Ivan Poupyrev

@ipoupyrev no problem! there's a fascinating history behind this bbc.co.uk/news/resources…

Flew back to the U.S. from Munich on another piece of obsolete technology—the Airbus A380. This thing is impressive; it’s sad to see them disappear.

@RahelJhirad Last time I talked there, they were very accommodating.

@ipoupyrev can I get off the waitlist? seems sold out.

Excited to be speaking at the @UCBerkeley MDes Forum on Emerging Technology tomorrow — tackling the shift from interface design to agent design, and what it means when systems act autonomously in the physical world. The speaker lineup is incredible: Andrew Pickering on cybernetics and agency, Maggie Gram on the evolution of design practice, Liz Danzico from Microsoft AI on responsible AI at scale, Jodi Forlizzi from CMU on interaction with Agents and Bruce Sterling, bringing his speculative lens to agentic systems. Looking forward to the conversation. If you're in the Bay Area, there's still time to register, see the thread ⬇️

Incredible, thank you for sharing this. I am going to look into this!

My hotel in München used to be an ice factory. They literally manufactured ice there. Then refrigerators came — and it disappeared. We didn’t ban fridges to protect ice factories. We moved forward. AI is no different.

Big news! We’re announcing a new partnership with @TMobile and @nvidia to bring Physical AI to the edge. AI is starting to move beyond digital workflows into the physical systems that run factories, infrastructure, and cities. That shift needs a different kind of tech stack where intelligence runs close to where data is generated and decisions are made. With T-Mobile and NVIDIA, we’re deploying Newton, our Physical AI foundation model, on distributed edge infrastructure powered by RTX PRO 6000 Blackwell GPUs. This brings real-time understanding of sensor data directly to the source, whether that’s a production line or a city block. How it works: 🔹 T-Mobile is turning its network into a platform for distributed AI 🔹 NVIDIA provides the accelerated compute for real-time inference 🔹 @PhysicalAI is building the model that can interpret and reason over the physical world Together, this new stack enables AI to move beyond observation into real-world impact — shifting systems from reactive monitoring to real-time understanding and control. As intelligence becomes embedded across infrastructure and environments, the result is not just more intelligent systems, but a new class of systems that can continuously learn, improve, and act at global scale. We’re excited to be among the first deploying on this. Learn more here: blogs.nvidia.com/blog/telecom-a… Book a demo to explore Newton use cases at #NVIDIAGTC: archetypeai.io/put-physical-a…

A step toward autonomous objects everywhere. Here’s TimeFusion, @PhysicalAI's new breakthrough 2B-parameter multimodal foundation model, allowing you to talk to any device, ask questions, and get both natural language and time-series output (which no other models are capable of). Here's how it works: TimeFusion processes raw IMU data from a smartphone. Identifies activities (shaking, raising your hand) without explicit training. Then, when queried in natural language, it generates a control signal: 0 for no motion-of-interest, 1 when detected. This closes the loop from insight → recommendation → automation. And it works on existing sensor infrastructure. No new hardware required.

Heading to NVIDIA GTC in San Jose? The @PhysicalAI team will be there with live demos of Newton, our Physical AI foundation model that helps teams understand complex sensor data in real time. Foundation models transformed the digital world. The physical world is next. Meet us at GTC to see Newton — the first general physical intelligence model — turn your existing sensor infrastructure into systems that perceive, reason, and adapt in real time. Book a demo with us at GTC → archetypeai.io/put-physical-a…

At @wef Industry Strategy Meeting in Munich, at Siemens’ global headquarters. AI and physical AI are at the center of all discussions. Everyone understands that AI is a tectonic shift—like electricity or communication tech once changed everything. No one thinks its hype or a fad.

Excited to be joining the World Economic Forum's Industry Strategy Meeting in Munich next week. Looking forward to conversations on how AI is reshaping manufacturing, infrastructure, and transportation – and where physical AI world models and physical agents fit into that transformation. If you're in Munich attending the @WEF meeting, drop me a message and let's catch up!

Just wrapped up at the P&P Germany Summit in Munich — what a setting. 🏛️ The BMW Museum provided a stunning backdrop for conversations about Physical AI, Newton and physical agents in the real world. Grateful for the sharp questions and energy in the room. Munich delivered!

🎉 Today we are excited to be welcoming @DongLin_ML as the new Principal Research Engineer, leading Archetype's Foundation Model Team! Dong brings over a decade of experience in developing, shipping, and optimizing world-class production models at Google and Apple. Dong, welcome to the team!

Excited to share a peek at our new home. 🏠 @PhysicalAI's new office is where the next chapter of Physical AI begins. Grateful for the team that shows up every day to make it real.

Latest from @PhysicalAI demo land: This is Newton distinguishing between two people using the same phone. But how? Everyone has a slightly different way of typing. Newton understands the micro-differences in motion, which are mapped clearly in the embedded space (on the right). We never explicitly trained Newton to do this. Our Physical AI foundation model is capable of detecting emergent behaviors — so instead of multiple vertical AI solutions, organizations can leverage Newton to generalize across hundreds of sensor types and support a broad range of use cases. #PhysicalAI #FoundationModel

I sat down with @richierocks from @DataCamp to talk about a question I get asked a lot: how is Physical AI different from robotics? Robotics focuses on embodied intelligence (arms, drones, humanoids) that acts in the world. Physical AI is much broader as it focuses on understanding the world itself through sensor data. 🎙️ Full episode here: datacamp.com/podcast/lets-g…](datacamp.com/podcast/lets-g…)

Over the last couple of years, AI models got really good at reasoning over text and images. Meanwhile, factories, fleets, and infrastructure generate continuous streams of vibration, motion, temperature, and current and these signals rarely make it into the conversation in a usable way. What if we bridge between these sensor signals and language? Today at @PhysicalAI we share new research demonstrating that meaningful integration between physical signals and language can be achieved with minimal additional computation. To do so, we built a lightweight 3.3M-parameter alignment network that connects our Newton physical world encoder to a frozen LLM (Llama-2-7B-Chat) in under 30 minutes on a single modern GPU. All of this is possible because Newton, our Physical AI foundation model, clusters together signals that mean similar things without being trained to do so. That emergent structure is what makes alignment with language practical and easy. This matters beyond the lab. The physical world generates more sensor data than any team of experts can interpret. The companies that win in manufacturing, infrastructure, transportation are those that can use existing infrastructure and turn raw signals into insights. --- Paper by Hasan Doğan, Jaime Lien, Laura I. Galindez Olascoaga, Muhammed Selman Artıran and O. Serdar Gedik 🔗👇 archetypeai.io/blog/language-…

Well said! "When intelligence enters the physical AI economy across robots, factories and infrastructure, it unlocks tens of trillions of dollars in opportunity. In this realm, AI doesn’t just generate content, it moves the world." @JohnSaw @TMobile! t-mobile.com/news/network/w…