China Launches ‘Brain-Inspired’ AI Model That Runs 100 Times Faster Without Nvidia Chips

A research team in Beijing has introduced what it describes as the world’s first “brain-like” large language model, designed to operate with lower energy consumption and without reliance on Nvidia hardware.

According to a report in the South China Morning Post, the system is named SpikingBrain 1.0. It was developed by the Institute of Automation at the Chinese Academy of Sciences. Unlike mainstream artificial intelligence tools such as ChatGPT, which activate entire networks during processing, SpikingBrain reportedly mimics the human brain by triggering only the necessary neurons. This selective approach allows the model to conserve power while responding more quickly to input.

The researchers reported that SpikingBrain requires less than 2% of the training data normally consumed by conventional AI systems. Despite the smaller dataset, the model maintained performance across lengthy inputs and, in some tests, ran up to 100 times faster than traditional systems. The findings were outlined in a technical paper posted on the open-access research repository arXiv, though the study has not yet undergone peer review, SCMP stated.

The model is powered by the MetaX chip platform, developed in Shanghai, making it independent of Nvidia’s widely used graphics processors. This is seen as significant as Washington imposes tighter export controls on advanced AI chips.

Li Guoqi, a lead researcher at the institute, told SCMP that the project “opened a new path for AI development” while creating a framework optimised for China’s domestic hardware.

Li said the system could be particularly effective in analysing long and complex material such as legal documents, medical records and scientific data. A smaller version of SpikingBrain has been released as open source, while a larger model is available for public testing.

On its demo site, the model introduces itself as, “Hello! I’m SpikingBrain 1.0, or ‘Shunxi’, a brain-inspired AI model. I combine the way the human brain processes information with a spiking computation method, aiming to deliver powerful, reliable, and energy-efficient AI services entirely built on Chinese technology.”

The system’s core relies on spiking computation, an event-driven technique that mirrors the brain’s method of sending signals only when needed. By avoiding continuous full-scale processing, the model reduces strain on memory and energy while still managing complex tasks.

As reported by the South China Morning Post, the researchers built two versions of the model, one containing seven billion parameters and another with 76 billion. Both were trained using hundreds of MetaX chips. The results, reportedly, matched the performance of several open-source large models.

The team highlighted SpikingBrain’s handling of long sequences. In one trial, the smaller model processed a prompt of four million tokens over 100 times faster than a standard system.