China's 5nm 'Fuxi' AI GPUs Reportedly in Development
As captured in a recently published report, China has apparently advanced in the direction towards 5nm AI GPU development. The local companies, Anfu Technology, and Xiangdi, are said to work together for the possibility of producing a new generation of chips that will be named 'Fuxi' to be suitable for high performance in parallel computing workloads for AI training and AI PC markets.
'Fuxi' GPU Performance and Manufacturing
The forthcoming 'Fuxi' GPUs are touted to have as much as 160 TFLOPS in FP32 compute performance. A critical point, as China has not, for its domestic semiconductor production, publicized the scale extending beyond the 7nm node.
The report speculates that Anfu Technology may have utilized nodes from an outside manufacturer like TSMC to get this end, but the veracity of this remains unproven. Nevertheless, it shows an intense commitment toward a computational breakthrough within China's AI industry.
Double Chip Strategy for Different Markets
The strategy here accounts for an even-chip pair for different markets coverage:
- Fuxi A0: This would be an execution-based chip with an orientation towards "AI PC" workloads.
- Fuxi B0: This is an AI chip designed with an on-board NPU that will be deployed at the endpoint model. It will support mainstream AI models such as DeepSeek R1.
Part of a Broader Self-sufficiency Initiative
Though much is yet to be known about these new solutions for artificial intelligence, their development is part of a much larger initiative by the Chinese government to organically develop technologies where it does not have to depend upon Western technology, especially NVIDIA's AI chips. In doing so, the capacity should be attained entirely by building a self-sufficient tech stack through domestic companies like Huawei and Cambricon.
This is an example of how rapidly the scaling of China's compute capabilities progresses, as demonstrated by this development.
