Chinese AI chip start-up Biren sees performance boost by teaming with Tencent-backed firm

Artificial intelligence (AI) chip start-up Biren Technology, considered key to China’s efforts to reduce reliance on US chipmaker Nvidia, said it has significantly increased the performance of its hardware when training large language models (LLMs) by working with a Tencent Holdings-backed supplier of computing power, or compute.

Biren, which Washington placed on a trade blacklist last year, has conducted a “deep” collaboration in LLM development and reasoning with Infinigence AI, boosting the training capacity of its graphics processing units (GPUs) by nearly 100 per cent, the chipmaker said in a statement on Tuesday.

Infinigence offers on-demand compute based on chips from brands including Nvidia, AMD, Huawei Technologies, Cambricon, Iluvatar Corex, and Moore Threads via its cloud platform Infini AI, which it unveiled in April.

The platform integrates 10 different chips, and has been used by leading Chinese AI start-ups Zhipu AI and MiniMax to develop models and applications. Infinigence added Biren’s GPU cluster to Infini AI on Tuesday, along with the associated software development tools, Biren said.

China’s chip development ecosystem continues to show resilience amid concerns of technological decoupling, as the US has continued to tighten access to advanced GPUs and other semiconductor technologies. Many Chinese AI start-ups face challenges in accessing enough compute resources for developing AI models. Difficulties include obtaining pricey Nvidia GPUs and lacking the technological skills needed to best optimise hardware for specific software.

Infinigence co-founder Xia Lixue said in September that his company intends to become a “computing power infrastructure operator” to provide compute in a way that is similar to utilities such as water and gas, according to local media outlet China Star Market Daily.

  

​Read More