Nvidia’s new rivals: 5 companies building next-gen compute silicon

Summary

A piece listing Huawei, Cambricon, Moore Threads, and Biren Technology as rising rivals to Nvidia in AI compute chips, triggered by US export restrictions and Beijing's push for technological independence.

Why it matters

The AI compute market is shifting from a single-supplier monopoly to a fragmented, geopolitically divided structure. Huawei's support for DeepSeek V4 on its Ascend 950 series chips via the CANN platform is the most concrete signal that a functional CUDA alternative is beginning to take shape. If even a fraction of these ecosystems matures, the global AI infrastructure stack could split along national lines.

Highlights

• Huawei's integration of CANN and DeepSeek V4 on its Ascend 950PR/950DT chips provides the strongest evidence of a functional CUDA replacement, despite being in early stages.
• Biren projects revenue growth of up to 2.107% in the first half. However, the article itself acknowledges that this sales surge does not imply performance or software compatibility with Nvidia.
• The core competitive advantage lies not in raw FLOP values, but in a complete ecosystem encompassing processors, network infrastructure, and developer tools. The article touches on this point but fails to support it with any quantitative data.

Critical take

The article reads more like a promotional list than an analysis. It lacks benchmark data, manufacturing process comparisons, and independent performance measurements. The claim that 'China still uses Nvidia' is supported only by a single SCMP source, with no other citations provided. Biren's 2.107% growth figure is presented without a baseline for comparison. Consequently, it is impossible to determine whether this increase represents a genuine gain in market share or is merely a percentage effect resulting from a low starting point.