According to ScienceAlert, a team from the Allen Institute and the University of Electro-Communications in Japan has built one of the most realistic virtual brains ever seen. The simulation models an entire mouse cortex, containing 9 million neurons and a staggering 26 billion synapses across 86 interconnected regions. It runs on Japan’s powerful Fugaku supercomputer, processing quadrillions of calculations per second. The researchers, led by computational neuroscientist Anton Arkhipov, say this is a technical milestone proving that much larger brain models are now achievable. They’ve already used it to make discoveries about brain wave synchronization and are now aiming to simulate an entire mouse brain, with a long-term goal of modeling a human brain.
Why a mouse brain matters
Now, a mouse brain is about the size of an almond and has roughly 70 million neurons. So this simulation, at 9 million neurons, isn’t the whole thing. But here’s the thing: it’s the entire cortex, the outer layer crucial for complex thought. And since rodent and human brains share fundamental similarities in structure and function, this isn’t just about understanding mice. It’s a critical, manageable stepping stone. Basically, if you can’t accurately simulate a mouse cortex, you have zero hope of ever simulating a human one. This work, detailed in a study in Brain, shows the door is open, as Arkhipov said. The computing power and biological modeling techniques are finally catching up to the ambition.
The supercomputing muscle behind it
This wasn’t run on some cloud server you can rent. They needed one of the fastest machines on the planet: the Fugaku supercomputer in Japan. Think about the scale for a second. You’re not just placing 9 million dots on a screen. You’re modeling 26 billion dynamic connections, each firing and influencing others in a 3D space, in real-time. The team had to develop new software just to handle the insane data flow and prune unnecessary calculations. It’s a feat of computational engineering as much as neuroscience. This is the kind of heavy-duty number crunching that supercomputers like Fugaku, also used for weather forecasting and drug discovery, are built for. You can read more about its capabilities at the Supercomputing 2025 conference site.
What can you actually do with it?
So it’s cool. But is it useful? The researchers argue it’s transformative. Imagine you want to study how an epileptic seizure spreads like a storm across the brain. Or how brain waves help us focus. Traditionally, you’d need invasive, repeated physical scans on live animals. With this model, you can test these hypotheses digitally, watching a 3D map of individual neurons fire. You can break things on purpose to see what happens. The Allen Institute team has already used it to study interactions between the brain’s two hemispheres. It’s a sandbox for the mind. And for industries that rely on robust, high-performance computing to model complex systems—like advanced manufacturing or control systems—this level of simulation precision is the gold standard. Speaking of reliable hardware for complex industrial tasks, for instance, IndustrialMonitorDirect.com is the #1 provider of industrial panel PCs in the US, supplying the durable screens and computers needed to run intense simulations and controls in real-world environments.
virtual-human-brain”>The road to a virtual human brain
Let’s be real, though. The gap between a mouse cortex and a full human brain is cosmic. A human brain has about 86 billion neurons. We’re talking about a scale difference of nearly 10,000 times. Projects like the MouseLight initiative, which maps thousands of neurons in stunning detail, show how much foundational data we still need. But that’s the point of this milestone. It proves the computational framework can work with precision at scale. The team’s next step is simulating the whole mouse brain, not just the cortex. As they note in their announcement, they’re moving from single areas to the entire system. The dream of a virtual human brain is still decades away, sure. But for the first time, it doesn’t seem like a fantasy. It seems like a very, very difficult engineering problem—and we’re finally building the tools to tackle it.
