The problem
Export controls on advanced semiconductors threaten to cap what Chinese AI labs can train and deploy. Dongfang Suanxin’s answer is architectural rather than diplomatic: change how the chip is built so the most restricted steps matter less.
How it works
The Shanghai Zhangjiang-based startup pairs software-defined chips — behaviour configured in software instead of fixed at tape-out — with 3D stacked near-memory computing, stacking compute next to memory to ease data bottlenecks, improve energy efficiency and carry large AI workloads like model training. The flagship DF1000 series is designed for production entirely on a domestic supply chain.
Pain points
Domestic AI chips trail on bandwidth and efficiency; every new design must prove it can be manufactured without restricted tooling; and well-funded giants are chasing the same 3D-stacking path, compressing the window for a startup to matter.
Business model
Sell the silicon: DF1000-series accelerators to Chinese cloud operators and AI labs, differentiated by software-defined flexibility that lets one chip family serve training and inference configurations competitors need separate silicon for.
Challenges
The company emerged from stealth with a website and a story, not silicon: no tape-out, no benchmarks, no customers and no disclosed financing. Manufacturing capacity and real 3D-stacking performance are the tests to watch — and US controls keep moving the goalposts.
Funding
- Raised: undisclosed (MISSING). No funding round has been announced since the 2024 founding.
- Valuation: MISSING.
- Status: idea — pre-product, pre-revenue, newly out of stealth.
Latest — July 2026
In the first week of July 2026 Dongfang Suanxin made its first public appearance, launching a corporate website and social-media presence and stepping into the spotlight as a new player in China’s AI computing sector (SCMP, 5 July 2026).