A thought does have weight, in two senses physics allows. By Landauer's principle and E = mc², one bit of information at body temperature corresponds to at least 3.3 × 10⁻³⁸ kg, so everything stored in your neocortex's synapses comes to roughly 2 × 10⁻²³ kg, about the mass of one small protein molecule. The energy your brain actually spends on thinking is far larger: about 17 watts, which burns roughly 3.7 grams of glucose an hour.
- 3.3 × 10⁻³⁸ kg is the mass equivalent of one bit at body heat, by Landauer's principle and E = mc².
- About 7 × 10¹⁴ bits sit in your neocortex's synapses (1.5 × 10¹⁴ synapses × 4.7 bits).
- 2.3 × 10⁻²³ kg, roughly one small protein molecule, is the Landauer mass of everything you know.
- 17 W of glucose power runs the whole brain; only about 0.1 W pays for cortical computation.
- 3.7 g of glucose an hour is the weight of thinking you can actually put on a scale.
The question sounds like poetry, but it has a clean physical version. Information is not abstract: it is stored as physical states, and changing states costs energy. Energy has mass.
So we can ask how much mass corresponds to the information in a head, and separately how much mass a thinking brain actually burns. The two answers differ by about 20 orders of magnitude, and the gap is the interesting part.
What is Landauer's principle, in plain words?
Landauer's principle says that erasing one bit of information must release at least \( k_B T \ln 2 \) of heat into the surroundings, where \( k_B \) is Boltzmann's constant and \( T \) is the temperature. The physicist Rolf Landauer's argument is simple: a memory that can hold 0 or 1 has two possible states, and resetting it to a known state halves the number of possibilities. The second law of thermodynamics says that lost entropy must go somewhere, and it goes out as heat.
It is not just theory. Antoine Bérut, Artyom Petrosyan and Sergio Ciliberto built a one-bit memory from a single glass bead in a laser trap with two wells, and erased it repeatedly. When they erased slowly, the average heat released settled at the Landauer bound, as their 2015 paper below reports.
The first person to try weighing thought with a real scale was the Turin physiologist Angelo Mosso. Around 1882 he built a "human circulation balance": a padded wooden table resting on a steel knife-edge, with a heavy counterweight to damp the rocking of each breath. A subject lay on it for at least an hour until it settled. In his 1884 report the subjects were V.G., a 22-year-old medical student, and Giorgio M., his laboratory technician; a 20 g weight placed by the knees was enough to tip it. When Mosso tapped a telegraph key, the head end sank. His daughter later wrote that it sank faster when the subject read a mathematics or philosophy manual than a newspaper or novel, and that when the student read a letter from an upset creditor the balance fell all at once. Mosso was weighing blood moving towards the head, not thought itself, and modern authors say there is no direct proof it tracked brain blood flow.
Sandrone S. et al.: Weighing brain activity with the balance, Brain (2014)A single colloidal particle in a modulated double-well potential served as a one-bit memory. Over long erasure cycles the mean heat dissipated saturated at the Landauer bound, and a detailed Jarzynski equality recovered the same bound independently. Limit: a slow, single-bit laboratory system; real computers and brains dissipate many orders of magnitude more per bit.
Read the paper (free)How much does one bit weigh?
One bit at body temperature corresponds to about 3.3 × 10⁻³⁸ kg. Here is the derivation, step by step.
Assumptions. (1) The "weight" of a bit is the mass equivalent of the minimum energy tied to it, the Landauer bound. (2) The memory is at body temperature, \( T = 310\ \text{K} \) (37 °C). (3) Mass and energy are related by Einstein's \( E = mc^2 \). Constants: \( k_B = 1.380649 \times 10^{-23}\ \text{J/K} \) (exact in the SI since 2019) and \( c = 2.998 \times 10^8\ \text{m/s} \).
- Landauer energy per bit. \[ E_{bit} = k_B T \ln 2 = (1.380649\times10^{-23})(310)(0.6931) = 2.97\times10^{-21}\ \text{J} \]
- Mass equivalent. \[ m_{bit} = \frac{E_{bit}}{c^2} = \frac{2.97\times10^{-21}}{(2.998\times10^{8})^2} = 3.30\times10^{-38}\ \text{kg} \]
- Check against the literature. At 300 K the same formula gives 3.19 × 10⁻³⁸ kg, the value Melvin Vopson published in AIP Advances in 2019.
