Matrioshka Brains and the Kardashev Scale: What Civilization-Scale Computing Actually Looks Like
April 15, 2026
We measure progress in product cycles. Quarterly earnings. GPU generations. The latest benchmark. Physicists and SETI researchers have a different unit of measure: civilizational energy consumption. By that metric, we're roughly 0.73 on a scale that goes to three, and the distance between where we are and where that scale tops out is not linear. It's almost incomprehensible.
The Kardashev Scale was proposed by Soviet astrophysicist Nikolai Kardashev in 1964 as a way to classify civilizations by energy use. It's still the clearest framework I've come across for thinking about what technological progress looks like at long timescales, and why the ceiling is so much higher than anyone tends to acknowledge.
The Kardashev Scale
Type I is a planetary civilization — one that has mastered all energy available on its home planet, including solar power hitting the surface, geothermal, oceanic, and atmospheric energy. That's roughly 10^16 watts. We're currently at about 2×10^13 watts of global power consumption. Close, but not there. The conventional estimate puts humanity around 0.73.
Type II is a stellar civilization — one that harnesses the full output of its star. Our sun radiates approximately 3.8×10^26 watts. That's ten billion times more than Type I. The canonical mechanism is a Dyson sphere: a megastructure that encases a star and captures its radiation. The more realistic version, a Dyson swarm, is a dense cloud of solar collectors orbiting in a coordinated pattern. Nothing about it violates physics. The materials science and orbital mechanics required are just several orders of magnitude beyond anything we've attempted.
Type III is a galactic civilization — one that harvests the energy output of an entire galaxy, roughly 10^36 watts. We don't have a clear mechanism for this. It likely involves Type II approaches scaled across millions of star systems. What a civilization even looks like at that scale is hard to reason about. The gap between Type II and Type III is larger than the gap between us and Type II.
The Matrioshka Brain
Once you've built a Dyson swarm around a star, you don't just have energy. You have substrate. Robert Bradbury, drawing on earlier work by Freeman Dyson, described where that leads: the Matrioshka Brain.
The concept is a nested set of concentric Dyson shells, each one capturing the waste heat from the shell inside it and using that energy for computation. The innermost layer runs at the highest temperature and does the most intense processing. Each outer layer operates at progressively lower temperatures — and since computation obeys thermodynamics, colder processors are more efficient. The outermost shell radiates waste heat into space at just a few degrees above absolute zero.
The total computational capacity is estimated somewhere in the range of 10^42 operations per second. Your laptop does around 10^11. The human brain does somewhere around 10^17. A large AI inference cluster — the kind running frontier models today — sits around 10^19. That's impressive by human standards, and two orders of magnitude above a single brain. It's also twenty-three orders of magnitude below a Matrioshka Brain. A Matrioshka Brain exceeds the combined computing capacity of every human brain that has ever existed by many tens of orders of magnitude. The number is so large it stops feeling like a number.
What does "thinking" look like at 10^42 operations per second? I genuinely don't know, and I don't think anyone does. You could model the quantum state of every particle in a human brain and simulate centuries of subjective experience per real second. You could run millions of independent civilizations in simulation. You could solve problems that would take every current computer on Earth longer than the age of the universe. If you're going to build a Dyson swarm anyway, adding computation to each shell as a secondary function is just the rational next step.
TerraFab: A small step in the right direction
None of this is happening soon. We're not even close to Type I. But there's a through-line between where we are now and what a Type II civilization looks like, and it starts with controlling your manufacturing substrate.
Elon Musk's TerraFab chip manufacturing project is an effort to build semiconductor fabrication capacity in Texas. The goal is to reduce dependence on foreign fabs. TSMC in Taiwan currently handles the overwhelming majority of advanced node production globally, which is a supply-chain vulnerability that's hard to overstate. TerraFab is framed as chip sovereignty: the ability to produce the hardware a civilization needs on its own soil.
In the civilizational computing context, this matters more than it might look. A Matrioshka Brain starts with silicon wafers. More precisely, it starts with a civilization that can produce computational substrate at scale without depending on anyone else. Every major leap in computing has started with controlling the fabrication layer. TerraFab doesn't put us on the path to a Dyson sphere, but it's on the same axis.
Where this leaves us
The Kardashev Scale is useful not as a roadmap but as a perspective adjustment. When you're thinking about the next GPU architecture or the next data center efficiency gain, occasionally zooming out to the point where the sun is the power supply and the solar system is the data center is clarifying. That's the ceiling. We've barely left the ground floor.
The physics allows it. Whether we're around to do it is the real engineering problem.
