The Industry’s Next Best Thing in Physical AI to Leapfrog Legacy GPUs

Physical AI is here now, scaling globally, and set to have a monumental impact on nearly every aspect of our lives. SiMa.ai is leading the charge. Our Gen 2 Physical AI platform is winning across three significant markets – robotics, automotive, and drones – where it’s shipping in volume production today.

I recently wrote about how we are dismantling the GPU moat to scale Physical AI with technology purpose-built for what this era demands. Rather than trying to retrofit a power-hungry data center GPU for the edge, we deliver energy-efficient performance in stark contrast to NVIDIA, reducing application development from months to days or hours. Our leadership rests on two pillars we’ve spent years building: Palette Neat, our agentic AI software environment, and our Gen 2 MLSoC Silicon Architecture, which delivers a 10x improvement in performance-per-watt over the competition.Together, they’re the reason Physical AI companies solving the hardest problems, like real-time perception, on-device autonomy, and power-constrained compute, keep choosing SiMa.ai.

We’re putting $150 million in new Series C financing straight into a Gen 3 platform that Physical AI actually needs to scale.

Here’s what that funding will unlock:

• One future-proof software across all generations and form factors.
• Software built right from the ground up for humanoids, L2++ and L4 ADAS, high-end drones and robotics, and the industry’s best AI-powered cockpit.
• The industry’s most performant Physical AI platform delivering 1,000 dense TOPS, which is 2X better than NVIDIA, at less than 80W, offered in three variants: IP, chiplet, and chips.

The Industry’s Best, Your Way

Hardware alone doesn’t scale Physical AI. Software does.The industry’s best semiconductor companies need easy to use, scalable and future-proof software. That is what SiMa.ai does better than others in the market as software is deep in our DNA. We are delivering today the best software solution for scaling Physical AI from Gen 2 platform, Gen 3 IP, chiplets, and chips and all under one future-proof software environment.

We’ll hand you Gen 3 chip, chiplet, or just the IP, whichever fits what you’re building, and it all runs on the same software environment. Our proprietary Gen 3 Machine Learning Accelerator IP is available now for high-volume customers in humanoid robotics, automotive, and drones. Production Chiplets and next-gen SoCs will arrive in the first half of 2028. 

That differentiation shows up in the numbers too. We’re pushing the envelope to 1,000 dense TOPS, the industry’s highest compute at the lowest power draw, roughly two times the performance of the leading GPU at under 80 watts. That is the kind of jump that changes what’s actually possible to build for robotics, mid-to-high level drones, and Level 2++ and Level 4 ADAS.

The foundation is proven. The software is ready. Our third generation of product architecture, purpose-built for Physical AI and shaped by years of real customer engagements is underway. What’s exciting is how many ways customers can take it. Contact us today to get started.

Related resources

• Palette Neat Open Source: Access on GitHub.
• Palette Neat Documentation: Get started at the Developer Center.
• Modalix MLSoC SoM: Read the full hardware specification.
• Webinar: Watch Scaling Physical AI on-demand now.
• Questions about our platform or dev kits? Ideas for a collaboration? Get in touch We’re eager to hear what you’re building.