We Can Now Simulate a Human Brain, Scientists Show

TL;DR

A new parallel GPU method enables simulation of up to 20 billion neurons, putting full human brain simulation within reach of next-generation supercomputers.

Key Points

• 1.Brain simulation has scaled from worms to near-human levels over decades. Early simulations covered a 300-neuron worm, then a 140,000-neuron fruit fly by 2023; the Allen Institute reached 10 million mouse neurons, and the new method targets 20 billion neurons on the upcoming JUPITER supercomputer in Germany.
• 2.The breakthrough is a massively parallel, local GPU architecture. Instead of instantiating the entire network across a cluster, each GPU handles ~225,000–800,000 neurons independently; JUPITER's full cluster could reach nearly 20 billion neurons, with the next generation expected to cover all 80 billion in a human brain.
• 3.The prior Human Brain Project (2013–2023) spent over €1 billion and largely failed. It aimed to simulate millions of neurons but collapsed due to unclear goals and a rift between project leadership and neuroscientists who wanted more specific research — producing little beyond reports.
• 4.Major obstacles remain before a meaningful simulation is possible. No complete human connectome (neural wiring map) exists, training the simulation requires converting inputs into neural signals (poorly understood), and ethical questions arise about whether a simulated brain could think or suffer.