- Chinese startup Dongfang Suanxin (DFSX) has officially introduced its self-designed AI chip, the DF1000, in Shanghai, while announcing a roadmap to develop its second-generation chip by late 2026 and its third-generation chip by 2027.
- Founded in 2024, DFSX previously operated quietly to focus entirely on research. Following its latest funding round, the company is valued at approximately 1.8 billion USD.
- DFSX’s investors include numerous Chinese state-backed institutions and industrial funds, including a venture capital fund co-founded by Jack Ma.
- DFSX stated that the DF1000 chip was developed entirely using a domestic supply chain to reduce dependence on foreign technology and bypass Western export restrictions on advanced chips.
- Instead of using High Bandwidth Memory (HBM) like Nvidia, DFSX adopts a customized memory stacking architecture directly on the processing layer to increase data bandwidth without importing HBM.
- Although the DF1000 chip is manufactured on a 14 nm process, the company claims it can achieve performance equivalent to certain Western chips using a 4 nm process in specific AI inference tasks. DFSX acknowledges that a gap still remains in the training phase, but expects the DF2000 to narrow this difference.
- Led by Professor Wei Shaojun of Tsinghua University, the company commercializes over 20 years of research by placing memory close to the processor and allowing the chip to self-adjust its internal architecture based on the software being run.
- Experts note that if DFSX succeeds, it will prove that China can build a competitive AI ecosystem using domestic technology. The trend of optimizing hardware for specific applications is also viewed as a crucial future direction for the AI industry.
- 📌 Conclusion: DFSX marks a new step in China’s ambition for semiconductor self-sufficiency. With the DF1000 chip utilizing an entirely domestic supply chain, a valuation of around 1.8 billion USD, and a continuous development roadmap leading to 2027, the company aims to prove that AI does not necessarily have to depend on advanced Western technologies. Although its performance claims still require independent verification, the approach of optimizing architecture rather than merely shrinking process nodes reflects a new direction in the global AI chip competition.
VIET NAM CONSULTING AND MEASUREMENT JOINT STOCK COMPANY
© 2026 Vietmetric