Beyond Energy: Rethinking the Kardashev Scale in the Age of AI

Genesis Perspectives · Opinion · 21 October 2026 · Dr Luke Soon

Abstract. In 1964 Nikolai Kardashev proposed classifying technological civilisations by the power they command: planetary (Type I), stellar (Type II) and galactic (Type III). Six decades later the scale is cited well beyond astrophysics. A rocket company now uses it in regulatory filings, and AI executives use it to justify building data centres in orbit. This essay updates the calculation with 2026 data, places humanity at approximately K = 0.73, and shows that even the most ambitious AI energy programmes barely change that figure. It reviews the results of technosignature surveys, which have so far found no confirmed evidence of Type II or Type III civilisations, and sets out three critiques of an energy-only metric. It argues that, for a civilisation developing artificial intelligence, energy should be assessed alongside two further measures, information and control. The defining question is not only how much power a civilisation commands, but whether it retains control of the systems that power enables.

Keywords: Kardashev scale; technosignatures; Dyson spheres; artificial intelligence; energy; Great Filter; AI governance; Barrow scale

---

I. Introduction

On 30 January 2026 SpaceX asked the US Federal Communications Commission for permission to launch up to a million satellites that would operate as orbital data centres. The filing described the constellation as a first step towards becoming a Kardashev Type II civilisation, one that harnesses the full power of its star (DCD). In June, presenting the company’s first AI satellite design, Elon Musk argued that civilisational progress is best measured by the power a civilisation harnesses, and that humanity uses far less than a trillionth of the Sun’s output (Dealroom). Google has launched a prototype satellite for Project Suncatcher, a research programme to run AI chips on solar power in orbit (GPB/NPR).

The Kardashev scale is now routinely used to argue that greater energy use means a more advanced civilisation. An earlier version of this essay, published here in 2025, took that view largely at face value. This revised version is more sceptical. Its argument is that energy capacity is a necessary measure of advancement but not a sufficient one, and that a 1964 metric designed to estimate the detectability of extraterrestrial radio transmissions needs to be supplemented for a civilisation now building artificial minds.

II. What Kardashev actually proposed

Kardashev’s paper in Soviet Astronomy was not a theory of progress. It addressed a practical question in radio astronomy: if a civilisation exists, how much power could it devote to a signal we might detect? He defined three types. Type I commanded roughly the energy then used on Earth, about 4 × 10¹² watts. Type II commanded the output of its star, about 4 × 10²⁶ W. Type III commanded the output of its galaxy, about 4 × 10³⁷ W [1].

Two later contributions turned a detection aid into a general measure of development. In 1960 Freeman Dyson had already argued that a civilisation with very large energy needs would surround its star with collectors and would be detectable through waste heat in the infrared [2]. In 1973 Carl Sagan made the scale continuous, defining K = (log₁₀P − 6) / 10, with P in watts, and set Type I at 10¹⁶ W, closer to the solar power a planet intercepts [3]. Sagan’s version is the one most widely quoted today. Because it is logarithmic, each increase of 0.1 in K corresponds to a tenfold increase in power.

Type	Kardashev (1964)	Sagan (1973)	Physical reference
I, planetary	~4 × 1012 W	1016 W	Sunlight reaching Earth: ~1.7 × 1017 W
II, stellar	~4 × 1026 W	1026 W	Solar luminosity: ~3.8 × 1026 W
III, galactic	~4 × 1037 W	1036 W	Milky Way: ~1036 to 1037 W

III. Humanity’s position in 2026

The Energy Institute’s 2026 Statistical Review reports that global primary energy supply grew 1.7 per cent in 2025 and exceeded 600 exajoules for the first time (Rice University/Energy Institute) [5]. Six hundred exajoules a year is an average power of about 1.9 × 10¹³ W, or 19 terawatts. Sagan’s formula gives K ≈ 0.728. The figure has changed very little in recent years: Jiang and colleagues, using 2018 data, also calculated 0.728 [4]. Fossil fuels still supplied close to 86 per cent of primary energy in 2025 (Connaissance des Énergies).

Reaching Type I would therefore require roughly a 500-fold increase in power. Jiang et al., modelling fossil, nuclear and renewable supply under climate constraints, estimate that humanity would reach Type I between 2333 and 2404 [4]. That corrects the claim, made in the earlier version of this essay, that Type I could be reached within one or two centuries.

Scenario	Power	Effect on K
Global primary energy, 2025 (600 EJ)	~19 TW	K ≈ 0.728
All data centres, 2025 (IEA, 485 TWh)	~55 GW, about 0.3% of total	< 0.001
All data centres, 2030 (IEA base case, ~950 TWh)	~108 GW	< 0.001
Hypothetical: 10 TW of orbital AI compute added	~29 TW total	K ≈ 0.746
Doubling all human energy use	~38 TW	+0.03
Type I threshold (Sagan)	10,000 TW	K = 1.0

IV. AI and energy demand

Artificial intelligence is the first technology since electrification to make energy growth a central strategic priority again. The International Energy Agency estimates that data centres used around 415 TWh in 2024, roughly 1.5 per cent of global electricity, and projects that figure to more than double to about 945 TWh by 2030 [6]. Its April 2026 update found that data centre demand rose 17 per cent in 2025, that AI-focused facilities grew faster still, and that the five largest technology companies spent more than $400 billion on capital expenditure, with a further 75 per cent increase planned for 2026 [7]. The Energy Institute, tracking data centres for the first time in the Statistical Review’s 75-year history, reports a higher figure of 788 TWh for 2025 on a broader definition (Crypto Briefing/Energy Institute).

