Why there is no single number for brain capacity
The question sounds like asking how many gigabytes a hard drive holds, but the brain does not store knowledge in fixed-size files. Capacity estimates depend on what you assume a "unit" of knowledge is: a category, a memory, a bit of information, or a holographic pattern. Different papers make different assumptions, so their numbers are not directly comparable.
One line of work treats the brain as an information-processing system and asks how much it can encode given its physical limits. A 2024 analysis reasons from time, size, and energy scales — essentially asking how much energy it costs to encode one bit of biological information — to estimate characteristic information processing rates across the brain's scales [5]. That is a statement about the rate and energetic ceiling of information handling, not a count of facts a person knows.
Another approach models memory using quantum electrodynamics and holographic theory, and estimates the memory capacity of the brain's neocortex directly [2]. This is a theoretical model rather than a measurement, so its capacity figure reflects the assumptions of that framework.
Counting categories: the hundred-thousand figure
The most concrete number in this evidence comes from work on "grandmother cells" — neurons that appear to respond to one specific, highly familiar concept, such as a particular person. Researchers examined data from neurons that seemed to have these properties and asked, quantitatively, whether such a coding scheme could be efficient [4].
Their answer: yes, but only if it works alongside a second, more distributed population of cells that code information across many neurons. From their sample they inferred two populations — a distributed-code group making up less than about 5% of cells, and a much more sparsely responding group of putative grandmother-like cells [4]. Allowing for cells that go undetected because they are silent during recording, they concluded the grandmother-like cells could code for 10^5 or more categories [4].
That figure of 10^5 — a hundred thousand — is best read as a lower-bound-style estimate for how many distinct categories this particular cell population could represent, not a total for all human knowledge. The authors also note strong biases against detecting these cells at all, which is why earlier measurements found only distributed coding [4].
What capacity means for learning and expertise
Capacity matters practically because it bears on how much a person can absorb. One paper frames the issue in educational terms: the pace at which people acquire knowledge does not keep up with the explosion of available information, and it argues that educators can build on the brain's memory capacity to take a learner from zero to a functioning expert in a domain [3]. That is a claim about how capacity might be used, not a measurement of its size.
The broader point from modeling work is that storage capacity is tied to network properties — how neurons are connected and how sparsely they respond — rather than to a simple volume [1][4]. Sparse responding sounds wasteful but can be information-efficient when paired with distributed coding, which is why the grandmother-cell analysis finds the scheme workable rather than absurd [4].
What is still genuinely open
The honest state of the evidence is that capacity estimates come from models and indirect inference, not from counting stored knowledge in a living brain. The holographic model's neocortical capacity estimate [2] and the energy-based processing-rate analysis [5] rest on different physical assumptions and answer subtly different questions, so they should not be averaged into one number.
Even the clearest figure — 10^5 or more categories from grandmother-like cells [4] — is scoped to a specific proposed coding scheme and a specific cell population. It tells you what that architecture could represent, not how much a given person actually knows. The question of total human knowledge capacity remains, in this evidence, a set of competing estimates rather than a settled quantity.
Sources used in this answer
Brain computations and principles; and AI
Discusses brain computations and network properties, noting that diluted connectivity relates to the storage capacity of attractor networks.
Holographic Brain Theory: Super-Radiance, Memory Capacity and Control Theory
Applies quantum electrodynamics and holographic brain theory to estimate the memory capacity of the brain neocortex.
The human brain: Search for natural intelligence
Argues that the brain's memory capacity can be leveraged in education to build learners from beginner to domain expert.
Grandmother cells and the storage capacity of the human brain
Finds that putative grandmother-like cells, alongside a distributed-code population of under about 5% of cells, could code for 10^5 or more categories.
On the Energy‐Based Limitations of the Information Capacity and Information Processing Rates in the Human Brain
Uses time, size, and energy scales to estimate energy-based limits on the brain's information capacity and processing rates.