New AI chip to boost computing hardware

DF1000, a 3D AI computing chip developed by Dongfang Suanxin, is on display at the 2026 World Artificial Intelligence Conference. [Photo/IC]

China's first near-memory computing 3D artificial intelligence chip was recently unveiled in Shanghai by AI chip startup Dongfang Suanxin, marking an important step forward in the nation's efforts to develop independent, high-end computing hardware.

Formally named DF1000, the chip enters the market at a time when computing power has become a crucial pillar of national competitiveness and a key driver of industrial growth. As global technological competition intensifies and external supply chain constraints tighten, China faces significant hurdles in obtaining high-end AI chips and advanced semiconductor manufacturing capabilities, according to industry experts.

The new chip addresses a fundamental bottleneck in chip design: memory access speed. In traditional computers, data constantly travels back and forth between a separate processor and the memory, which creates a slowdown known as the "memory wall". Dongfang Suanxin's chip uses near-memory computing — placing memory components directly alongside the processing unit — and stacks them vertically in a 3D design. This structure significantly speeds up data transfer while lowering power consumption, allowing high performance without relying on ultra-advanced manufacturing equipment.

"As the nation's first software-defined 3D AI chip powered by near-memory computing technology, its launch marks an important step toward overcoming one of the key bottlenecks in high-end computing chips, namely their dependence on advanced semiconductor manufacturing processes," said Wei Shaojun, chairman and CEO of Dongfang Suanxin.

The technology behind the chip builds on more than two decades of research. Founded on May 20,2024, Dongfang Suanxin specializes in integrated circuit design, focusing on self-sufficient software ecosystems and original hardware architectures.

Looking ahead, the company plans to introduce next-generation chips designed for AI model training — the computing-heavy process where AI algorithms learn from massive datasets.

"Our current chips focus on the training level, while the next-generation products are expected to tape out — meaning they will complete the final design phase before manufacturing — by the end of this year, with performance reaching the level of mainstream international peers," Wei said during an interview earlier in July.

Wei added that the company will follow a strategy of "mass-producing one generation, developing the next and researching future generations", while collaborating with supply chain partners to build a secure and self-controlled computing ecosystem.

The technological breakthrough highlights Shanghai's growing role as a national hub for chip innovation, particularly in the Pudong New Area.

Li Hui, deputy head of Pudong New Area, said Pudong's integrated circuit industry expanded by about 23 percent year-on-year in 2025, with total industrial output exceeding 360 billion yuan ($53.11 billion). The district accounts for about one-fifth of China's total integrated circuit industry and roughly three-quarters of Shanghai's overall output.

Overall, Shanghai has built a complete innovation ecosystem supported by strong talent and advanced infrastructure. In the first eleven months of 2025, Shanghai's revenue generated from the integrated circuit industry totaled 391.2 billion yuan, up 23.72 percent year-on-year. Major breakthroughs like the DF1000 chip are expected to give fresh momentum to Pudong's 3D integrated circuit cluster.

Shanghai is currently home to more than 1,200 integrated circuit companies, roughly 40 percent of China's semiconductor talent, and nearly half of the country's innovation resources in the sector.