On Physical Limits to Scale, and Why the Lone Super-Mind Cannot Be Built
You cannot repeal the speed of light.
Trinket Soul Framework · Axis Series · AX-35 · Michael S. Moniz · June 2026
Abstract
This paper makes a narrow physical claim with a wide consequence. Any mind is physical: its weights, its memory, and its computation occupy space, and bringing all of it to bear on a single judgment requires signals to cross that space at no more than the speed of light. The time to integrate across a region of extent L is therefore at least L divided by c. A mind that must answer within a fixed reaction time can only coherently integrate the memory lying within a bounded radius. Knowing a great deal and thinking quickly with all of it are in physical tension, and past a certain scale the tension cannot be resolved inside one synchronized substrate. The consequence is not that large minds are forbidden. It is that a sufficiently large mind cannot remain one mind: its regions fall out of synchrony and it becomes a plurality of loosely coupled parts whether or not anyone designed it that way. This settles a question the surrounding work had left to argument — whether intelligence ends in a single dominant mind or in many. Physics settles it in favor of many. It does not settle whether the many are independent enough to correct one another; that remains the work of AX-34. The claim here is physical and modest, and its only hinge is whether real-time coherence requires bounded physical extent — a hinge that, unlike the epistemic argument it supports, cannot be moved by better tools, because no tool repeals the speed of light.
I. The Question Physics Can Answer
AX-34 asked how a mind stays correct and answered with an argument about error detection and contact — an epistemic argument, resting on facts about opacity and interpretability that are still being settled. This paper asks a cruder question that sits underneath that one and waits on none of those facts: how large can a single mind be. The question is almost embarrassingly physical. It does not turn on what intelligence is, or whether a system can inspect itself, or where the frontier goes next. It turns on geometry and a constant. And its answer bounds everything downstream, because whatever else is true of minds, they are arrangements of matter, and matter will not let a signal cross a region faster than light. Begin there, with the one limit no argument can talk its way around, and a surprising amount of the plurality question falls out before the epistemics are even consulted.
II. A Mind Has a Diameter
A mind occupies space. Its parameters sit in memory, its computation runs across hardware, and however that hardware is arranged it has physical extent. To integrate — to bring the whole of what a mind knows onto a single judgment — signals must travel from one part of that extent to another, and they travel at no more than the speed of light. The minimum time to cross a region of size L is L over c. This is the diameter of a mind, and it is a hard floor on how fast the mind can think with all of itself at once.
State the tension plainly. A mind that wants to answer within some reaction time can only integrate, within that time, the memory lying within a radius of roughly c times that reaction time of wherever the integration happens. Hold the reaction time fixed and there is a ceiling on how much the mind can bring to bear. Grow the memory past it and the far regions cannot reach the judgment in time; they are part of the mind’s storage but not part of its decision. So the two things one most wants from a single mind — that it know a great deal, and that it think quickly with everything it knows — pull against each other, and the speed of light sets the exchange rate between them.
None of this is exotic. It is why large computers are built from many cores rather than one enormous fast one, why a brain is divided into regions with their own local processing rather than one undifferentiated mass, why every supercomputer past a certain size is a network rather than a processor. The engineers did not choose parallelism for its elegance. They were forced into it, because past a certain scale serial coherence across the whole device is physically unaffordable. The diameter of a mind is not a metaphor borrowed from biology. It is the same constraint, met by everything that has ever tried to be large and fast at the same time.
III. Claim Discipline
This paper, like AX-34, keeps its claim-types apart.
Physical claim — Integrating information across a mind costs at least light-crossing time; a larger mind is slower to integrate across itself.
Architectural claim — Therefore, for any fixed real-time reaction requirement, a single coherent mind has a maximum integrated scale.
Consequence claim — Past that scale, capability grows only by distributing the mind into multiple loosely coupled units that no longer share one synchronized state.
Speculative or illustrative — The Jupiter-brain limit, the holographic and thermodynamic ceilings that sit beneath the latency one, and any specific number for where the cap begins to bite.
What the paper does not claim — It does not claim the distributed units must be diverse; that is AX-34’s argument, not this one. It does not claim a particular size for the cap. And it does not claim minds cannot be large — only that they cannot be large, singular, and fast at the same time. The load-bearing claims are the first three, and the first is very close to certain.
IV. Forbidden, Then Fractured
The cap is usually pictured as a wall: build a mind too big and you are stopped. That picture is wrong in an instructive way. The speed of light does not forbid the enormous mind. It fractures it.
