Hi there,
Welcome to the 138th edition of Heartcore Insights, curated with 🖤 by the Heartcore Team.
If you missed the past newsletters, you can catch up here. Now, let’s dive in!
Today, we are stoked to officially welcome Naza Metghalchi to Heartcore Capital as Partner! 🔥
Naza has spent her career around ambitious people building movements before the rest of the world fully understands them. From political systems to startups, she has always been drawn to conviction, speed and obsessive belief.
She joins us from EQT Ventures, where she backed founders across the B2B AI transition and built a reputation for backing exceptional people early.
She deeply aligns with how we think about investing from inception at Heartcore: believing in exceptional people long before the evidence arrives.
She also spent a surprising amount of time stress-testing one particular Heartcore value: Love. Safe to say, she eventually got stuck with it. ❤️🔥
Hear from Naza herself on her journey, investment philosophy and how she works with founders from the earliest stages 👇🏻
There are moments in technology where several things align at once. In Physical AI, three shifts that would each independently excite investors are happening simultaneously: a migration of world-class talent from research labs to commercial ventures, a series of compounding technology breakthroughs and structural tailwinds that are pulling the entire ecosystem forward.
We have been tracking this space closely, and the pace of change over the past twelve months has been unlike anything we have seen before in robotics.
The talent signal is unmistakable. Academic research in robotics has increased exponentially, and the best researchers are no longer staying in labs - they are founding companies. Physical Intelligence, started by former Google DeepMind researchers, is raising $1B at an $11B valuation. Skild AI closed $1.4B in January. In Europe, Germany’s NEURA Robotics raised $1.2B in March, becoming the continent’s most funded humanoid robotics company, and launched Europe’s largest scientific training centre for Physical AI - TUM RoboGym. This pattern of talent exodus from academia into startups mirrors everything we have seen in previous inflection points - and historically, once the talent moves, the market follows.
The hardware economics are shifting fast with the most consequential shift in edge compute. Ten years ago, $500 bought you roughly 1 TOPS (trillions of operations per second) of edge compute capability. Today, NVIDIA’s latest Jetson chips deliver +2,000 TOPS - a 2,000x performance improvement at only a 7x increase in price. This matters because Physical AI demands real-time intelligence. A chatbot can tolerate a half-second delay; a surgical robot or an autonomous vehicle cannot. Intelligence has to live on the robot itself.
This creates an interesting dynamic that separates Physical AI from digital-based AI. Robots are battery-constrained, thermally constrained and latency-constrained systems. You cannot simply deploy ever-larger models. Instead, the industry is being pushed toward smaller, faster, more efficient models - and toward more open ecosystems. NVIDIA’s GR00T N1 foundation model for humanoids is open-source and integrated with Hugging Face. Cosmos provides synthetic data and world modelling. Isaac Sim enables large-scale simulation. The infrastructure players benefit when more teams experiment and deploy, and that is accelerating the entire space.
This is also creating opportunities for a new infrastructure layer around Physical AI. Companies like Qualia 🖤, which we backed last year, are building the developer tooling robotics teams to work across different models, and environments. As experimentation accelerates, interoperability and workflow abstraction may very well become one of the most important layers.
But let’s be intellectually honest: robotics is still hard. Generalisation remains a fundamental challenge - a warehouse robot trained on one box shape may fail when packaging or lighting changes. Manipulation requires hardware breakthroughs as much as software ones. And unlike LLMs, there is no internet-scale dataset for robotics. Data collection is expensive, embodiment-specific, and often proprietary. Lastly there is the whole issue around “catastrophic forgetting” - when a robot learns to navigate a new kitchen layout, it risks overwriting the skills it previously mastered elsewhere, because memory and learned knowledge are encoded in the same model weights. Future architecture will need to separate long-term memory from core reasoning, but we are not there yet.
The emerging paths are promising. World models - neural networks trained on massive amounts of video that learn physical intuition - are showing early results. Meta’s V-JEPA 2 achieved 80% zero-shot success on real robot arms with just 62 hours of robot-specific data. Simulation and RL work well for locomotion. And the gap between demonstration and deployment is genuinely closing: Figure AI went from producing one robot per day to one per hour at its BotQ facility, and 1X’s consumer humanoid NEO sold out its first-year production run of 10,000 units in five days.
Where does this leave us as investors? We believe open-source will commoditise model architectures faster than most expect, but the data and deployment layers will remain proprietary long enough to matter. The companies that understand which parts of their stack to open and which to protect will have a meaningful strategic advantage. Commercial adoption is emerging fastest in constrained, vertically specialised environments - healthcare, industrial automation, defence, autonomous vehicles – closed loops, where robots do not need to do everything, they need to do one thing reliably.
Robotics may not be having its ChatGPT moment just yet. Our bar for what is “good enough” for a robot remains orders of magnitude higher than for a chatbot. But the convergence of talent, declining hardware costs, exponentially improving edge compute and open infrastructure is creating the conditions for something significant.
As always, if you are building in this space, we would love to talk.
~Lærke Hansen, Investor, Heartcore Capital
Predicting AI Job Exposure, Benedict Evans
The Art of the Interview, David Rubenstein
State of AI: May 2026, Nathan Benaich
AI Eats the World, Benedict Evans
Davis AI 🖤 announced a $5.5M pre-seed round led by Heartcore Capital and Balderton with participation from Yellow, Evantic Capital and a group of exceptional angels! ⚡️
Corti 🖤 launched Symphony for Speech-to-Text, the world’s most accurate medical speech-to-text engine, outperforming OpenAI, ElevenLabs, Google, Amazon Web Services (AWS)! 🔥
TIME has named Fora Travel 🖤 one of its #TIME100 Most Influential Companies 2026. 🏆
Our very own Michael 🖤 organised the first of Europe’s Quantum Builders with IQM’s Jan Goetz in Munich. 🎤
Thank you for being a loyal subscriber of our monthly Insights. Please feel free to share this newsletter with anyone you’d think would appreciate it!
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.