NextFin News - In a strategic move to decentralize artificial intelligence from the cloud to the palm of the hand, the entrepreneurial minds behind some of the last decade’s most viral AI applications have joined forces to launch Mirai. Dima Shvets, co-founder of the face-swapping sensation Reface, and Alexey Moiseenkov, the creator of the neural style transfer app Prisma, announced on February 19, 2026, the closing of a $10 million seed funding round for their new venture. The London-based startup, which currently employs a 14-person technical team, aims to solve the "inference gap"—the performance lag and high energy consumption that occur when complex AI models are forced to run on consumer hardware like smartphones and laptops.
According to TechCrunch, the seed round was led by Uncork Capital, with participation from a high-profile roster of angel investors including ElevenLabs co-founder Mati Staniszewski and Groq investor Scooter Braun. Mirai has developed a specialized inference engine built in Rust, designed specifically for Apple Silicon and soon Android, which the company claims can increase model generation speeds by up to 37% without compromising output quality. By providing a Software Development Kit (SDK) that requires as few as eight lines of code to integrate, Shvets and Moiseenkov are positioning Mirai as the "Stripe for on-device AI," enabling developers to deploy summarization, classification, and voice models locally with minimal friction.
The emergence of Mirai comes at a pivotal moment for the global technology sector. While the initial wave of the generative AI boom was defined by massive data centers and cloud-based Large Language Models (LLMs), the economic and physical limits of this model are becoming apparent. For developers, the cost of cloud inference is a recurring tax that erodes margins; for users, the latency and privacy concerns of sending data to remote servers remain significant hurdles. Shvets noted that while the industry has been obsessed with Artificial General Intelligence (AGI) and server capacity, the "missing piece" has been the optimization of consumer hardware. This sentiment is echoed by the current political climate in Washington. U.S. President Trump has frequently advocated for American technological self-reliance and the reduction of energy-intensive infrastructure costs. Mirai’s focus on making existing hardware more efficient aligns with a broader national interest in maintaining a competitive edge in edge computing without solely relying on the expansion of massive, power-hungry server farms.
From a technical perspective, Mirai’s choice of Rust as its foundational language is a deliberate play for memory safety and execution speed. Unlike traditional Python-heavy AI stacks, which often struggle with the resource constraints of mobile devices, Mirai’s engine optimizes throughput directly at the silicon level. This is particularly critical as companies like Apple and Qualcomm release chips with dedicated Neural Processing Units (NPUs). However, hardware alone is not a panacea. The software layer must be able to orchestrate tasks between the local device and the cloud—a "mixed mode" of operation that Mirai is currently building. This orchestration layer ensures that while simple tasks like text summarization happen instantly on-device, more complex reasoning can still be offloaded if necessary, preserving the user experience.
The market implications of Mirai’s success could be profound. As Andy McLoughlin, managing partner at Uncork Capital, pointed out, the venture capital community is beginning to shift its focus from "rocketship" foundation model companies to the "picks and shovels" of the AI era. The underlying economics of cloud-only AI are increasingly viewed as unsustainable for mass-market consumer apps. By moving inference to the edge, Mirai allows developers to bypass the per-token costs associated with cloud providers, potentially unlocking a new category of "always-on" AI assistants that do not require a constant internet connection or a massive operational budget.
Looking ahead, the trajectory for Mirai involves expanding beyond text and audio into the more computationally demanding realm of computer vision. As augmented reality (AR) hardware matures, the need for real-time, low-latency spatial intelligence will become the next frontier for on-device inference. If Mirai can successfully port its performance gains to Android and diverse chip architectures, it may become the standard runtime for the next generation of intelligent software. In an era where U.S. President Trump’s administration is pushing for leaner, more efficient domestic tech ecosystems, Mirai’s mission to squeeze more power out of the devices we already own represents a vital evolution in the AI value chain.
Explore more exclusive insights at nextfin.ai.