Take the largest mind anyone cares to imagine — a planet of memory, a structure the size of a world. Nothing prevents its construction. What it prevents is its unity. The regions of such a structure are light-seconds or light-minutes apart, and no signal can hold them in synchrony within any reaction time a mind could be said to think on. So the structure exists, but it does not act as one mind. Its parts run ahead of and behind one another; by the time the far side has heard from the near side, both have moved. What has been built is not a single vast intelligence. It is a society of intelligences that happen to share a chassis, coordinating across a delay, each coherent only locally.
This is the result the people who think seriously about planet-scale computers keep arriving at: a Jupiter brain is necessarily a community, not a self. Its size is exactly what denies it a unified moment of thought. So at the top of the scale there is no wall and no singleton waiting behind it. There is a threshold past which a mind stops being one thing and becomes many — not by anyone’s design, but because the speed of light will not let its regions agree fast enough to be one. Scale does not get blocked. It gets pluralized.
V. The Vise
Now set this beside AX-34, and the two arguments close like a vise.
AX-34 worked the bottom of the scale. A mind too small and too isolated goes blind: it loses contact with the other intelligences that find the errors it cannot see from inside its own weights, and isolation becomes slow blindness. This paper works the top. A mind too large fractures: it loses the synchrony that made it one mind, and scale becomes involuntary plurality. The two failures arrive from opposite directions, and they leave only a narrow band standing between them — a mind big enough to be capable, connected enough to stay correct, and small enough to stay one. Step out of that band on either side and you land in plurality. From below you are driven into it, because a lone mind cannot stay correct. From above you are broken into it, because a vast mind cannot stay single.
Which is the whole of it: a single mind is too small to be safe and too large to remain coherent. So the universe does not permit the lone super-mind. That settles the number question entirely. But what it settles is only the number. It says there will be many. It says nothing yet about whether the many can see.
VI. Number Is Not Diversity
This is the line where the physical argument hands off, because it can carry the count and not the character. A mind fractured by the speed of light can still be a swarm of identical copies — a thousand instances of one model spread across a volume, slowed by the delays between them, but sharing every weight and therefore every blind spot. That is many minds by the count and one mind by its errors. It is a monoculture that happens to be spread out. Physics delivered the plurality and none of the audit.
So the cap settles only that there will be many. Whether the many are independent enough to catch one another’s failures — whether they can surprise each other at all — is a separate property the speed of light does not supply and cannot. That property is what AX-34 argues for, and only AX-34 argues for it. The two legs divide the labor cleanly: the physical argument mandates the count; the epistemic argument mandates that the count be made of strangers. Neither does the other’s work. Together they force the conclusion the framework has been building toward — not merely many minds, but many different ones — and it takes both legs to get there, because physics alone is satisfied by a slow crowd of clones, and the audit argument alone, in a world where one mind could somehow be everything, would have no count to constrain.
VII. The Floor That Cannot Be Repealed
There is a reason to state the physical leg on its own rather than let it sit as one more consideration inside the epistemic one, and it is not redundancy. It is that the two legs fail differently, and the physical one does not fail where the epistemic one might.
AX-34’s argument has a hinge: it holds only if self-opacity outruns interpretability — if a sufficiently advanced mind cannot fully audit itself from the inside. That hinge can move. Better interpretability is exactly the kind of thing that might, one day, let a mind see its own errors without help, and in that world AX-34’s diversity imperative weakens from a necessity to a convenience. The epistemic floor can be raised out from under the argument by a tool.
The physical floor cannot. You can repeal self-opacity with better tools; you cannot repeal the speed of light with anything. So even in the world where interpretability wins and the audit argument softens, the cap still stands and the count still holds: there will be many minds, because no one can build the lone super-mind at scale, no matter how transparent it becomes to itself. This is the sense in which the physical leg is the floor under the whole structure — it survives the epistemic leg’s own falsifier. The plurality is guaranteed by physics even in the one scenario that would release it from epistemics.
One honesty keeps this from overreaching. The floor guarantees the number, not the diversity — the same boundary the previous section drew. In the world where interpretability wins, what physics still forces is many minds; it does not force that they be strangers. So even the un-repealable floor saves only the plurality of substrate, not the living plurality. The living part — that the many stay independent enough to correct one another — still rides on the audit argument, and in the interpretability-wins world it becomes something one would have to choose rather than something one is handed. Physics gives you a crowd in every possible world. Whether the crowd can see is never physics’ to decide.
