SiMa.ai, a Silicon Valley-based startup specializing in embedded ML system-on-chip (SoC) platforms, has announced securing a significant $70 million extension funding round. This investment aims to facilitate the launch of its second-generation chipset, tailored specifically for multimodal generative AI processing.
The decision to focus on this cutting-edge technology aligns with market trends, as illustrated by Gartner’s projection of the AI-supporting chip market to surpass $119.4 billion by 2027. SiMa.ai seeks to capitalize on this opportunity by providing edge AI solutions across various sectors, including industrial manufacturing, retail, aerospace, defense, agriculture, and healthcare.
SiMa.ai’s initial success with its ML SoC, targeted at the 5W–25W energy usage segment, was propelled by its proprietary chipset and no-code software, Palette. Already adopted by over 50 global companies, SiMa.ai’s first-generation ML SoC achieved impressive results, boasting the highest FPS/W performance on the MLPerf benchmark.
Anticipating growing demand for GenAI applications, SiMa.ai is poised to introduce its second-generation ML SoC in early 2025. This chipset will offer multimodal GenAI capabilities, accommodating various frameworks, networks, models, and sensors, thereby serving as a unified platform for AI tasks spanning computer vision, transformers, and multimodal GenAI.
SiMa.ai’s strategic partnerships with industry giants like TSMC and Arm Holdings underscore its commitment to delivering cutting-edge solutions. Leveraging TSMC’s 6nm process technology and Synopsys EV74 embedded vision processors, SiMa.ai aims to maintain its competitive edge against incumbents and emerging AI chip startups.
The latest funding round, led by Maverick Capital and joined by Point72 and Jericho, marks a significant milestone for SiMa.ai, bringing its total funding to $270 million. With plans to expand its team and global presence, SiMa.ai is well-positioned to capitalize on the transformative potential of edge computing in the AI landscape.
