Lingsi raises ₽35 billion for chips that run AI on the user's own device

Lingsi Tech, a Chinese startup specializing in chips for local AI inference, has raised a Series B of about 500 million yuan (roughly ₽35 billion). The money came from state investment funds of Anhui Province and the city of Hefei, as well as venture funds Deep Report Capital, Sky Intelligence Ventures, Scientia Venture, Yingke Capital, Dongrui Capital and Yongxin Ark. Venture firm Taihé Capital acted as the company's financial adviser.

Where the money will go

Lingsi will put the funds into a new generation of chips for edge AI inference, that is, running neural network models on peripheral devices. The company plans to move on from its current focus on chips for perception models (image recognition, sensor data processing) toward supporting large language models and reasoning systems that run directly on the user's device. This would let users run powerful AI systems locally, without a connection to the cloud.

• Series B: about 500 million yuan (≈₽35 billion)
• Main investors: state funds of Anhui and Hefei
• Current direction: edge inference for LLMs and cognition-level models
• Product evolution: from perception models to cognition-oriented chips

Why it matters for the AI market

Edge computing chips are gaining ground because cloud computing runs into problems with latency, privacy and cost. Users and enterprises increasingly demand that critical data and models run locally. Lingsi competes with major players such as Qualcomm, Apple (Neural Engine) and NVIDIA, but targets the Asian market and specialized applications. The state backing shows that edge AI is also seen as a strategic area at the government level.

What this means

Growing investment in edge AI inference signals that local inference is establishing itself as a separate segment of the chip market. Distributed processing replaces the centralized cloud. For developers, this means new optimization libraries and APIs for running models on devices. For end users, it means faster and more private AI.