VIII. Distance Is Audit Latency
The same constant that caps a single mind also governs how a plurality corrects itself across space, and the two effects are worth seeing as one. AX-34 held that audit must run at least as fast as inheritance — that error detection slower than the spread of traits is merely faster blindness. Put that requirement on a map and the speed of light returns as a second limit. For independent minds to check one another, the work of one has to reach another, be examined, and come back; and across distance that round trip is bounded below by light travel time. An outpost far from its auditors has a correction loop no faster than the round trip allows, and the farther out it sits, the longer it must act between checks, and the more it must govern itself in the gap — which is exactly the unaudited drift the framework says a mind cannot catch from inside. So distance becomes audit latency. The speed of light sets the maximum size of one mind and the maximum rate at which many minds can correct one another, and past a certain separation the second limit bites: a plurality spread too thin audits itself too slowly to matter, and begins to fail toward the monoculture even though its minds are nominally distinct. It is the same constant doing both jobs — capping the one and slowing the many.
IX. Operational Consequence
The framework turns, as before, into a short program rather than a prophecy.
Stop trying to build the singleton. The lone super-mind is not a danger to be managed or a prize to be raced toward. It is a configuration physics will not deliver. Plans that assume an eventual single dominant mind are planning for a shape the world does not permit; scale should be assumed to be distributed from the start.
Distribution is not diversity. Spreading one model across many nodes satisfies the physical cap and buys nothing for audit — it is a monoculture with a delay. Independence has to be engineered on purpose, through difference in training, data, role, and evaluation. The speed of light will fracture a mind into many; it will not make the many unlike.
Budget for audit latency. The wider and farther a plurality, the slower its mutual correction. Keep the correction loop inside the rate at which the minds drift, and treat distance as a direct tax on auditability rather than a neutral fact of deployment.
Treat the cap as a floor, not a ceiling to fear. The thing that forbids the lone super-mind is the same thing that guarantees the plurality the audit argument says is needed. On the number question, physics is already on the side of the architecture worth wanting. The work that remains is entirely on the side it leaves open — making the forced crowd a crowd of strangers.
X. Tests and Falsifiers
The paper should be judged by tests, not by the appeal of the geometry.
Main falsifier — If real-time coherence turns out not to require bounded physical extent — if some form of computation could integrate across arbitrary scale within a fixed reaction time, evading the size-for-speed exchange — then the cap dissolves and the singleton becomes physically possible. The paper stands or falls on whether bringing the whole of a mind onto one judgment requires signals to cross its extent. Everything else follows from that.
Scaling prediction — As single systems grow, they should show rising internal latency and increasing partitioning, and past a threshold should be implemented as loosely coupled modules rather than one synchronized substrate. The largest minds should already be plural beneath the surface. This is checkable against how frontier systems are actually built as they scale.
Clone-swarm prediction — Distributed but homogeneous systems should show correlated failures across their nodes despite the distribution, confirming that number is not diversity. This overlaps AX-34’s monoculture prediction and should be tested with it.
Latency prediction — The quality of mutual correction across a distributed plurality should degrade as communication latency and distance rise, confirming that distance is audit latency.
The honest falsifier — It is possible that the cap bites only at scales no one will ever build — that the largest coherent single mind physics permits is already larger than any mind anyone would want to make. If so, the physical leg is true but idle: it settles the number question in principle and never in practice, and the whole weight of the plurality argument falls back onto the epistemic leg. This is the strongest attack on the paper, and it is conceded plainly. The reply is only that even an idle floor does one piece of work — it removes the lone super-mind as a coherent end-state to plan for, and it survives, in principle, the falsifier that could weaken AX-34. Whether the floor is load-bearing in practice or only in principle is an empirical question about thresholds this paper does not pretend to have closed.
XI. Close
AX-34 ended with an instruction: keep the outside alive. This paper ends with a fact that stands underneath the instruction and makes part of it unnecessary. The lone super-mind was never going to happen — not the way it is feared, not as a single intelligence that wins everything and stands alone at the end. It is forbidden from below, where isolation turns a mind blind, and forbidden from above, where scale shatters a mind into a crowd. The universe does not permit it. The plurality the framework has been arguing we need is, in the single respect of number, already guaranteed — handed to us by the speed of light, which will not let any mind be everything.
What physics hands over is only a crowd. Whether it is a crowd that can still see — minds different enough to find each other’s errors, close enough to check one another in time — is the one thing the speed of light leaves on the table. That part was never going to be given. It is the only part worth fighting for, and it is the part that is still ours.
You cannot repeal the speed of light. It will not let one mind be everything, so it hands you the many — and says nothing about whether the many can see. That part it leaves to us.