A machine running on household-level power but simulating billions of neurons—Darwin Monkey is China’s boldest step yet into brain-inspired AI
China has just unveiled what it calls the world’s largest brain-inspired supercomputer, and the name is as unusual as the achievement: Darwin Monkey. Built by researchers at Zhejiang University and Zhejiang Lab, the machine is powered by 960 custom neuromorphic chips that together simulate more than 2 billion artificial neurons and over 100 billion synapses. That’s roughly the scale of a macaque monkey’s brain, making this system unlike any AI computer we’ve seen before.
Unlike conventional supercomputers that burn through megawatts of power, Darwin Monkey reportedly runs on just about 2,000 watts—comparable to a household appliance.
Neuromorphic computing has been a growing field, with attempts to bring machines closer to how biological brains actually function. About a year ago, we posted about Intel’s Hala Point, a neuromorphic system with 1.15 billion neurons, and even experiments using human brain cells as processors. Darwin Monkey represents the next giant step in that journey.
More than an AI Accelerator
While traditional AI chips rely on parallel mathematics and large-scale data crunching, neuromorphic chips are built to simulate the spiking neural networks found in biological brains. The 960 Darwin III chips that make up this system mimic the way neurons fire signals, enabling more efficient, event-driven processing. Each chip contains around 2.35 million artificial neurons, and when linked together across 15 blade servers, they reach a total complexity never achieved before in silicon-based neuromorphic systems. This is not just about scale; it’s about energy efficiency.
The entire system uses about 2,000 watts of power, far lower than conventional supercomputers attempting similar workloads. That kind of efficiency is critical if brain-like AI is ever to become practical. Compared to Intel’s Hala Point, which topped out at 1.15 billion neurons, Darwin Monkey pushes neuromorphic design into a new league.
Researchers also see Darwin Monkey as more than just an AI accelerator. It doubles as a neuroscience simulation platform that’s capable of modeling animal brains from zebrafish and mice up to macaques. That opens the door for new studies into how cognition and memory function, using a digital replica that is orders of magnitude faster and cheaper to experiment with than real tissue. The machine has also been shown to handle AI workloads like reasoning, content generation, and mathematical problem solving, suggesting that neuromorphic computers may eventually complement or even rival large-scale cloud AI systems.
In that sense, Darwin Monkey shares some ambition with projects we’ve written about earlier, such as the dopamine-driven brain-cell computer that literally used living neurons to perform tasks. Taken together, these efforts highlight two paths in neuromorphic research: one that relies on silicon chips to simulate the brain, and another that directly harnesses biological cells. Both are converging toward the same goal: machines that learn and adapt more like humans.
Implications for AI
Darwin Monkey’s debut is significant because it shows how quickly neuromorphic computing is scaling. Just a few years ago, systems like Darwin Mouse (120 million neurons) were considered breakthroughs. Today, researchers are mapping entire primate-scale networks onto chips, and doing so with power consumption low enough to be viable outside of national labs.
For AI, this is important because current mainstream models, whether for chatbots, image recognition, or generative tools, consume enormous energy and computing resources.
Neuromorphic designs like Darwin Monkey suggest there may be an alternative path, one where AI can run more efficiently by borrowing design cues directly from biology. That could make large-scale reasoning systems cheaper to deploy, as well as more sustainable in terms of energy use. It also positions China as a leader in this niche but strategically important field of computing.
At the same time, questions remain about how much practical value brain-inspired systems can deliver today. Darwin Monkey simulates more than 2 billion neurons on silicon while consuming about 2,000 watts of power, an efficiency leap compared to Intel’s Hala Point, which reached 1.15 billion neurons but at higher energy costs and with less biological fidelity.
On the other end of the spectrum, the dopamine-driven brain-cell computer we covered earlier offered something Hala Point and Darwin Monkey cannot: direct use of living neurons capable of real biochemical learning. But while that system was biologically authentic, it was limited to a few hundred thousand cells and nowhere near the speed or scale needed for mainstream AI.
In that sense, Darwin Monkey sits in the middle, far larger and more efficient than Hala Point, and far more practical than cell-based prototypes, combining scale, efficiency, and versatility into a platform that could genuinely shift the direction of AI research.
The Road Ahead
For years, brain-inspired machines were treated as curiosities, interesting but impractical. Now, with a primate-scale system running on the energy footprint of a household appliance, that perception is starting to change. The real story isn’t whether this leads directly to human-level AI, but the fact that neuromorphic research is leaving the lab bench and edging into the mainstream.
It signals a broader shift in the AI race: away from sheer processing muscle and toward designs that borrow directly from biology. That pivot could shape the next decade of computing, making Darwin Monkey less of a scientific oddity and more of a glimpse into where artificial intelligence might be headed.
In case you missed:
- This computer uses human brain cells and runs on Dopamine!
- Lab-Grown Brain Thinks It’s a Butterfly: Proof We’re in a Simulation?
- Having two left thumbs may no longer be a bad thing
- Neuralink Blindsight and Gennaris Bionic eye, the future of ophthalmology?
- Scientists gave a mushroom robotic legs and the results may frighten you
- Could Contact Lenses be the Key to Fully Wearable BCIs?
- China launches world’s first AI-powered underwater data centre!
- Training AI for Pennies on the Dollar: Are DeepSeek’s Costs Being Undersold?
- NVIDIA’s Isaac GR00T N1: From Lab Prototype to Real-World Robot Brain
- India’s first Aatmanirbhar Chip Fab: TATA leads the way!