For scale, an electron has a mass of about 9.1 × 10⁻³¹ kg (NIST CODATA), so one bit at body heat weighs roughly 28 million times less than an electron.
Calculated in this post from the electron mass above and the 3.3 × 10⁻³⁸ kg per bit derived in the steps above (9.1 × 10⁻³¹ ÷ 3.3 × 10⁻³⁸ ≈ 2.8 × 10⁷).
How much information is stored in a brain?
A defensible estimate is around 7 × 10¹⁴ bits in the neocortex's synapses, though the true number is uncertain by at least an order of magnitude. Two measurements make it possible.
- Number of synapses. Bente Pakkenberg and colleagues used stereology on human brains and reported about 0.15 quadrillion, 1.5 × 10¹⁴, synapses in the neocortex (Experimental Gerontology, 2003).
- Bits per synapse. Thomas Bartol, Terrence Sejnowski and colleagues reconstructed rat hippocampal tissue under the electron microscope and compared pairs of synapses that share the same input and output. Their sizes matched so precisely that the authors estimated at least 26 distinguishable synaptic strengths, about 4.7 bits per synapse (eLife, 2015).
Serial-section electron microscopy of rat hippocampus found pairs of synapses from the same axon onto the same dendrite with nearly identical spine-head sizes. From that precision the authors estimated at least 26 distinguishable synaptic strengths, or 4.7 bits per synapse, which requires averaging activity over minutes. Limit: one brain region in rats; applying it to every human cortical synapse is an extrapolation.
Read the paper (free)
So what does everything you know weigh?
About 2.3 × 10⁻²³ kg, or 23 zeptograms. Multiply the bit count by the energy and mass per bit:
- Total Landauer energy. \[ E = N_{bits}\,k_B T \ln 2 = 7.05\times10^{14} \times 2.97\times10^{-21} = 2.09\times10^{-6}\ \text{J} \]
- Total mass equivalent. \[ m = \frac{E}{c^2} = \frac{2.09\times10^{-6}}{8.99\times10^{16}} = 2.33\times10^{-23}\ \text{kg} \]
- In familiar units. One dalton is 1.66 × 10⁻²⁷ kg (NIST CODATA), so this is about 14,000 daltons: the mass of a single small protein molecule.
Calculated in this post from the 2.33 × 10⁻²³ kg total above and the dalton value from NIST CODATA: 2.33 × 10⁻²³ ÷ 1.66 × 10⁻²⁷ ≈ 14,000.
Two micro-joules is roughly what a 17 W brain spends in about a ten-millionth of a second. Your entire stored knowledge, valued at the thermodynamic minimum, is lighter than one molecule of the enzymes that maintain it.
The Landauer mass of a memory
Could any scale ever weigh a thought?
A scale could resolve that mass in isolation, but never inside a head. In 2012 Julien Chaste, Adrian Bachtold and colleagues in Barcelona built a carbon-nanotube resonator about 150 nm long, vibrating near 2 GHz, that detected mass changes of 1.7 yoctograms (1.7 × 10⁻²⁷ kg), roughly one proton.
The Landauer mass of all your knowledge, 2.3 × 10⁻²³ kg, is about 14,000 times that resolution. The problem is the background: the instrument works only when the thing being weighed lands on a nanotube in a vacuum, and a living brain swaps far more mass than that with its blood every microsecond.
Calculated in this post from the 2.3 × 10⁻²³ kg Landauer mass above and the 1.7 yoctogram resolution reported by Chaste et al. (2012) (2.3 × 10⁻²³ ÷ 1.7 × 10⁻²⁷ ≈ 14,000).
Vopson proposed a cleaner test: weigh a data drive empty and full. For a 1 TB drive his formula predicts about 2.5 × 10⁻²⁵ kg, which he notes is far beyond today's balances, including the ultra-precise Kibble balance.
Proposes, as an explicitly speculative extension of Landauer's principle, that a stored bit has a rest mass of kBT ln 2 / c², about 3.19 × 10⁻³⁸ kg at 300 K, and suggests weighing a storage device empty and full; for 1 TB the predicted change is about 2.5 × 10⁻²⁵ kg. Limit: no experiment has confirmed it, and mainstream thermodynamics ties the Landauer energy to erasure, not to storage.
Read the paper (free)Where does the analogy break down?
