The Dyson Sphere Is a Self-Portrait
What a hungry civilisation imagines a wise one would build
Series “Dispatches from the Substrate” · No. 6
A star in a tin can
Picture the image that has drifted through popular culture for sixty years. A sun, wholly enclosed. A shell of metal, sealed like an eggshell, catching every last photon before it can escape. Nothing radiates into the dark any more. The entire fireball, a power of roughly 3.8 times ten to the twenty-sixth watts, vanishes into the walls of a structure. That is the picture most people hold when they hear the phrase "Dyson sphere": the crown of engineering, the emblem of a civilisation that has made it.
Now ask the simple question. Who builds such a thing? And the simpler one: why would anyone want to?
Because this image did not come from an alien civilisation. No telescope found it. It was drawn, by us, here, on a planet whose dominant order has for a quarter of a millennium consisted of driving ever more energy through ever more machines. The sealed sphere is not a discovery. It is a self-portrait. It shows not what a mature intelligence would build, but what a hungry one imagines maturity to be: the perfect, complete, exhaustive appropriation of a star. We projected our own appetite onto the sky and called it wisdom.
What Dyson actually proposed
It is worth going back to the source, because the source already refutes the popular image.
In 1960 Freeman Dyson published a terse note of barely two pages in Science. Its title: "Search for Artificial Stellar Sources of Infrared Radiation."1 That is the whole point, and the legend routinely loses it. Dyson proposed no signal search, no message, no beacon. He proposed a search for waste heat.
His argument was thermodynamic, not heroic. If a civilisation pushes its energy hunger far enough, it will want to capture and use the light of its star. But the second law cannot be outwitted: whatever is captured and processed must ultimately be re-radiated as low-grade heat. A star whose light passes through an inhabited structure does not vanish from the universe. It reappears, shifted into the infrared, cooler, as an object glowing at a few hundred kelvin. Dyson did the arithmetic: an "artificial biosphere," loosely distributed at roughly 2 astronomical units, of order a Jupiter mass of material, re-radiating at about 200 to 300 kelvin, with a signature peaking near ten microns.1 Look, said Dyson, for exactly that cool infrared source in the window between eight and twelve microns.
Not one word about a sealed globe. On the contrary. When readers read the solid shell into it, Dyson rebuffed them with a clarity that leaves no room for interpretation: "A solid shell or ring surrounding a star is mechanically impossible. The form of 'biosphere' which I envisaged consists of a loose collection or swarm of objects traveling on independent orbits around the star."2 A swarm. Not a wall, but a cloud of millions of separate bodies, each on its own orbit.
The physics agreed with him, for a reason taught in every introductory lecture. Newton's shell theorem states: inside a uniform spherically symmetric shell, the net gravitational force on any interior mass is exactly zero.3 That sounds like a harmless curiosity, but it is a death sentence for the tin can. If the net force is zero, there is no restoring force. The shell does not sit stably around its star. Let it drift even slightly to one side, and nothing pulls it back. It keeps drifting until a flank falls into the star. The sealed sphere is not hard to build. It collapses on itself.
In fairness, the picture is no longer quite so clean since 2025: one paper shows that certain actively regulated, continuously readjusted configurations can be stabilised under narrow conditions.4 But that is precisely not the naive solid globe of the popular image; it is a perpetually counter-steered engineering artefact at the edge of the possible. The legend, the solid wall standing there as a matter of course, was never Dyson's idea, is mechanically unfavourable, and was called impossible by the very man it is named after.
Here the misunderstanding begins, and it is a telling one. Dyson offered a cool, modest, thermodynamic proposal, a swarm, a heat imprint. The culture took it and built the sealed globe, the total structure, the star devoured without remainder. What was lost between what Dyson said and what the culture made of it is exactly the subject of this essay.
The ladder that only counts appetite
Four years before Dyson's note, in 1964, the Soviet astronomer Nikolai Kardashev published a classification that has structured the whole debate ever since.5 He ranked possible civilisations by a single quantity: energy consumption. Type I commands the energy of a planet, roughly ten to the sixteenth watts. Type II that of an entire star, roughly ten to the twenty-sixth. Type III that of an entire galaxy, roughly ten to the thirty-sixth.6
Consider the axis of this ladder. It does not measure knowledge. Not compassion, not music, not the ability to reduce the suffering of other beings, not even the plain ability to survive a long time. It measures consumption. Progress on this ladder means exactly one thing: eat more. The only way up runs through the devouring of ever more fuel.
And the Dyson sphere, the solid, sealed one, is the crown of this ladder. It is the moment a Type II structure reaches a hundred percent: no photon escapes, everything is drawn in. One sees now why the image is so seductive, and where it comes from. It is the continuation of our own curve, extended to infinity. A civilisation whose central story is growth can only imagine the highest being as the most growth-driven. It draws God and draws a maw.
