Mirai Raises $10M for On-Device AI

- The founding team consists of Dima Shvets, co-founder of the viral face-swapping app Reface, and Alexey Moiseenkov, co-founder of the AI photo-filter app Prisma, bringing experience in scaling consumer AI applications with millions of users. - Mirai’s core technology is a proprietary inference engine built in Rust, initially optimized specifically for Apple Silicon, which it claims can improve model performance by up to 37% without altering model weights or quality. - The $10 million seed round was led by Uncork Capital and includes angel investments from notable figures such as the CEO of Dreamer, a YC Partner, a co-founder of Snowflake, and a co-founder of ElevenLabs. - Mirai is developing a hardware-aware runtime that acts as an interaction layer between AI models and device hardware, aiming to abstract away the complexity of memory management and kernel optimization for developers. - The company plans to use the funding to expand beyond Apple Silicon and support text, voice, vision, and multimodal workloads, with the goal of making on-device processing the default for AI applications. - This investment comes as the edge AI chip market is forecasted to grow significantly, with some projections expecting it to reach over $80 billion by 2036, driven by applications in automotive and AI-enabled smartphones. - The competitive landscape for on-device inference includes established solutions like Google's MediaPipe and open-source frameworks such as ONNX Runtime, highlighting the growing importance of optimizing AI performance on consumer hardware. - Mirai is also building an orchestration layer to create a hybrid system, allowing AI workloads that are too complex for the device to be seamlessly sent to the cloud for processing.