China’s New 5nm AI GPUs: Fuxi A0 and B0 Target PCs and Model Training

China’s semiconductor ambitions have taken another high-profile step forward, with reports surfacing that domestic firms Anfu Technology and Xiangdi are preparing 5nm-class AI GPUs under the ‘Fuxi’ brand. The chips – codenamed Fuxi A0 and Fuxi B0 – are said to be designed specifically for different computing needs: AI training workloads, AI PCs, and rendering tasks.

If realized as described, these chips would represent one of the most advanced semiconductor efforts yet announced from within China’s borders.

The Fuxi A0 reportedly targets AI PC and rendering markets, potentially offering local alternatives to GPUs typically imported from Western manufacturers. Meanwhile, the Fuxi B0 is aimed at AI inference and training, equipped with an onboard NPU to accelerate popular models such as DeepSeek R1. The claims of up to 160 TFLOPS FP32 performance put these chips in the same conversation as some Western-designed accelerators, though details on yield rates, efficiency, and real-world performance remain unknown.

The use of a 5nm process node is what truly raises eyebrows. Until now, China’s domestic semiconductor ecosystem had only been confirmed at 7nm, with speculative chatter about experimental 5nm production via multi-patterning DUV lithography. The possibility that TSMC’s technology may be indirectly involved hasn’t been dismissed either, fueling skepticism about whether this is a fully homegrown milestone or partially reliant on foreign fabs. Still, the symbolic importance of aligning with a 5nm-class label is clear: it signals China’s determination to claim parity in a sector where Western and Taiwanese firms, particularly TSMC, dominate.

Experts caution, however, that nanometer classifications are not always straightforward indicators of performance. The efficiency gap between Samsung’s so-called 8nm (used in NVIDIA’s RTX 3000 GPUs) and TSMC’s true 7nm and 5nm nodes demonstrates that marketing can blur real engineering capabilities. The difference in power consumption and performance-per-watt between NVIDIA’s 3000 and 4000 series – where TSMC processes delivered cooler, quieter, and far more efficient GPUs – is a reminder that numbers alone do not tell the whole story.

For Beijing, the political and economic stakes are high. Building a self-sufficient chip supply chain is seen as essential to shield China’s AI ecosystem from restrictions on NVIDIA and other Western suppliers. Even if yield rates on advanced nodes remain challenging, the announcement of Fuxi A0 and B0 positions Anfu and Xiangdi as players in China’s push for technological sovereignty. Whether these chips can truly rival established giants in efficiency, scalability, and adoption remains uncertain, but the narrative is unmistakable: China intends to push forward regardless of hurdles, betting that breakthroughs will arrive sooner rather than later.