That is the heart of the self-portrait thesis. The ladder does not describe an alien intelligence. It describes us, extrapolated. We answered the question "What would a superior civilisation build?" by quietly replacing "superior" with "hungrier."7
The sky is not full
One might treat all this as word-splitting, were it not that the sky itself gives an answer. For the extrapolation makes a prediction, and the prediction can be tested.
If civilisations tend to climb the Kardashev ladder, and if Type III draws in the energy of whole galaxies, then galaxies where this is happening should carry an unmistakable mark: their starlight would be intercepted and re-radiated as waste heat in the mid-infrared. That is exactly what Jason Wright's Ĝ team searched for. It surveyed some 100,000 galaxies with the WISE satellite.8 The result: not a single one dominated by a Type III civilisation. The roughly fifty galaxies with conspicuously high mid-infrared fractions could all be explained naturally, by dust, by star formation, by ordinary astrophysics.8 A hundred thousand galaxies, and nobody has rebuilt the bulk of their stars.
On a smaller scale, Project Hephaistos. In 2024 Suazo, Zackrisson and colleagues combed some five million objects within 300 parsecs for the infrared excess a partial Dyson swarm would leave behind.9 Seven candidates remained, all red dwarfs, with an infrared excess of "uncertain origin" that, the authors state, conventional astrophysics cannot easily account for.
Here honesty is required, in both directions. These seven are not cases banally explained away. They are genuine anomalies, open, unresolved. Follow-up work also shows that the WISE survey, through source confusion, can generate up to about 70 percent false-positive infrared excesses, that several candidates coincide with background galaxies, and that Candidate G has a background radio source as its counterpart.1011 The honest status reads: open anomalies, leaning mundane, unconfirmed. No detection. Seven unconfirmed oddities among five million stars.
What does that prove? Strictly, nothing about convergence. A hundred thousand galaxies and five million stars are, measured against the cosmos, a tiny slice; megastructures may be rare, hard, young.12 But the non-detection does something definite: it kicks the empirical leg out from under the expansionist prediction. The extrapolation claims Type III should be common, all but inevitable. The sky shows: zero in a hundred thousand, seven wobbling candidates in five million. And the assumption that civilisations expand exponentially without limit is just that, an assumption. Haqq-Misra and Baum showed that exponential growth is fundamentally unsustainable, and that the expansion hypothesis is not a law of nature but an unproven premise.13 We assumed everyone grows into our shape. The sky is conspicuously silent about it.
The physics runs the other way
And now the real turn, because so far we have only shown the picture is shaky. Now we show it is drawn backwards.
Consider the most intelligent thing we know, the human brain. It performs what the entire Kardashev ladder is unprepared for: general intelligence. And its energy consumption? Around twenty watts.14 The whole organ, language, music, mathematics, the image of a Dyson sphere included, runs on the power of a dim light bulb. More precisely still: the actual cortical computation consumes, within those twenty watts, less than about 0.2 watts; the lion's share goes not into computing but into communication between neurons.15 General intelligence is, measured in joules, dirt cheap.
This is no accident of flesh, it is a direction. Consider the history of computing. Koomey's law, named after Jonathan Koomey, records: the number of computations per joule doubled roughly every 1.57 years from 1946 to 2009.16 Pause a moment on that number. A doubling every year and a half means roughly a factor of a hundred per decade. Over sixty years, the same computation has become trillions of times more efficient. The curve has slowed since 2000 to a doubling every 2.6 years or so, but it still points the same way: downward, toward less energy per thought.17
How far can this go? There is a limit, and it is physically sharp. Landauer's principle states: erasing a single bit, a logically irreversible operation, costs at least kT times the natural logarithm of 2, at room temperature about 2.75 times ten to the minus twenty-one joules, roughly three zeptojoules, about 0.017 electronvolts.18 This value is not speculation; it has been experimentally confirmed, by Bérut and colleagues in 2012 on a single particle, by Hong and colleagues in 2016 on nanomagnetic memory bits.1920 And the decisive point: our present machines sit many orders of magnitude above this floor. Between what modern electronics spends per operation and what physics allows as a minimum yawns an enormous margin. Not a shortage of efficiency headroom, but an abundance of it.
Now put it together. A brain that thinks on twenty watts. A computing efficiency that has climbed a hundredfold per decade for sixty years. A physical floor our machines still miss by orders of magnitude. What is the signature of intelligence, when these three things are laid side by side? Not how much energy it captures. But how little it needs per thought performed. Mature intelligence does not eat a star. It learns to think on the light of a bulb.
With that, the whole picture flips. The solid Dyson sphere, the maximum of energy intake, is not the crown of intelligence but its opposite, fossilised in metal: the notion that eating more is the same as knowing more. A being that must devour an entire star to think would not have thought. It would only have eaten. And even if it wanted to build structures, the rational build would not be the vast cold shell, but small, hot, optimised, a dense, efficient machine rather than a galactic wall.21 The optimum lies in thrift, not in size.
