SiMa.ai Hits 1.45 Billion Dollar Valuation As Investors Bet On Chips That Let Robots Think On Their Own

SiMa.ai is pitching its silicon as the brains of what the industry now calls physical AI, the category of systems that pair AI models with sensors and actuators so they can reason about and act in the real world. Examples range from self driving cars to industrial drones inspecting infrastructure to humanoid robots. Counterpoint Research projects cumulative shipments of physical AI devices reaching 145 million units by 2035, a forecast the company cites as evidence of the market it is chasing.

The new capital has two main destinations. The first is Palette Neat, which SiMa.ai describes as an agentic development environment for physical AI, designed to help engineers build and deploy models on its hardware more quickly. The second is Modalix, the next generation system on chip aimed at embedded robotics. SiMa.ai says the new dedicated physical AI chips will arrive in the first half of 2028 and deliver up to 1,000 tera operations per second, a common measure of AI compute throughput.

That target puts the company in direct comparison with Nvidia, the dominant supplier for robotics compute. Nvidia's top end Jetson AGX Thor modules reach 2,000 FP4 TOPS and are built for full scale humanoid robots and complex autonomous machines, while its mainstream robotics and heavy drone modules sit at roughly 400 and 865 TOPS. SiMa.ai's 1,000 TOPS goal therefore lands between Nvidia's mid tier and flagship parts, and the company is betting that customers will trade some peak performance for lower cost, lower power draw and simpler software. Rangasayee has said the funding will help the company take a leading position by scaling its purpose built platform against incumbents whose hardware is more expensive and tied to the CUDA ecosystem.

Rangasayee has framed the opportunity in sweeping terms, describing a 50 trillion dollar market spanning humanoids, automotive and drones that he says modern innovation has largely left untouched. That is a promotional estimate rather than an independent forecast, and the company's disclosed customer base is more grounded. Named customers lean industrial, including Bosch, Emerson, Micron, Synopsys, ARK Electronics and AverMedia, rather than marquee humanoid or automotive brands. That mix suggests SiMa.ai's current revenue comes from factory, imaging and embedded applications, with the higher profile robotics and vehicle markets still largely ahead of it.

The company's strategy rests on the observation that many physical AI deployments cannot rely on a data centre. A drone inspecting a pipeline may have no reliable connection, a robot on a factory floor cannot tolerate network delays, and a vehicle must make decisions in fractions of a second. In each case, processing has to happen on the device, within a tight power and thermal budget. Efficient chips that can run modern vision and language models at the edge are therefore becoming a distinct hardware category, separate from the large accelerators used to train and serve models in the cloud.

Investor appetite for that category has grown alongside enthusiasm for robotics, and a roster that includes Fidelity, J.P. Morgan and a US state investment office signals that institutional money is now looking at edge silicon as well as data centre chips. The harder question is execution. Chip development is slow and capital intensive, a 2028 launch date leaves a long runway in which rivals will ship new generations, and software support often decides whether a processor is adopted at all. SiMa.ai's decision to put its agentic development environment alongside the next chip reflects that reality, since ease of use for developers may matter as much as raw compute in winning designs across robots, drones and vehicles.