At the word "stored". Landauer's bound is about the heat released when information is erased, and that heat leaves the system. A memory full of 1s does not have to contain more energy than a memory full of 0s; that depends entirely on how the two states are built.
So the mainstream reading is that information, as such, has no rest mass; only energy differences between physical states do. Vopson's principle is the minority view that says otherwise, and it remains untested.
| Sensitivity check | Change | Mass of "all you know" |
|---|---|---|
| Baseline | 1.5 × 10¹⁴ synapses, 4.7 bits, 310 K | 2.3 × 10⁻²³ kg |
| One bit per synapse | Bits ÷ 4.7 | 5.0 × 10⁻²⁴ kg |
| 10× more synapses (my estimate if you count beyond neocortex) | Synapses × 10 | 2.3 × 10⁻²² kg |
| Room temperature instead of body | 310 K → 300 K | 2.25 × 10⁻²³ kg (−3%) |
| Real neurons, not ideal erasure | ~10⁸ times the Landauer cost (Levy & Calvert) | ~2 × 10⁻¹⁵ kg |
Every reasonable input moves the answer by a factor of up to about ten. None of them moves it anywhere near something you could feel.
What does thinking actually cost in real mass?
A lot more than information theory demands: about 3.7 grams of glucose an hour, which leaves your body as carbon dioxide and water. William Levy and Victoria Calvert, in PNAS in 2021, revised the brain's energy budget to about 17 W of glucose power, of which roughly 6.2 W ends up as usable ATP.
Only about 0.1 W pays for cortical computation itself; around 3.5 W pays for communication, sending spikes down axons. They also compare neurons with Landauer's limit of about 3.4 × 10²⁰ bits per joule at body temperature and find a simple neuron model about 10⁸ times less efficient.
- Energy per hour. \( E = 17\ \text{W} \times 3600\ \text{s} = 61{,}200\ \text{J} = 14.6\ \text{kcal} \)
Calculated in this post from the brain's 17 W of glucose power reported by Levy and Calvert (2021), over one hour (3,600 s), converted at 4,184 J per kcal.
- Glucose burnt. With the Atwater factor of 4 kcal per gram of carbohydrate: \( 14.6 / 4 \approx 3.7\ \text{g/h} \), or about 88 g a day.
- Mass-energy equivalent of that hour. \[ m = \frac{61{,}200\ \text{J}}{(2.998\times10^8)^2} = 6.8\times10^{-13}\ \text{kg} \approx 0.7\ \text{ng} \]
- The "pure thought" part. At 0.1 W for computation, one hour is 360 J, a mass equivalent of 4 × 10⁻¹⁵ kg, still about 170 million times the Landauer mass of everything you know.
Calculated in this post from the 0.1 W for cortical computation reported by Levy and Calvert (2021), over one hour, and the 2.3 × 10⁻²³ kg Landauer mass above (0.1 × 3,600 = 360 J; 360 ÷ c² ≈ 4 × 10⁻¹⁵ kg; ÷ 2.3 × 10⁻²³ ≈ 1.7 × 10⁸).
A bottom-up energy budget for the human cortex: the whole brain runs on about 17 W of glucose, about 6.2 W as ATP; cortical computation costs about 0.1 W and communication about 3.5 W. The authors compare neural efficiency with the Landauer limit (about 3.4 × 10²⁰ bits per joule) and find neurons about 10⁸ times less efficient. Limit: many parameters are estimates, and the split between "computation" and "communication" depends on their definitions.
Read the paper (free)See the data as a table
| log10 of mass in kg (log10 kg) | |
|---|---|
| One bit at 37 °C (Landauer) | -37.5 |
| Best mass sensor resolution (1.7 yg) | -26.8 |
| All neocortical knowledge (Landauer) | -22.6 |
| Mass equivalent of 1 hour of cortical computation (0.1 W) | -14.4 |
| Mass equivalent of 1 hour of brain power (17 W) | -12.2 |
| Glucose burnt by the brain in 1 hour | -2.4 |
Source: Author calculations from Landauer bound, Pakkenberg 2003, Bartol 2015, Levy & Calvert 2021, Chaste 2012 · Each step left is ten times lighter. Bars show log10 of mass in kg, so a value of -22.6 means 2.3 × 10⁻²³ kg.

Does that mean thinking makes you lighter?
