednesday. Follow Matt and Will for more on the distributed future.

Until then, keep deploying.

We'll catch you next Friday!