Chinese AI Chip Startup Dongfang Suanxin Embraces 3D Stacking Amid Export Restrictions

Dongfang Suanxin, a Shanghai startup specializing in artificial intelligence chips, has revealed its presence publicly for the first time, marking a step in China's efforts to establish an independent AI computing ecosystem.

Founded in 2024, the company launched its corporate website and social media accounts this week, signaling its entry into the competitive semiconductor market.

Wei Shaojun, a professor at Tsinghua University and vice-president of the China Semiconductor Industry Association, leads the startup.

Although the AI chip sector is heavily dominated by established firms such as Huawei Technologies, Dongfang Suanxin's emergence coincides with ongoing US export restrictions that limit China's access to advanced semiconductor manufacturing equipment and high-end AI processors.

The startup emphasizes two primary technologies: “software-defined chips” and “3D stacked near-memory computing.” It asserts that its products are developed entirely through a domestic supply chain, aligning with China's objective to reduce reliance on foreign semiconductor technologies.

3D Stacking as a Strategic Semiconductor Approach

Dongfang Suanxin’s adoption of 3D chip architecture mirrors a broader industry trend. As the pace of transistor scaling slows beyond Moore’s Law, chip manufacturers increasingly explore stacked designs to enhance performance without depending solely on smaller manufacturing nodes.

This method has gained traction in China following Huawei's introduction of the “Tau Scaling Law” in late May, which advocates for 3D architectural innovations to achieve performance improvements comparable to increased transistor density, potentially circumventing restrictions on advanced-node technologies.

For China, 3D stacking serves not only as a performance enhancement but also as an industrial workaround to existing limitations.

By integrating compute and memory components more closely through near-memory computing, data transfer bottlenecks can be minimized, energy efficiency improved, and support for large AI workloads—such as model training and inference—enhanced.

The company’s flagship product, the DF1000 series, utilizes a “3D near-memory computing architecture.” According to Dongfang Suanxin, this accelerator supports both training and inference tasks for large AI models and is compatible with AI server platforms from leading domestic equipment manufacturers.

Company Profile and Expansion

Dongfang Suanxin is based in Shanghai’s Zhangjiang high-tech zone, a major hub for semiconductors and biotechnology in China. Despite its recent establishment, the startup has expanded to over 500 employees.

Its research and development operations extend across seven Chinese cities, including Beijing and Shenzhen, indicating ambitions beyond a niche design firm toward becoming a significant domestic AI computing player.

The startup’s investors reportedly include the National AI Industry Investment Fund, Yunfeng Capital backed by Jack Ma, and the venture capital arms of Xiaomi and JD.com, reflecting substantial financial support.

Emphasis on Developing Semiconductor Talent

Wei Shaojun has been a prominent advocate within China’s semiconductor sector. At an industry conference in Shenzhen last month, he called for enhanced policies to cultivate elite chip professionals, citing a shortage of high-end talent.

He also pointed out the fragmented state of China’s chip design industry, noting that 87% of domestic chip design companies employ fewer than 100 people.

Earlier this year, Wei stated in an interview with the Global Times that China’s semiconductor industry must maintain “high vigilance” while advancing domestic capabilities in cutting-edge technologies.