MIT researchers propose astrocytes could unlock vast untapped memory capacity in the brain

For decades, neuroscience has operated under a straightforward assumption: the brain's roughly 86 billion neurons do the heavy lifting when it comes to memory and thought. Everything else—the supporting cast of cells that fill the spaces between them—was understood as infrastructure. Maintenance work. But researchers at MIT have begun to challenge that hierarchy, proposing that star-shaped cells called astrocytes, which exist in numbers roughly equal to neurons themselves, may be far more than mere scaffolding. They may, in fact, be active participants in the very process of storing and retrieving memories.

The work, led by Leo Kozachkov in collaboration with Jean-Jacques Slotine and Dmitry Krotov of the MIT-IBM Watson AI Lab, emerged from a mathematical puzzle that has long nagged at memory researchers. The dominant models used to explain how brains store information—systems called Hopfield networks—work by distributing memories across the connections, or synapses, between neurons. They're elegant and they explain a lot. But they have a hard ceiling. Feed enough memories into a Hopfield network and the system begins to choke; new memories interfere with old ones, and the whole apparatus degrades. Scientists developed more sophisticated models, called dense associative memories, that could hold vastly more information. The catch: these advanced systems require something neurons alone cannot naturally provide. They need higher-order interactions—three or more cells communicating at once, not just the traditional two-neuron synapse.

This is where astrocytes enter the picture. Unlike neurons, which fire electrical signals, astrocytes communicate through calcium waves and release chemical messengers called gliotransmitters. Each astrocyte extends thousands of delicate processes that wrap around individual synapses, creating what neuroscientists call tripartite synapses—three-way junctions involving a sending neuron, a receiving neuron, and the astrocyte itself. The MIT team proposes that these three-part structures function not as passive observers but as computational units. By inserting themselves into the conversation between neurons, astrocytes could generate exactly the kind of higher-order interactions that dense associative memory models require.

If the model holds, the implications for memory capacity are striking. Traditional neuron-based models predict a storage ceiling—a point beyond which the brain cannot reliably add new memories. The neuron-astrocyte framework suggests no such limit exists. Instead, memory capacity appears to scale with the size of the network itself. The brain, in other words, may not be bumping up against some fundamental storage constraint. It may simply be far larger than we thought. The researchers are careful to note this does not mean infinite memory. Rather, it suggests the brain possesses an architecture capable of holding vastly more information than neuron-only theories would predict—which may help explain why humans have never identified a hard upper limit to what we can remember.

The hypothesis does not emerge from pure theory. Recent neuroscience has accumulated hints that astrocytes do more than housekeeping. Studies have shown that disrupting the connections between astrocytes and neurons in the hippocampus—a region critical to memory formation—impairs both the laying down and retrieval of memories. Advances in calcium imaging have allowed researchers to watch astrocytes coordinate their activity with neurons in real time. None of this proves the MIT model. But it does support the broader claim that astrocytes are involved in information processing, not merely maintaining the neural environment.

Yet the researchers themselves are clear about what remains unknown. Their work is a mathematical model. The proposed mechanism has not been directly observed in living brains. No experiment has yet shown astrocytes performing the memory-related computations the theory predicts. Krotov has expressed hope that the study will prompt experimental neuroscientists to test the hypothesis directly. Such work will be essential before the theory can reshape how neuroscience understands memory itself.

If future experiments bear out the model, the field may need to reconsider one of its most basic assumptions: that the synapse between two neurons is the fundamental unit of memory storage. Memory might instead depend on a more intricate system in which neurons and astrocytes work in concert. This would not erase what neuroscience already knows. But it could substantially enlarge the picture—and finally explain why the human brain, unlike any computer we have built, seems to have no known limit to what it can hold.