SiMa.ai is a semiconductor and software company building purpose-built AI systems-on-chip to bring power-efficient AI into physical-world applications such as robotics, drones, automotive, medical, aerospace and defense.
What was the original vision behind the company?
I had mostly been in large public companies servicing what I broadly call edge opportunities: robotics, automotive, medical, aerospace and defense, vision systems, and technologies that touch our lives. At the same time, AI from a cloud construct was booming. Even in 2018, it was clear this would be a large trend, though perhaps nobody guessed it would become this big.
What I wanted to do was bring AI to the physical world. We saw cloud AI increasing, but not many people building purpose-built platforms to scale AI in the physical world. That was the simple thesis behind SiMa.ai.
Do you believe 2026 could be the year of physical AI?
Yes. A few tipping points are coming together. For 60 or 70 years, we have lived on rule-based compute, whether CPU, GPU or DSP constructs. For the last 10 years, we have lived with nascent AI and machine learning efforts, such as CNN-based and transformer-based architectures.
The biggest shift in the last year and a half or two years is that, for the first time, we are beginning to build reasoning-based architectures. Today’s flavor is large language models, but it is becoming multimodal. Whether it is 2026 or 2027, we are near a tipping point, and I believe physical AI will be a much larger construct than cloud AI.
What are the potential benefits of physical AI in terms of delivering ROI?
There is a little sense of jadedness because many consumer-centric AI applications in the cloud, and even some business-centric ones, have weak economic outcomes. People are now trying to understand where AI can justify ROI.
One benefit of the physical world is that these markets have existed for tens or even hundreds of years. Automotive is not new. Robotics is not new. Medical is not new. The economic streams are well understood. What is new is the underlying technology shift from rule-based compute to AI-based systems, which brings better accuracy, productivity, total cost of ownership and overall capability.
Physical AI is very different because you are dealing with the real world. First, there is real-world sensing. Like a human being, you are sensing visual, textual or audio data, processing it, and converting it into a construct that an AI engine can work with.
Second, anything that touches human beings has a safety and security context that you do not have in the same way in the cloud. If ChatGPT gives you a wrong answer, you move on. But if you are driving a car or a robot is operating near you and it makes the wrong decision, that is very different. Third, physical AI involves a huge diversity of applications, with tens of thousands of customers doing radically different things.
Where does SiMa.ai fit into that transition?
Most peer companies have looked at AI as a product. For us, AI is really a capability, not a product. We are not building a standalone AI accelerator. In the physical world, you need to build a system on a chip that comprehends the AI capability.
The idea is to capture data in whatever form, audio, video or text, and compute and make a decision on a single chip. We are the only startup really doing that. We also benefited from being purpose-built. We do not have a GPU, DSP or CPU legacy, so from both a software and silicon perspective, we built something specifically for scaling AI.
What differentiates SiMa.ai’s technology?
Performance matters in the physical world, but performance per watt is critical. On performance per watt, we are 10 times ahead of anybody, including the largest GPU company in the world. That has given us huge differentiation.
Software is also key. In my previous company, we scaled to 30,000 or 40,000 customers, and I learned that software experience is essential. The combination of performance per watt and scalable ease of use gives us a one-two punch. In the physical world, a proof point of one or two is easy, but scaling to volume is the hard part, and software is what often gets in the way.
Can you give concrete examples of where this applies?
Our two volume-focused markets today are drones and robotics. In two years, automotive will be added. We also participate in medical, aerospace and defense, and visual inspection systems. Aerospace and defense will also be a key growth vector for us.
The engagement with STIGA Group, the lawnmower company, is a robotics engagement. We are collaborating closely with STIGA to bring AI-powered solutions to robotic lawn mowers. SiMa.ai delivers real-time decision-making on an ultra-low-latency and low-power platform, providing a scalable solution architecture for all robotic lawn mower products.
How do you see the drone market evolving?
Drones have become much more significant than I would have guessed two years ago. I am not talking about commercial drones, but industrial drones and drones for defense. Ukraine, in particular, has pushed the envelope and is the market leader today in drones. The rest of the world is learning from them, from datasets to deployments, cost efficiency and power efficiency.
For a drone application, it is not just performance alone. It is a flying robot or flying vehicle, so power efficiency really matters. We are one of the most power-efficient solutions in the world today, and our performance matches everyone else. If I had to identify poster-child use cases, they are drones and robotics today.
What role will automotive play for SiMa.ai?
There are two things happening in automotive. The classic infotainment system is evolving into an AI assistant. You will talk to your car, and the car will talk to you like a human being. The big difference is that it will all be on-device. There will be no need for 5G or Wi-Fi, and you will have an on-device agent that talks to you at human-level latency.
The second area is driver assistance. I still think classic Level 4 and Level 5 autonomous cars are a long way away. But Level 2-plus driver assistance will absolutely happen in volume and scale. It is already happening now, and it will scale further in 2028 and 2029.
Which industries are most favorable for robotics adoption today?
Factory floor automation, no doubt. Warehouse robots and industrial environments will see the fastest adoption. There is a lot written about robots in the home folding laundry, but that is still some way away economically, by capability and from a human safety perspective.
Industrial automation is different. Hazardous conditions and repetitive jobs can be streamlined. Robots for automotive manufacturing, pharmacy, gripper sort handling applications, agriculture and warehouses are absolutely happening globally. I believe we will see a boom in robotics, not necessarily in the humanoid context, but across industries, manufacturing and agriculture.
Can you share a few words on the implications of your partnership with Micron Technology?
I cannot disclose the details of our partnership, but compute and memory are tied at the hip. In the physical AI world, we need to partner very closely around memory.
Our entire software construct is built around how tightly we integrate the memory subsystem with our compute subsystem. Micron is an amazing company, and we have partnered with them deeply to build a solution we can jointly scale for physical AI.
What are your priorities for the next 12 months?
For us, we have built best-in-class technology. We have the best silicon and software for scaling physical AI compared to other companies. My headwind is that we are a startup. If we were a public company, life would be a lot easier.
The key priority now is scaling. That means scaling our go-to-market, our partner ecosystem, our sales and support structure, and our software so it becomes easier to use every day. If I had to use one word for the next 12 months, it would be scaling.
