A Supercomputer Just Simulated a Whole Human Brain at Full Scale

A team of researchers has published a paper describing CerebroSim, software that simulates an entire human brain at full scale: 86 billion neurons (brain cells) and 100 trillion synapses (the tiny junctions where brain cells pass signals to each other). It ran on LineShine, a Chinese supercomputer spanning 18,432 machines and 11.2 million processor cores. For context, that's more computing power than most countries own, pointed at one problem — and it hummed along at roughly 24 quadrillion calculations per second.

Why is that hard? Because brain wiring is messy. Signals arrive at different times, and the connections are mostly empty space — imagine a road map where only a tiny fraction of streets ever have traffic, but you still have to track every one of them. The team's clever bit was three software tricks: letting millions of processors work at once without tripping over each other, storing that mostly-empty wiring compactly, and rebuilding connections on the fly instead of keeping them all in memory.

The payoff matters to anyone whose family has been touched by brain disease. The model is built from real MRI scans (detailed pictures of a living brain), so researchers can run experiments that would be slow, expensive, or unethical on people: switch off a gene and watch what breaks, mimic a stroke, or test whether a drug calms overactive circuits. Clinical trials cost billions and take a decade; a simulation can screen thousands of ideas first. It could also reduce animal testing.

Now the honest caveat. This is a simulation of structure and electrical activity — not a thinking, feeling brain. There's no consciousness here, no personality, no mind. It's also slow: simulating one second of brain activity still takes far longer than one second. The underlying data comes from a limited set of scans, and anything it suggests must still be checked in real patients. Think flight simulator, not actual airplane.