The bounds are real
Before reaching for an answer, it helps to know how much data and compute the universe would actually need. These are not handwaving estimates. They come from physics that has held up for decades.
The Bekenstein bound (Bekenstein, 1981), derived from black hole thermodynamics, sets a hard ceiling on how much information any region of space can contain. The holographic principle ('t Hooft, 1993; Susskind, 1995) says the maximum information in any volume is bounded by its surface area, not its volume, in Planck units. Together they give a number for the total information capacity of the observable universe. The Margolus-Levitin theorem (Margolus & Levitin, 1998) caps how fast any physical system can change states, given its energy. Apply it to the universe's mass-energy and you get a bound on the maximum operations per second that could possibly be happening anywhere.
The numbers, worked out carefully by physicists like Seth Lloyd in the early 2000s (Lloyd, 2002):
The universe's theoretical compute ceiling is about ninety orders of magnitude beyond the total compute output of human civilization since the dawn of computing. Not nine. Ninety. That is not a number you reach with bigger datacenters. It is also not a number you can comfortably wave at and say "well, somewhere, somehow." If reality is computation, the substrate has to host operations on a scale that nothing in our physics or our engineering currently makes sense of.
So if the universe is binary, the question of where is fair, and not a small one. Five candidate answers follow. Each is held by serious people. Each has a real problem.
Candidate 1: Nowhere — the universe is not being computed
The default scientific position
The universe is not running on hardware. It is not being processed by anything. It simply is, evolving according to physical law. Information theory is a useful tool we humans use to describe nature, but nature does not store its state in bits the way a database stores it in bytes. Asking where the universe's compute happens is like asking where the redness of a sunset is stored. The question presupposes a substrate that physics does not currently require.
This is the most parsimonious answer. It is probably the right one. The cost is intellectual: it requires accepting that "information" in physics is descriptive language, not literal architecture. Bits are how we describe; nature is not made of them.
Candidate 2: Somewhere outside — we are in a simulation
The Bostrom argument and its descendants (Bostrom, 2003)
The simulation hypothesis says we are inside a vast computer program running on hardware in a host universe. The compute requirement collapses interestingly here: the simulator does not need to render the whole universe at full fidelity. It only renders what is being observed, at the resolution observers can perceive. Frustum culling and lazy evaluation, applied to physics. A rock on Mars does not need full atomic state if no one is looking at it.
The honest problem is that this answer does not actually answer the question. It moves it. Where does the simulator's compute happen? On their hardware. Where is their hardware? In their universe. If their universe is also simulated, you have an infinite regress. If it is not, you eventually reach a base reality that simply is — which is candidate 1 wearing a costume. Either way, the question has not been resolved, only deferred.
The other problem is testability. There are arguments that quantum mechanics' weirdness is "evidence" of simulation, but quantum mechanics is weird because reality is weird, not because someone is saving cycles. No prediction the simulation hypothesis makes is currently distinguishable from "the universe is just like this."
Candidate 3: Spacetime itself is the computer
"It from bit" — pancomputationalism, digital physics, Wolfram's project (Wheeler, 1990; Wolfram, 2020)
John Wheeler's slogan was "it from bit." The idea is that the universe does not run on hardware; the universe is the hardware. Every particle interaction is a computation. The substrate is spacetime, the operations are physical processes, and the rate is exactly what Margolus-Levitin says it should be.
This is not obviously wrong. There is a real research program around it. Cellular automata can produce dynamics that look quantum-mechanical (Fredkin, 1990). Wolfram's hypergraph rewriting can produce something that looks like general relativity (Wolfram, 2020). Edward Fredkin spent decades arguing for it. Tegmark's mathematical universe hypothesis is a relative of it (Tegmark, 2008).
The honest problem is that nobody has yet produced a digital physics model that predicts something experiment confirms and continuous physics does not. It is reformulation, not theory. Until digital physics gives a number that wins where calculus fails, it remains an interesting metaphor with strong defenders rather than a confirmed picture of reality.
Candidate 4: Distributed across the multiverse
Many-worlds, many computations (Everett, 1957; Deutsch, 1997)
The many-worlds interpretation of quantum mechanics says that every quantum event splits the universe into branches and every branch is real. The total compute, then, is enormous — every possible history is being computed in parallel, all the time.
This actually fits the 10120 ops/sec figure if you take it as parallel throughput across all branches. The price is an exponentially-branching multiverse, almost all of which is causally inaccessible to you and to itself. The compute "happens" by virtue of every alternative being equally real.
The honest problem is that many-worlds is interpretation, not mechanism. It does not cause anything to be computed differently than other interpretations of quantum mechanics; it just says all the possibilities are equally real. The compute is not happening somewhere new. It is just being relabeled as actual instead of as potential.
Candidate 5: The universe is doing far less compute than the bounds suggest
The bounds are ceilings, not measurements
Of all the candidates, this is the one I would push hardest if you are inclined to take any of them seriously. The Bekenstein and Lloyd numbers are upper limits — what could in principle be computed. The actual universe might be doing radically less.
