Fangle Technology, Led by Jun Liang, Raises Over 500 Million Yuan in Multiple Funding Rounds to Forge New Path in AI Chips

By Songdao Tu | Produced by AI Nao

Intro

Shanghai Fangqing Technology, an AI chip startup led by former Cambricon CTO Liang Jun, today announced it has raised over 500 million yuan across multiple funding rounds, backed by an impressive roster of investors.

The Pre-A round was led by a major internet company, with participation from Xinlian Capital, Hundsun Electronics Industry Fund (Yima Capital), GF Xinde Investment Management, and a top-tier VC. Existing shareholders Shanghai Lingang Science and Technology Investment Management Co., Ltd. and 37 Interactive Entertainment continued to increase their stakes. Additionally, the angel+++ round was completed by Dawu Ventures.

Founded in late 2022, Fangqing aims to carve out a differentiated path through system-level innovation in a global AI computing landscape still dominated by NVIDIA. Unlike most companies in the industry that choose to optimize or substitute within existing architectures, Fangqing is attempting to change the rules from a more fundamental level: its core focus is a "decoupled distributed AI computing architecture."

Specifically, Fangqing has proposed a new technical direction: decoupling the key computational components in current mainstream Transformer models — the "context-aware" attention mechanism and the "context-free" feedforward network (FNN) — into two independent modules. In traditional designs, these two are tightly coupled and executed in series. Fangqing's vision is to split them apart and assign them to the most suitable heterogeneous hardware for distributed processing, thereby pursuing greater computational efficiency and scalability at the system level.

This is not a simple improvement on existing chips, but an attempt to reconstruct the AI computing paradigm. Fangqing's ultimate goal is to change existing AI hardware design thinking through this new computing system based on a decoupled-separation architecture, thereby creating entirely new markets and ecological niches.

To date, Fangqing has assembled a lineup of dozens of investors, which AI Nao has preliminarily categorized into industrial resources, star financial investors, and major internet companies. Industrial investors include at least Xinlian Capital, Hundsun Electronics Industry Fund, NIO, 37 Interactive Entertainment, and Hua Capital. The convergence of these industrial resources points to priority scenarios for Fangqing's technology deployment — intelligent driving, finance, interactive entertainment, and broader edge AI devices are all vertical domains extremely sensitive to real-time response and computing costs, with gap markets that existing general-purpose computing cannot adequately serve. Fangqing is attempting to achieve high efficiency and low latency through architectural innovation.

On the major internet company front, in addition to Xiaomi, which entered at the angel round, this Pre-A round was exclusively led by a platform-level internet company. AI Nao has learned that this lead investor has drawn significant attention this year for its extensive deployments in AI, particularly in embodied intelligence and other areas.

Liang Jun, Former Cambricon CTO, Initiating a New Game

In August 2024, Liang Jun — former Cambricon CTO and former chief architect of HiSilicon's Kirin SoC — joined Fangqing, then less than two years old, as CEO. This star architect, who lived through China's golden two decades of chips and possesses rare hands-on experience with both top-tier SoCs and AI chips, chose to join a startup with a clear objective: not merely to participate in existing competition, but to define a new set of computing rules.

Liang's career move came at a delicate moment.

On one hand, Chinese tech companies are facing unprecedented pressure in accessing advanced computing power. On the other hand, the explosive demand for AI applications domestically is forcing the industry to find solutions that boost efficiency beyond established technical frameworks.

Liang's choice can be seen as a response to the latter challenge.

Unlike his previous career focused on building benchmark single-point products, the technical direction Liang is pushing at Fangqing represents a shift in industrial logic. His publicly stated core technical path has been summarized as a "context-decoupled distributed computing architecture." This is not optimization of existing chips, but a turn in system design philosophy: from pursuing peak performance of a single chip to pursuing the overall efficiency of a system composed of multiple heterogeneous units collaborating effectively.

In the view of Fangqing's investors, this innovative architecture will drive the iterative advancement of domestic AI infrastructure.

This path means choosing a harder but potentially more expansive track. It requires simultaneously solving two core problems: first, at the hardware level, achieving efficient decomposition and collaboration of different computing modules; second, at the software and ecosystem level, building development toolchains and application ecosystems that support this new hardware paradigm. This is fundamentally different from the current industry mainstream model of building software-hardware ecosystems around powerful general-purpose computing cores.

The challenges this choice faces are undoubtedly systemic: it tests not only engineering execution capability, but also the composite ability to define new standards, attract developers, and build industrial ecosystems.

But Fangqing's opportunity also lies here. If this system can be proven to provide a significantly better efficiency-cost curve in key scenarios — such as intelligent vehicle applications where cost and latency are extremely sensitive, or high-concurrency scenarios requiring precise resolution of efficiency loss — then it will have the chance to bypass head-to-head competition in the general-purpose track and establish new moats in industrial gaps.

Liang's personal resume lends credibility to this attempt. His rare complete experience spanning from general-purpose mobile computing to dedicated AI acceleration gives him firsthand understanding of the strengths and weaknesses of both types of systems. Liang's all-in commitment has also attracted concentrated investment from internet platforms, automakers, semiconductor industry capital, and top-tier financial investors for Fangqing. What the investors are betting on, rather than a product already on the verge of success, is the validation of a differentiated underlying technology breakthrough path — within China's broader narrative of seeking technological self-reliance.

And Liang's "new departure" thus becomes a case worth watching: it tests whether, within existing global technological and supply chain constraints, architecture-level innovation can open up a substantive new option for China's AI computing industry.