Table 2 shows why this changes the Kardashev figure far less than the public statements suggest. Even if orbital AI systems of the kind SpaceX and Google propose delivered 10 terawatts of compute, more than half of today’s total human energy use, K would rise from about 0.73 to about 0.75. SpaceX’s stated target of 100 gigawatts of orbital AI compute a year, rising eventually to a terawatt (SmallSat News), would be a major industrial achievement, but its effect on K would be negligible. The economics also remain contested: analysts question whether orbital data centres offer advantages over terrestrial ones once launch costs, debris risk and heat rejection are included (CNBC).

The first conclusion is therefore straightforward. AI energy demand will not materially change humanity’s Kardashev rating this century. What AI may change is which measures of advancement matter.

V. Evidence from technosignature searches

If civilisations commonly expanded their energy use as Kardashev assumed, astronomers should be able to detect them. Dyson’s argument was that large-scale energy use must produce waste heat. The Glimpsing Heat from Alien Technologies survey (G-HAT) used NASA’s WISE infrared telescope to examine about 100,000 galaxies. None showed signs of a civilisation reprocessing 85 per cent or more of its galaxy’s starlight, and only about fifty showed mid-infrared emission consistent with more than half (Centauri Dreams) [8]. The survey’s founder, Jason Wright, concluded that none of the galaxies WISE could resolve is widely populated by a civilisation using most of its starlight (Universe Today).

Within our own galaxy, Project Hephaistos in 2024 analysed roughly five million stars from Gaia, 2MASS and WISE and reported seven red dwarfs with unexplained infrared excess as partial Dyson sphere candidates [9]. Other astronomers soon showed that background dust-obscured galaxies along the line of sight could account for the excess, and concluded that several candidates had very likely been misidentified [10]. A 2026 archival study is using Gaia astrometry to test the remaining candidates for the same contamination [11]. Sixty-six years after Dyson’s paper, there is still no confirmed observational evidence of a Type II or Type III civilisation.

Absence of evidence is not evidence of absence, but it does constrain the possibilities. Such civilisations may be rare, they may not expand their energy use in the way Kardashev assumed, or something may prevent them from reaching that scale. Each possibility is relevant to our own development.

VI. Three critiques of an energy-only metric

Efficiency. Computation has a thermodynamic minimum. Landauer showed in 1961 that erasing one bit of information must dissipate at least kT ln 2 of energy [12], and Seth Lloyd later calculated the physical limits of computation [13]. Current chips operate many orders of magnitude above that minimum. A civilisation that prioritised computational efficiency could become far more capable without a proportional increase in energy use, and would be correspondingly harder to detect. Milan Ćirković, reviewing the scale fifty years on, describes such cases as scenarios in which an advanced civilisation’s waste heat is indistinguishable from the natural background [14]. The counter-argument is the Jevons paradox: in practice, more efficient energy use has usually increased total consumption [15]. The AI build-out so far fits that pattern.

Miniaturisation and information. The cosmologist John Barrow proposed measuring civilisations by the smallest scale at which they can manipulate matter: bodies, then genes, molecules, atoms and below [16]. On that measure, humanity’s progress since 1964 has been rapid, from the transistor to gene editing to nanometre-scale chips, even though per-capita energy use in high-income countries has changed little. Clément Vidal proposed combining the two into a two-dimensional metric, with energy on one dimension and miniaturisation on the other [17]. Sagan himself proposed a parallel information scale, lettered A to Z, with each letter representing a tenfold increase in the information a civilisation stores [3].

Survival. Robin Hanson’s Great Filter argument holds that at least one step between non-living matter and a galaxy-wide civilisation must be extremely improbable, and asks whether that step is behind us or ahead of us [18]. John Smart’s transcension hypothesis proposes that advanced intelligence may develop towards ever denser computation rather than expanding through space [19]. In both cases, the lack of detections suggests that the principal difficulty in becoming an advanced civilisation may not be acquiring power, but surviving the consequences of having it.

Kardashev asked how much power a civilisation commands. The question for this century is whether a civilisation still commands its power.

VII. A proposal: three measures instead of one

For a civilisation developing artificial intelligence, I propose that the Kardashev value be reported as one of three measures.

1. Energy (K). Kardashev and Sagan’s measure, unchanged: the power a civilisation commands. Humanity: approximately 0.73.
2. Information (I). How much a civilisation knows and computes per unit of energy, following Sagan’s information scale and Barrow’s measure of miniaturisation. This is where AI is producing the fastest change, and where the Kardashev scale says nothing.
3. Control (C). The proportion of a civilisation’s most capable systems that remain reliably under the direction of those who built them: verifiable, interruptible and accountable. If the Great Filter lies ahead of us, this is the measure on which it will be decided.