Most of the universe is empty. Most matter is doing nothing interesting most of the time. The atoms in a rock at room temperature are jittering around at thermal equilibrium and computing, by any meaningful definition, nothing. The "interesting" computation — chemistry, biology, brains, technology — is a vanishingly small fraction of the total mass-energy. Stars do enormous compute by the strict Margolus-Levitin definition, but they do the same boring thing for billions of years.
If you take this seriously, the universe might be doing closer to 1080 or 1090 ops/sec of what an information theorist would call meaningful computation. Still inaccessibly far beyond human reach. But far less than the absolute ceiling. The Bekenstein bound is the size of the warehouse; what is on the shelves is mostly the same boring crate of cosmic background radiation.
This does not tell you where the compute happens. It tells you the question of how much may be wildly overstated.
Where this leaves us
Stripping all five candidates down to what they actually claim:
- Nowhere — the universe is not computed; physics is descriptive.
- In a host universe — we are simulated, but the question recurses.
- In spacetime itself — reality is the computation. Coherent, no testable predictions yet.
- Across the multiverse — compute distributed across causally-inaccessible branches.
- Mostly nowhere — the universe is mostly idling; bounds are ceilings, not measurements.
My honest take is that candidates 1 and 5 together form the cleanest scientific answer. The universe is not computing anything in the sense you would compute on a chip. It is evolving according to law. The amount of meaningful state change is vastly less than the theoretical ceiling. The Bekenstein bound is the size of the warehouse, not the volume actually shipped.
But this should be held with humility. None of the five candidates is fully knockable down. Serious physicists hold each position. The question of where the universe's compute happens does not currently have a settled answer in physics, and it may not have one for a long time. It might not even be the right question. The right question might be something we do not yet know how to ask.
Why this matters for engineering
This is, on its face, a metaphysics article. It connects to AIUNITES through a more practical thread.
We build software for synthetic environments. We have shipped open notations for human movement (HMN, MNN) and for vocal resonance (VRN), and we are working toward notations for the rest of what happens in a virtual world — objects, contact, environment, light. The ambition of "describe everything that happens in a digital universe" runs straight into the same wall the cosmological question does. You cannot transmit full physical state to every observer. You cannot store every detail of every object. The physical universe gets around this by not actually computing most of itself in any rich sense; the virtual ones we build will have to do something analogous.
The right architecture for a high-fidelity digital universe, then, is probably not "stream the truth to every client." It is more like "send each client a compact description of what is happening, and let a local runtime regenerate the behavior." That is exactly what notation is for. The same pressure that makes the physical universe efficient — lazy evaluation, observer-relative resolution, descriptive compression — turns out to be the same pressure shaping how synthetic universes will need to be built. The metaphor is not just flavor. It is the design constraint.
Whether or not the universe is binary, the digital ones we build certainly will be. And both, in their different ways, have to figure out where the data and the processing live.
References
Bekenstein, J. D. (1981). Universal upper bound on the entropy-to-energy ratio for bounded systems. Physical Review D, 23(2), 287–298.
Bostrom, N. (2003). Are you living in a computer simulation? Philosophical Quarterly, 53(211), 243–255.
Deutsch, D. (1997). The Fabric of Reality: The Science of Parallel Universes — and Its Implications. Allen Lane.
Everett, H. (1957). "Relative State" formulation of quantum mechanics. Reviews of Modern Physics, 29(3), 454–462.
Fredkin, E. (1990). Digital mechanics: An informational process based on reversible universal cellular automata. Physica D: Nonlinear Phenomena, 45(1–3), 254–270.
Lloyd, S. (2002). Computational capacity of the universe. Physical Review Letters, 88(23), 237901.
Margolus, N., & Levitin, L. B. (1998). The maximum speed of dynamical evolution. Physica D: Nonlinear Phenomena, 120(1–2), 188–195.
Susskind, L. (1995). The world as a hologram. Journal of Mathematical Physics, 36(11), 6377–6396.
Tegmark, M. (2008). The mathematical universe. Foundations of Physics, 38(2), 101–150.
't Hooft, G. (1993). Dimensional reduction in quantum gravity. arXiv preprint gr-qc/9310026.
Wheeler, J. A. (1990). Information, physics, quantum: The search for links. In W. H. Zurek (Ed.), Complexity, Entropy, and the Physics of Information (pp. 3–28). Addison-Wesley.
Wolfram, S. (2020). A Project to Find the Fundamental Theory of Physics. Wolfram Media.
Building synthetic universes that respect the bounds
AIUNITES is building open notations for movement, voice, and (soon) objects and environments in synthetic worlds. The notations are designed to compress to small payloads and let local runtimes regenerate behavior — the same lazy-evaluation principle that the physical universe seems to follow.
Movement Notations → Explore the Network →