According to MyDrivers, the pair are cooking a new generation of ‘Fuxi’ GPUs with “excellent performance” in parallel workloads, pitched at AI training and the ‘AI PC’ crowd.

Two parts are on the slate, the Fuxi A0 for rendering and AI PC jobs, and the Fuxi B0 with an onboard NPU for end-side model deployment.

The headline number is up to 160 TFLOPS FP32 and, more eyebrow-raising, a 5nm process claim.

That process talk jars with China’s domestic shops being stuck at 7nm and persistent chatter about TSMC 5nm turning up in the region, which the companies have not confirmed.

The B0 is said to back mainstream models such as DeepSeek R1, which would make it a tidy turnkey for local deployments if the software lands on time.

The sales pitch fits Beijing’s “not rely” stance on Nvidia and pushes the familiar narrative of swapping in Huawei and Cambricon for Western kit.

Hard details are thin, timelines thinner, and the small matter of who their foundry will be has not been answered.