BEIJING (Diya TV) — Chinese scientists say they have built the world’s first “brain-like” artificial intelligence large language model that runs without U.S. chips and uses far less energy than today’s leading systems.
Researchers at the Chinese Academy of Sciences’ Institute of Automation in Beijing have unveiled SpikingBrain 1.0, a large language model inspired by the human brain. Unlike ChatGPT and other popular AI tools, it does not activate an entire network to process every word. Instead, it selectively responds to nearby words, much like how neurons in the brain fire only when needed.
This design allows the system to work faster while using less power. The team said SpikingBrain can run up to 100 times faster than traditional AI models in some cases. It also operates on China’s MetaX chip platform, avoiding dependence on Nvidia’s specialized processors.
Most large language models, such as ChatGPT and Meta’s Llama, rely on a method called “attention.” This means they compare every word in a sentence with every other word to make predictions. The approach is powerful but demands huge amounts of data, computing power, and energy.
For example, when given a sentence like, “The cat went under the bed after seeing the stranger because it was scared,” ChatGPT processes the entire sentence at once to understand that “it” refers to the cat. While effective, this technique consumes vast resources when applied to long texts, such as books or legal records.
By contrast, SpikingBrain uses what the researchers call “spiking computation.” This mimics the way neurons send short bursts of signals only when triggered. The system stays quiet most of the time, firing only when it needs to process new information. That makes it leaner, faster, and more energy-efficient.
Training a large language model typically requires billions of words and enormous computing power. The researchers behind SpikingBrain say their system needs less than 2 percent of the data that mainstream models use. Despite the smaller input, it still achieved results on par with popular open-source systems.
The team tested two versions of SpikingBrain: one with 7 billion parameters and another with 76 billion. Both ran on China’s MetaX chips, produced by Shanghai-based MetaX Integrated Circuits Co. In one test, the smaller version processed a prompt with 4 million tokens more than 100 times faster than a standard model.
The study, published on the open-access site arXiv but not yet peer-reviewed, reported that the system ran stably for weeks across hundreds of MetaX chips while using far less power than conventional AI systems.
The launch of SpikingBrain comes as China works to reduce its reliance on American technology. Since former President Donald Trump’s administration, the U.S. has tightened export controls on advanced chips and tools used for artificial intelligence.
Nvidia, the California-based chipmaker that dominates the AI hardware market, has been hit especially hard by these restrictions. Its most advanced GPUs, which power many of the world’s leading AI models, can no longer be exported to China.
By building SpikingBrain on the MetaX platform, researchers hope to show that China can create advanced AI systems using its own technology. Lead researcher Li Guoqi said the model opens a new path for AI development optimized for Chinese chips.
The researchers say SpikingBrain could be useful for processing long sequences of information, such as medical records, legal documents, or scientific data. Its efficiency may also help reduce the environmental impact of AI, which currently consumes vast amounts of electricity and water for cooling.
Li’s team has released a smaller open-source version of the model and made a larger one available online for public testing. The demo greets users with the message: “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.”
Today’s leading AI systems, including ChatGPT, require massive data centers filled with high-end chips to function. They remain costly to run and difficult to scale. SpikingBrain offers a possible alternative by using brain-inspired mechanisms to cut energy use while maintaining strong performance.
The researchers say their work shows the promise of efficient, scalable AI models that can operate outside the Nvidia ecosystem. As the U.S. continues to tighten technology exports, China’s efforts to build a homegrown AI industry are gaining momentum.
“Overall, this work demonstrates the potential of brain-inspired mechanisms to drive the next generation of efficient and scalable large model design,” the team wrote in its study.
At a time when the global race for AI dominance is intensifying, SpikingBrain represents both a technological achievement and a strategic step for China.