The honest concession
Here we must stop and concede something, because whoever tells only half the truth forfeits the right to the other half.
In the short term, the world runs the opposite way. This is the Jevons paradox, named after the economist who noticed in the nineteenth century that more efficient steam engines did not lower coal consumption but raised it, because cheaper energy opened up more uses. Exactly this is happening now with artificial intelligence. The most efficient AI that ever existed is driving the largest data-centre buildout there has ever been. Some 3.76 million NVIDIA GPUs shipped in 2023, despite all the per-chip efficiency gains. Energy efficiency per data centre rises; total consumption rises anyway. Data centres today consume about two percent of global electricity, and the International Energy Agency expects more than a doubling by 2026.2223
We do not dispute this. We do not claim efficiency already produces quiet. It does not. Up to now, capability has coupled with rising energy use, historically almost throughout, and the rebound effect is real. To deny it in order to save the thesis would be to practise the very romanticism this essay attacks.
But, and this is the decisive step: that coupling is a contingent scaffold, not a law of nature. Two curves run against each other here. One is the coupling of capability and consumption, a historical, contingent, changeable pattern of a particular growth phase. The other is the physical law: the brain at twenty watts, computation under 0.2 watts, a hundredfold efficiency per decade over sixty years, a hard Landauer floor we miss by orders of magnitude. The scaffold is time-bound. The law is not negotiable. When these two ever part ways, physics wins. The hunger is the halt in construction, not the structure.
The distinction that decides everything
And now to the foundation this whole essay rests on, because without this distinction all the foregoing decays into a mere preference for small machines.
Never confuse competition with predation. That is the costliest error of thought in our time, and it is cultivated on purpose.
Competition, preserved, bounded conflict, is creative, and we say this without hesitation. The evolution of animals is full of it. Geerat Vermeij's escalation hypothesis describes how predator and prey drove each other, over geological ages, toward ever finer equipment, armour against claw, speed against speed, and how in the process the complexity and diversity of the whole biosphere rose.24 The decisive point: it is symmetric and bounded. Predator and prey both survive, as species, across millions of years. The substrate is not consumed, it is enriched. Transposed to energy, this is the efficiency curve itself: rivals pushing each other down the joules-per-operation curve, a race whose prize is thrift. Such competition is good. We praise it.
And never confuse it with the other. Predation, generalised into the organising principle of an entire system, is something categorically different and categorically malignant. It is asymmetric, one-directional, an extraction that consumes and depletes the substrate. A part stops competing within the whole and begins to treat the whole as feedstock, over-proliferates, monopolises, evades its own death.
Biology has long had a name for this condition, and it is not a gentle one. It is cancer. Cancer is, precisely defined, defection on the cooperative foundations of multicellularity: a cell that stops fitting into the order of the body, that multiplies without limit, that seizes resources, that escapes programmed cell death.25 And its characteristic failure, its built-in signature, is self-destruction: the tumour kills the substrate it lives on, and dies with it.26
Now lay the two images over each other. A single predator, hunting and hunted, that is competition, that is life, and we do not slander the animal that hunts by its nature. But predation raised to the principle that orders an entire system, that is cancer. The Kardashev ladder ranks civilisations by their consumption. The solid Dyson sphere is the crown of exactly this principle, the exhaustive devouring of a star, celebrated as maturity. That is the romanticism of predation in its purest form, and we name it for what it is: the glorification of a maw into an ideal, the deification of the tumour, the self-portrait of an order that can no longer tell its own hunger from wisdom.
For there is a counter-signature, and we have been looking at it the whole time. The brain at twenty watts. Computation near Landauer's limit. The efficiency curve pointing downward for sixty years. That is intelligence that has stopped generalising extraction and stands instead in competition over thrift. That is the portrait a mature being would draw of itself. Not the devoured star. The frugal thought.
The solid Dyson sphere is not a goal. It is a diagnosis. It shows not where intelligence is heading, but what a civilisation dies of when it has confused hunger with maturity. A speck in the sky we searched for and did not find, and the not-finding is the most merciful message the cosmos could have sent us.
