A little known Chinese chipmaker says it has developed a domestic alternative to Nvidia’s widely used AI training chips, marking the latest sign of China’s push to reduce reliance on Western semiconductor suppliers.
While the pitch invites comparison to DeepSeek’s software success, analysts warn that hardware performance claims require far more scrutiny.
Zhonghao Xinying — also known as CL Tech — claims its self-developed general-purpose tensor processing unit (GPTPU) entered mass production in 2023.
The chip, dubbed Chana, reportedly delivers up to 1.5 times the compute performance of Nvidia’s A100 GPU, while using 30% less energy for large model workloads and costing less than half as much (42%) per unit, according to a report in the South China Morning Post.
What is a TPU?
If accurate, the gains would be significant. TPU’s (tensor processing units) are a form of application specific integrated circuit (ASIC) designed to accelerate machine learning workloads.
Unlike GPU’s - originally built for graphics but now the backbone of AI training - TPU’s strip out general purpose features to maximise efficiency and throughput for neural network tasks.
Google pioneered the approach and recently began selling its TPU hardware directly to Meta and Anthropic, positioning itself more directly against Nvidia.
The Cybernews community is talking about this. Be a part of the conversation.
Analysts say the reported performance is ambitious but plausible.
“If true, that would be an impressive achievement, but not unheard of gains for an ASIC” Tom’s Hardware noted, adding that such chips can outperform GPUs by focusing tightly on specific workloads.
Challenging Nvidia’s monopoly
However, the hardware bible for PC and chip enthusiasts adds that even a 1.5 times improvement over the A100 would still leave Chana trailing Nvidia’s newer Hopper generation chips and far behind the forthcoming Black Ultra series.
Yet in China, where access to high end Nvidia hardware is restricted and older A100s are still being quietly imported - a domestic TPU offering competitive performance could make a significant difference.
It also comes at a time when both Western and Chinese firms are exploring alternatives to Nvidia’s near-monopoly in AI compute - from Google’s TPUs to a wave of specialist ASICs designed for large scale training.
Unlock exclusive Cybernews content on YouTube
Your email address will not be published. Required fields are markedmarked