The control measure is already the working agenda of AI safety institutes and, as argued in Both Empires Want to Win the AI Race. Neither Wants AI to Win It., it is the one issue on which Washington and Beijing are converging. It can be assessed, at least in part, with the same rigour astronomers apply to infrared excess: incident reporting, independent evaluations, published capability thresholds and a demonstrated ability to halt a deployment. A practical principle follows. Each increase in energy and information capability should be matched by an increase in control. A civilisation whose energy and information capabilities grow while its control declines is not becoming more advanced. It is increasing its exposure to catastrophic risk.

VIII. Conclusion

The Kardashev scale remains an elegant idea and a useful tool for astronomers deciding where to search. It is a poor guide to policy. Read literally, it tells AI developers that greater civilisational advancement simply means more power, and it lends scientific legitimacy to every gigawatt campus and satellite constellation. The data support a more modest conclusion: humanity is at about K = 0.73, AI energy use will not change that figure materially this century, and no survey has yet found evidence of a civilisation operating at Type II or Type III scale.

This is where the Fork, a framework I return to often, applies: the choice between a future of shared abundance and one of scarcity and conflict. Both outcomes are possible at the same Kardashev value. What distinguishes them is not energy but judgement, and specifically whether humanity keeps its most capable systems accountable to it. If the transition brings short-term turbulence, the long-term abundance it promises will need to be measured on all three dimensions. Kardashev asked how much power a civilisation commands. The question for this century is whether a civilisation still commands its power.

---

Box 1. A note on Types IV to VII

The earlier version of this essay described Types IV to VII: civilisations commanding the observable universe, a multiverse, all logically possible realities and beyond. These extensions come from futurists and fiction rather than the scientific literature, and they quickly move beyond known physics. The total starlight of the observable universe is a meaningful, if vast, quantity. “Multiversal” or “omniversal” energy is not a defined physical quantity at all. These ideas appear in works from Olaf Stapledon’s Star Maker to Liu Cixin’s Three-Body trilogy, and readers can enjoy them as such. They are not part of the scale Kardashev proposed.

References

1. Kardashev, N. S. (1964). Transmission of information by extraterrestrial civilizations. Soviet Astronomy, 8, 217.
2. Dyson, F. J. (1960). Search for artificial stellar sources of infrared radiation. Science, 131(3414), 1667–1668.
3. Sagan, C. (Ed.) (1973). Communication with Extraterrestrial Intelligence (CETI). MIT Press.
4. Jiang, J. H., Feng, F., Rosen, P. E., Fahy, K. A., Das, P., Obacz, P., Zhang, A. & Zhu, Z.-H. (2022). Avoiding the Great Filter: predicting the timeline for humanity to reach Kardashev Type I civilization. Galaxies, 10(3), 68. doi:10.3390/galaxies10030068
5. Energy Institute (2026). Statistical Review of World Energy 2026. London: Energy Institute, with KPMG, Kearney and Ember.
6. International Energy Agency (2025). Energy and AI. Paris: IEA.
7. International Energy Agency (2026). Key Questions on Energy and AI. Paris: IEA.
8. Griffith, R. L., Wright, J. T., Maldonado, J., Povich, M. S., Sigurðsson, S. & Mullan, B. (2015). The Ĝ infrared search for extraterrestrial civilizations with large energy supplies. III. Astrophysical Journal Supplement Series, 217, 25.
9. Suazo, M. et al. (2024). Project Hephaistos II: Dyson sphere candidates from Gaia DR3, 2MASS, and WISE. Monthly Notices of the Royal Astronomical Society.
10. Ren, T., Garrett, M. A. & Siemion, A. P. V. (2024). Background contamination of the Project Hephaistos Dyson sphere candidates. arXiv:2405.14921.
11. Archival diagnostics for potential background contaminants of Project Hephaistos Dyson sphere candidates (2026). arXiv:2607.03619.
12. Landauer, R. (1961). Irreversibility and heat generation in the computing process. IBM Journal of Research and Development, 5(3), 183–191.
13. Lloyd, S. (2000). Ultimate physical limits to computation. Nature, 406, 1047–1054.
14. Ćirković, M. M. (2015). Kardashev’s classification at 50+: a fine vehicle with room for improvement. Serbian Astronomical Journal, 191.
15. Jevons, W. S. (1865). The Coal Question. London: Macmillan.
16. Barrow, J. D. (1998). Impossibility: The Limits of Science and the Science of Limits. Oxford University Press.
17. Vidal, C. (2011). Black holes: attractors for intelligence? arXiv:1104.4362; and Vidal, C. (2014). The Beginning and the End. Springer.
18. Hanson, R. (1998). The Great Filter: are we almost past it?
19. Smart, J. M. (2012). The transcension hypothesis: sufficiently advanced civilizations invariably leave our universe. Acta Astronautica, 78, 55–68.

Dr Luke Soon is an AI futurist, ethicist and author of Genesis: Human Experience in the Age of Artificial Intelligence. This essay is a revised and expanded version of his August 2025 piece on the Kardashev scale.