It’s time for a new adventure. I’m joining one of the top AI clouds, Nebius (@nebiusai), as their Head of Physical AI. I’m also leaving my first professional love — spatial computing. At least in its most direct and advertised form, i.e. AR, VR, 3D, aka ‘XR’. But in another sense, I’m not leaving at all. I’m going deeper. As I've hammered on my newsletter — Dream Machines — spatial computing is far more than XR. This is just one end-point; just one form of a computer relying on sensors to localize, track, and understand the physical world. Self-driving cars, mobile robots, humanoids, autonomous drones... These are all ‘spatial computers’, each relying on a similar technical foundation, with similar computer vision tech, simulation tools, and ‘spatial’ data types, e.g. lidar, point clouds, 3D, video, IoT/sensors. The key difference? Autonomy and their ability to impact/interact with the physical world. So again, I’m not leaving. I’m just moving downstream; from a ‘spatial tributary’ to a gushing river, surging with capital, progress, and commercial opportunity. Jensen Huang popularized the term 'physical AI' about a year ago, and as it goes... Interest and awareness has exploded ever since. To be sure, Jensen likes to live 10+ years in the future. Heck, so do I. But physical AI remains early and is entering a hype cycle of its own. One that, at first glance, feels like the hype cycle that sucked me into XR. But this feels different. The timing feels right: early, but not too early... If you want to know why... check out the essay below 👇👇 It'll give you a L100-200 grasp of this new tech stack and the key innovations worth monitoring. And if you enjoy, do subscribe! There'll be a Part II... covering some of the challenges that remain, including the ones that I’ll be trying to solve in my new role at Nebius, as their Head of Physical AI. Lastly, if you work in this space and want to connect. Drop me a DM. Otherwise... See you on the front lines🫡 @EndicottInvests @mvcinvesting

Within AI, rarely do we step back, close our eyes, and project all of this stuff forward. Our industry has become too frenetic and too competitive; too anxious about the next GPU, the next research talent, the next benchmark, the next viral moment. When we do, the implications aren’t just staggering, they’re difficult to even comprehend. They require immense imagination and suspension of disbelief, especially for all the big AI use cases: personal agents, AI doctors, AI scientists. But the implications of world models? Something about this feels different. It’s not just the manifestation of net new things in the world. It’s the manifestation of new worlds themselves, of new realities, of new… perspective. And this might be the most upstream implication, which is a shift in the most upstream variables of our behavior: perspective, understanding, foresight. Could it be ungrounding? Sure, especially in the beginning. But it could also be the thing that gives us the imagination and intuition we need to properly navigate this future? Helping us deploy this technology for good, while diminishing (predicting/forecasting) the bad? I think so, with a new group of (robotic) friends to help us build it along the way. Full essay here: dreammachines.ai/p/world-mod… And if you enjoy, be sure to subscribe! More to come from the front lines of physical AI and robotics.

Appreciate the partnership! @antiochrobotics

Everyone is talking about world models. But nobody seems to agree on what a world model actually is. So I sat down with Matthias Niessner, CEO of SpAItial AI, for one of the clearest conversations I’ve had on what world models are, how they’re being built, and why they could matter enormously for robotics. If you’re trying to understand where world models are going — and what they could unlock for Physical AI — this episode is worth the time... (Link below) We cover the spectrum of approaches being built today; from video models that generate pixels, to 3D spatial models, to systems learning physics and cause-and-effect in latent space. And we dig into some of the most important questions in the field: Why might video models be hitting a wall? What’s the difference between worlds that look right and worlds that behave right? How could world models improve robotics simulation and close the sim-to-real gap? How close are we to virtual environments where robots can learn before ever entering the real world? I learned a ton from this conversation and walked away even more optimistic about how quickly we may be approaching general-purpose robotics. If you enjoy, be sure to subscribe to Dream Machines! Much more to come on physical AI and the future of the human experience. dreammachines.ai/p/world-mod…

Had a blast running it back for round 2 with @KingDSky on Sky Kings Mental Playground. We dive deep into the physical AI stack, the key trends to follow, and the overlap between XR + robotics. And in classic Sky King fashion, we spin off in all kinds of directions, from fatherhood, to anime, to the state of the world and what to do about it. Full episode 👇 youtu.be/1gLxEQpwnfs

I ask the same question at the end of every podcast: in an era of exponential change, what's the most important question we should all be asking ourselves? I most recently posed this question to Charles Wong and Aravind Kandiah: two founders deep in the trenches of robotics, AI, and the future of work (Bifrost AI) Their answer wasn't what you'd expect from technologists building at the frontier. No grand theory of civilization. No AGI prophecy. No moonshot manifesto. Just this: What quest are you on? And who's your tribe? In a world drowning in signal, noise, and 10X productivity expectations… the thing that actually matters is the oldest thing in the book — your people, your community, the life you're choosing to live. It's the perfect note to end on. It's also the whole reason Dream Machines exists — to explore not just where technology is going, but where it's taking us. Full episode here: dreammachines.ai/p/synthetic…

Imagine when intelligence is "solved". Imagine there's a final API for world: a single brain that can do anything, anywhere; the physical AI holy grail. Even then… you've still got a problem. You can't just load it into a robot, walk it into your lab, and hope it works. You need rigorous, domain-specific evals. CI/CD for robots. Thousands of scenario tests before a single one touches real infrastructure — or real lives. In this clip, the Bifrost co-founders argue the real bottleneck of the robot age isn't training. It's evaluation. And they're already living it: most of their customers use them for evals more than training. The moonshot? An "imagination engine" running on edge compute — where robots roll out 40 versions of a scenario in their own heads before deciding what to do. Wild stuff. And maybe closer than we think. For a deeper dive, check out the full episode below👇 Apple Pod: podcasts.apple.com/us/podcas… Spotify: open.spotify.com/episode/6R9…

One of the most common debates in robotics... Can you really train robots in simulation… and have them just work in the real world? The answer is far from a clean yes or no, and the conundrum has a name: Sim2Real. In this clip, Bifrost's co-founders demystify it. We break down: - The 4 layers of Sim2Real (perception, dynamics, contact forces, deformables) - Where it actually works - Where it absolutely doesn't - And the mind-bending trick of "sim → sim → sim → real": bouncing a policy across physics engines until reality becomes just one more. Best line of the whole convo? "Reality is just another type of physics engine." Make of that what you will. Full episode here: dreammachines.ai/p/synthetic…

"Sim-2-real" is one of the most common buzz words in robotics. But very few truly understand what it means, why it matters, and how to harness it. This episode will change that.