Mirai launches to boost on-device AI inference speeds by 10x

The London-based startup Mirai has secured $10 million in seed funding, led by Uncork Capital, to pursue its goal of dramatically accelerating artificial intelligence on consumer devices. According to TechCrunch, the 14-person company was founded last year by Dima Shevts and Alexey Moiseenkov, technical founders with a proven track record in building viral, AI-powered consumer applications. Shevts co-founded the face-swapping app Reface, which was backed by venture firm a16z, and Moiseenkov was the CEO and co-founder of the AI photo-editing app Prisma.

The founding team’s background is central to Mirai’s mission. TechCrunch reports that both founders had been focused on the challenges of on-device machine learning long before the current generative AI boom. Their experience building consumer apps revealed a significant gap in the market. While the industry conversation is dominated by cloud infrastructure and massive data centers, Shevts identified a missing piece: performant on-device AI for consumer hardware. This realization, stemming from a meeting in London, was the catalyst for Mirai. They sought to create a technological pipeline that would enable complex AI tasks to run efficiently on phones, addressing a need they also observed among other developers for better cost optimization and margins.

Mirai’s technical approach involves developing a framework to make AI models perform better on devices. The company has specifically built an inference engine optimized for Apple Silicon to enhance on-device throughput. As reported by TechCrunch, the upcoming software development kit (SDK) will allow developers to integrate this runtime into their applications. This focus on a core piece of technology, rather than an end-user product, represents a strategic pivot for the founders toward becoming an infrastructure provider for the next wave of AI applications.

The push for on-device AI is gaining momentum across the technology sector as a counterweight to the dominant cloud-centric model. Companies like Apple and Qualcomm are investing heavily in making local AI inference more capable and efficient. Mirai’s entry into this space highlights a growing recognition of the cloud’s limitations, which include latency, recurring costs per token usage, and potential privacy concerns. Running AI locally on a device can mitigate these issues, offering users faster response times and greater data security while providing developers with more predictable and scalable cost structures.

What comes next for Mirai is an execution challenge in a competitive and rapidly evolving field. The success of its SDK will depend on its ability to deliver the promised tenfold speed increase and to attract developers away from established cloud providers and other on-device solutions. The company’s initial focus on Apple Silicon targets a premium segment of the market known for its powerful hardware and users willing to adopt cutting-edge applications. Details on a public release timeline for the SDK or specific model compatibilities were not disclosed in the available sources.

The broader industry context underscores the significance of this technological shift. As noted in coverage by VentureBeat, the proliferation of smart devices, or the Internet of Things (IoT, has led to a dramatic expansion of the attack surface for cyber threats, with sensors and connected devices being among the fastest-growing attack vectors. Enhancing on-device processing capabilities could play a crucial role in improving security for this ecosystem by reducing the constant data transmission to and from the cloud that can be vulnerable to interception. Mirai’s work, therefore, touches on not just performance and cost but also on the pressing need for more secure architectural approaches to AI.