Chinese chip startup Dongfang Suanxin has unveiled a new processor architecture that it says can compete with NVIDIA by reducing dependence on advanced semiconductor manufacturing restricted under US export controls. The company is betting on software-defined computing and 3D-stacked near-memory technology to improve computing performance without relying on the latest fabrication nodes.
The Shanghai-based startup introduced its flagship DF1000 processor during a launch event on Monday. Built on a 14-nanometer process, the chip is designed for computing workloads and is claimed to deliver 520 teraflops of BF16 performance, along with 6.4 TB/s of memory bandwidth and 900 GB/s of inter-chip communication bandwidth.
The company said the DF1000 is ready for mass production, with shipments expected to begin before the end of the year. It also outlined plans for two more processors, the DF2000 in late 2026 and the DF3000 in 2027, with the goal of surpassing NVIDIA’s H200 and B300 processors, respectively.
Rather than relying solely on smaller manufacturing nodes, Dongfang Suanxin aims to improve performance by redesigning how chips process and move data.
The strategy reflects a broader shift among Chinese chip companies that are exploring architectural innovations instead of relying solely on smaller manufacturing processes. As access to the most advanced fabrication technologies remains restricted, firms are increasingly looking at software optimization, memory design, and chip packaging to improve overall system performance.
Rethinking chip architecture
Its software-defined computing approach dynamically reconfigures computing and data-flow resources to match different workloads. At the same time, the company’s 3D-stacked near-memory architecture places memory closer to processing cores by stacking it vertically, reducing the distance data must travel.
According to the company, this helps lower latency, improve memory access and reduce energy consumption while easing dependence on the most advanced chipmaking technologies currently affected by US restrictions.
“We have to forge a path of our own,” founder Wei Shaojun said during the launch.
“That path cannot be about passively catching up within a framework set by others. We need independent architecture, original technology, a self-sustaining ecosystem and a secure, controllable supply chain.”
The startup also introduced supporting hardware around the DF1000, including the Dianfeng accelerator module, the TY64 supernode and the QY100 integrated computing appliance. To attract developers, it launched an open software stack called CAAP that supports mainstream development frameworks along with custom programming for operators, supernodes, and computing clusters.
Ambitious roadmap ahead
Despite the aggressive roadmap, Wei acknowledged that the company’s approach does not solve every challenge facing China’s semiconductor industry.
He noted that stacking multiple silicon layers can reduce manufacturing yields, while limited domestic access to advanced fabrication nodes remains the industry’s biggest obstacle to higher performance.
Dongfang Suanxin was founded about two years ago and is backed by state investment funds as well as venture capital firms linked to Xiaomi, JD.com, and Yunfeng Capital. The company said it plans to continue expanding its hardware and software ecosystem as it works toward competing with established global chipmakers.