Real-time human-brain-scale neural simulation
Compute, memory and interconnect able to simulate ~1e11 neurons and ~1e15 synapses at biological speed, continuously and affordably.
State of the art
Published projections place cellular-level whole-mouse-brain simulation around 2034 and human later than 2044 on supercomputer trend lines; gigawatt-class AI compute campuses are the nearest industrial analogue.
- Synaptic state alone needs petabytes of high-bandwidth memory, which dominates cost
- All-to-all spike traffic makes the problem interconnect-bound, not FLOP-bound
- Energy per synaptic event is ~1e6 times worse than biology, so a real-time human emulation implies large power draw
Who’s working on it
- Anton ArkhipovResearcherAllen Institute for Brain Science
Co-led the November 2025 whole-mouse-cortex simulation of 9M biophysical neurons and 26B synapses on Fugaku, presented at SC25
- Tadashi YamazakiResearcherUniversity of Electro-Communications, Tokyo
Co-led the same work and built Neulite, the simulator that made biophysical neuron dynamics tractable at that scale
- Allen Institute for Brain Science ↗Academic lab
Supplies the Cell Types Database, Connectivity Atlas and Brain Modeling ToolKit that turn measured anatomy into runnable models
- RIKEN Center for Computational Science ↗Government lab
Operates Fugaku (>400 petaflops), the machine on which the current largest biophysical cortex simulation ran, still far slower than biological real time
- SpiNNcloud Systems ↗Startup
Dresden spin-out commercialising SpiNNaker2; the TU Dresden system launched April 2025 with ~35,000 chips and 5M cores, with multi-rack configurations claimed to support 10B neurons in real time
- Christian MayrResearcherTU Dresden
Chair behind the SpiNNaker2 architecture, which targets sub-millisecond real-time operation rather than offline supercomputer runs
- Intel Neuromorphic Computing LabCompany
Hala Point (2024) reached 1.15B neurons on Loihi 2, the largest neuromorphic deployment, though at point-neuron rather than biophysical detail
What to invest in
| Kind | Investment | Scale |
|---|---|---|
| research | Neuromorphic and in-memory-computing architectures benchmarked on spiking workloads at mouse-brain scale | $$$ $100M – $1B |
| infrastructure | A dedicated brain-simulation machine with petabyte-class fast memory and spike-optimized fabric | $$$$ > $1B |