Yes, in the plain biological sense, though not because of the thoughts. The 3.7 g of glucose your brain burns each hour is converted to carbon dioxide and water that you breathe out and excrete, so a working brain makes your body lose mass continuously, as every organ does.
Thinking hard does not change that much: the brain's power budget is dominated by keeping neurons ready and sending signals, and Levy and Calvert's 0.1 W for computation is a small slice of the total.
The "weight of a heavy thought" is a metaphor; the weight of a thinking brain is about a teaspoon of sugar an hour.
For the brain-as-battery version of this sum, see the follow-up below; for what the brain does with all that energy at rest, read the default mode network, and for why it can only juggle a few items at once, the working memory limit. The reward chemistry that makes some thoughts feel heavier than others is in dopamine and prediction error.
Information itself has no measured rest mass, and the idea that it does (Vopson's principle) is unconfirmed. But if "the weight of a thought" means the mass equivalent of the minimum energy needed to write it, physics gives a firm number: 3.3 × 10⁻³⁸ kg per bit at body heat, and about 2.3 × 10⁻²³ kg for everything in your neocortex. That is within reach of the best nanotube balance in principle, and hopelessly buried inside a living brain. The real, weighable cost of thinking is the glucose: about 3.7 g an hour.
What's new
What if next?
- What if you could stop time for everyone but yourself? Frozen air becomes a wall and frozen light becomes darkness; the maths is brutal.
- What if your brain ran on a phone battery? At 17 W, a flagship phone battery lasts you about an hour of being yourself.
- What if you took only one breath a minute? The oxygen arithmetic of a very slow breath.
Questions people ask
Does information have mass?
Not in mainstream physics. Information is carried by physical states, and those states have energy and therefore mass, but the information itself adds mass only if the states holding it differ in energy. Vopson's mass-energy-information equivalence principle proposes otherwise and has not been tested.
Is a full hard drive heavier than an empty one?
Only if its written state stores more energy than its blank state, which depends on the technology. Even on Vopson's hypothesis, a full 1 TB drive would be about 2.5 × 10⁻²⁵ kg heavier, far below what any balance can measure.
How much energy does a thought use?
The whole brain uses about 17 W of glucose power. Levy and Calvert estimate that cortical computation itself is about 0.1 W, with most of the rest spent on communication between neurons and housekeeping.
What is the Landauer limit?
The minimum heat released when one bit is erased: kBT ln 2, about 2.9 × 10⁻²¹ J at room temperature. It has been confirmed experimentally with a single trapped particle and in a quantum many-body system.
Sources
- Bérut A., Petrosyan A., Ciliberto S.: Experimental verification of Landauer's erasure principle, J. Stat. Mech. (2015) (source for the experimental confirmation)
- NIST CODATA: Boltzmann constant, 1.380649 × 10⁻²³ J/K, exact (accessed 2026)
- Vopson M.M.: The mass-energy-information equivalence principle, AIP Advances (2019) (source for 3.19 × 10⁻³⁸ kg at 300 K and the 1 TB prediction)
- Bartol T.M. et al.: Nanoconnectomic upper bound on the variability of synaptic plasticity, eLife (2015) (source for 26 strengths, 4.7 bits per synapse)
- Pakkenberg B. et al.: Aging and the human neocortex, Experimental Gerontology (2003) (source for 0.15 × 10¹⁵ neocortical synapses)
- Chaste J. et al.: A mechanical mass sensor with yoctogram resolution, Nature Nanotechnology (2012) (source for the 1.7 yg resolution)
- Levy W.B., Calvert V.G.: Communication consumes 35 times more energy than computation in the human cortex, PNAS (2021) (source for 17 W, 0.1 W, 3.5 W and the 10⁸ efficiency gap)
- Sandrone S. et al.: Weighing brain activity with the balance: Angelo Mosso's original manuscripts come to light, Brain 137(2):621-633 (2014) (source for the balance, its 1882 and 1884 reports, the subjects and the reading experiments; author post-print read via this copy)
- FAO: Food energy, methods of analysis and conversion factors, Atwater factors (2003) (source for 4 kcal per gram of carbohydrate)
- TU Wien: The quantum physics of forgetting (Aimet et al., Nature Physics) (2025)
- University of Portsmouth: Is our universe the ultimate computer? (2025)
More on the science of the mind on the Manthan world page.
