A Fruit Fly Brain on a Laptop
What 139,255 neurons tell us about the future of compute
In October 2024, a team at UC Berkeley published something quietly remarkable in Nature: a complete simulation of an adult fruit fly brain — all 139,255 neurons and roughly 50 million synaptic connections — running on a single laptop.
Not a cluster. Not a supercomputer. A laptop.
The simulation accurately predicted which neurons would fire in the real fly brain when stimulated. Same paper, same issue of Nature, as the FlyWire consortium's announcement of the full connectome (Dorkenwald et al. 2024). Two papers, one day, one shift in what's possible.
The most complex object ever fully simulated in biology runs at laptop scale. The code is open. The hardware is off-the-shelf.
I want to sit with what this means, because the headline is "we simulated a brain" and the actual story is "we did it on hardware that fits in a backpack."
The compute vs. interpretation question
The standard narrative in 2026 is that artificial intelligence is winning because it can throw more compute at every problem. Bigger models, bigger clusters, bigger electricity bills. The data centres being built this year in Virginia, in the UAE, in Galicia, are sized for the assumption that the next breakthrough will need the next order of magnitude of power.
The fruit fly simulation is the counter-evidence. The most complex object ever fully simulated in biology runs at laptop scale. The paper's lead, Phil Shiu, was a postdoc when he did this. The team is small. The code is open. The hardware is off-the-shelf.
This is the kind of result that doesn't just advance one field — it tells you something about the shape of what's about to be possible. If a fly brain — 139,255 neurons, the result of 400 million years of evolution compressing intelligence into a creature that fits on your fingertip — can be simulated on a laptop, then a lot of what we currently assume requires a data centre is, in fact, over-provisioned.
What the headlines missed
There is a second thing in this paper that the headlines missed.
A full connectome is not just a map. It is also an argument. To simulate a brain cell-by-cell, you have to decide which connections matter, which are noise, which are signals. The FlyWire consortium spent over a decade drawing the diagram. The simulation, once the diagram existed, took the time of a single project.
The work was not "compute." The work was interpretation. Reading a brain. Deciding what to count.
This matters because the same pattern is showing up in the other stories I've been following this year: the mycelium sector turning into a real industry, the small labs growing building materials from fungi, the climate DAOs that are buying tropical dry forest in Colombia, the experiments showing plants reorganise their cellular state under anaesthesia. None of these are compute problems. They are interpretation problems. We are not short of signal. We are short of people who can read signal.
The question this leaves me with
If I had to pick one question this leaves me with, it's this:
If a fly brain runs at laptop scale,
what else is over-provisioned?
Not just in AI. In everything we assume needs a bigger machine, a bigger budget, a bigger team. The history of technology is full of moments where the thing everyone thought needed a mainframe turned out to fit in a pocket. The fly brain is one of those moments. It's just happening in biology first.
What's something you read recently that made you recalibrate your sense of what's possible?
Sources
- Berkeley News — Researchers simulate an entire fly brain on a laptop
- Nature — Simulated fly brain is powered by the data behind it
- FlyWire browser demo
- Dorkenwald et al. 2024, "Neuronal wiring diagram of an adult brain," Nature (full connectome paper)
Published on the Sotabosc blog, 2026. Image credit: FlyWire consortium, © MRC LMB, University of Cambridge.