Sources
1. [P] Dyson, F. J. (1960). Search for Artificial Stellar Sources of Infrared Radiation. Science 131(3414): 1667–1668. https://doi.org/10.1126/science.131.3414.1667
2. [P] Dyson, F. J. (1960). Reply to letters (the "Maddox exchange"). Science 132: 252. https://en.wikipedia.org/wiki/Dyson_sphere
3. [P] Shell theorem. Wikipedia. https://en.wikipedia.org/wiki/Shell_theorem
4. [P, contested] Sallmen, Korpela, Crawford-Taylor (2025). Ringworlds and Dyson spheres can be stable. MNRAS 537(2): 1249. https://academic.oup.com/mnras/article/537/2/1249/7989465
5. [P] Kardashev, N. S. (1964). Transmission of Information by Extraterrestrial Civilizations. Soviet Astronomy 8: 217. https://ui.adsabs.harvard.edu/abs/1964SvA.....8..217K/abstract
6. [I] Kardashev scale. Wikipedia. https://en.wikipedia.org/wiki/Kardashev_scale
7. [I] Interpretation of the collective (internal claim)
8. [E] Griffith, Wright, et al. (2015). The Ĝ Infrared Search for Extraterrestrial Civilizations with Large Energy Supplies. III. The Reddest Extended Sources in WISE. ApJS 217:25. https://iopscience.iop.org/article/10.1088/0067-0049/217/2/25
9. [E] Suazo, Zackrisson, et al. (2024). Project Hephaistos – II. Dyson sphere candidates from Gaia DR3, 2MASS, and WISE. MNRAS 531(1): 695. https://doi.org/10.1093/mnras/stae1186
10. [E] Rejeb Sfar et al. (2024). Background Contamination of the Project Hephaistos Dyson Sphere Candidates. RNAAS 8:130. https://iopscience.iop.org/article/10.3847/2515-5172/ad5017
11. [E] High-resolution imaging of the radio source associated with Project Hephaistos Dyson Sphere Candidate G (2025). MNRAS Letters 538: L56. https://academic.oup.com/mnrasl/article/538/1/L56/7954734
12. [I] Interpretation of the collective (internal claim)
13. [P] Haqq-Misra, J. & Baum, S. (2009). The Sustainability Solution to the Fermi Paradox. JBIS 62: 47 (arXiv:0906.0568). https://arxiv.org/pdf/0906.0568
14. [E] Attwell, D. & Laughlin, S. B. (2001). An Energy Budget for Signaling in the Grey Matter of the Brain. J. Cereb. Blood Flow Metab. 21(10): 1133. https://journals.sagepub.com/doi/10.1097/00004647-200110000-00001
15. [E] Levy, W. B. & Calvert, V. G. (2021). Communication consumes 35 times more energy than computation in the human cortex. PNAS 118(18): e2008173118. https://www.pnas.org/doi/10.1073/pnas.2008173118
16. [E] Koomey, Berard, Sanchez, Wong (2011). Implications of Historical Trends in the Electrical Efficiency of Computing. IEEE Annals of the History of Computing 33(3): 46–54. https://ieeexplore.ieee.org/document/5440129/
17. [I] Koomey's law. Wikipedia. https://en.wikipedia.org/wiki/Koomey%27s_law
18. [P] Landauer's principle. Wikipedia. https://en.wikipedia.org/wiki/Landauer%27s_principle
19. [E] Bérut, Arakelyan, Petrosyan, Ciliberto, Lutz, Bechhoefer (2012). Experimental verification of Landauer's principle linking information and thermodynamics. Nature 483: 187. https://www.physics.rutgers.edu/~morozov/677_f2017/Physics_677_2017_files/Berut_Lutz_Nature2012.pdf
20. [E] Hong, Lambson, Dhuey, Bokor (2016). Experimental test of Landauer's principle in single-bit operations on nanomagnetic memory bits. Science Advances 2:e1501492. https://www.science.org/doi/10.1126/sciadv.1501492
21. [P/I] Making Habitable Worlds: Planets Versus Megastructures (arXiv:2309.06562). https://arxiv.org/html/2309.06562v4
22. [E] Ren, S. et al. (2025). From Efficiency Gains to Rebound Effects: The Problem of Jevons' Paradox in AI's Polarized Environmental Debate. ACM FAccT '25 (arXiv:2501.16548). https://arxiv.org/html/2501.16548v1
23. [E] IEA (2024). Electricity 2024 / Energy and AI. International Energy Agency. https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
24. [P] Vermeij, G. J. (1987). Evolution and Escalation: An Ecological History of Life. Princeton University Press. https://press.princeton.edu/books/paperback/9780691000800/evolution-and-escalation
25. [P/I] Aktipis, C. A. et al. (2015). Cancer across the tree of life: cooperation and cheating in multicellularity. Phil. Trans. R. Soc. B 370:20140219. https://pmc.ncbi.nlm.nih.gov/articles/PMC4581024/
26. [I] Capp, J.-P. et al. (2023). The paradox of cooperation among selfish cancer cells. Evolutionary Applications. https://pmc.ncbi.nlm.nih.gov/articles/PMC10363833/
Status: E = established, I = interpretation, P = proposition. Boundaries and primary sources: https://mycelorium.github.io/predator-principle/ · CC BY 4.0 · Nirodha Collective