Oh, and Here’s Why Communism is Physically Impossible
A thermodynamic proof that central planning is physically incoherent
As a special bonus edition, here is a thermodynamic proof that central planning is physically incoherent. Happy 250th America.
Well, well, communism. I’ve been working through social thermodynamics for my upcoming book, The World Destroyer’s Handbook, when it dawned on me that communism and socialism cannot be doing what they say they are doing — it isn’t possible.
TL;DR: Coordinating N agents in real-time requires power that scales linearly with N. Beyond a critical population size, required power exceeds available power. No amount of better computers, AI, or benevolence changes this. The problem is recursive, the data is the high resolution readout of each person’s life, it accumulates with every passing nanosecond, and it physically cannot be transmitted from individuals to planners. Markets work because they distribute the computation rather than bottlenecking it at a central node.
A Summary Dismissal of Fantastical Regimes
Central planning at scale is not a policy option that sometimes works and sometimes fails. It isn’t real. It cannot exist as a physical process.
When central planners claim to be coordinating an economy of 100 plus million people, they are engaged in a collective fantasy — writing plans that cannot be instantiated in physical reality any more than those blueprints for your perpetual motion machine will become a reality. The “central planning” isn’t even happening. What’s actually happening are arbitrary directives, imperfectly imposed, on top of a real economy that preserves energy, movement, and coordination through local decisions below the threshold of official scrutiny, black markets, informal networks, and adaptation around impositions. Do you know what that is? A phenomenon known as a distributed market economy. That’s right — the same type of economy that the regime is pretending to have eliminated. Every communist and centrally-controlled socialist regime that claimed to “centrally plan” was both lying and delusional. The Soviet Union didn’t fail because of poor implementation or corruption; it failed because what it claimed to be doing—the central coordination of 280 million people—cannot exist as a physical process. The plans were fiction. The coordination was impossible. The whole apparatus was theater. An aspirational performance, driven by men living out a fantasy of impossible abstractions, on top of a real economy that continued working in spite of — and not because of — their central ahem “direction.”
The entire century of debate between free markets and communism or socialism is not a debate between two viable alternatives. It is a debate between a physically possible system and an incoherent fiction dressed up as policy. The communists and socialists are arguing for something that cannot exist, has never existed, and will never exist. This is not because human nature won’t allow it, or because of the political challenges in fomenting “the revolution,” but because it violates physics.
No central planner has access to sufficient data, computational capacity, real time feedback, and authoritarian control for the idea to actually occur. When we say an economy is centrally planned, we are talking about a free market economy that continues to function in spite of ridiculous constraints being put upon it by delusional enforcers, acting the part of a “tribal signaling nexus,” not economic planners. The central planner lacks the real world information to be able to react to constantly changing conditions, unfolding as a thermodynamic and irreversible process. The information each person in an economy possesses is high definition, multivariate and updating in real time across dimensions that planners cannot physically access — the information is ontologically moored to each individual. The best any government can possibly do is get out of the way, so that local agents can coordinate freely. Any economic system that depends on government intervention in order to persist is already a dead man walking. The condition of needing central control is indistinguishable from an economy that is already on life support, that cannot persist, has already metastasized, and is now being animated by electrical shocks in the form of coerced extraction. It is economic misdirection at gunpoint. A system that needs government control to persist is already in thermodynamic decay, it is metastable and primed for collapse. Healthy economies are those that emerge naturally, exactly like how a healthy forest needs no human intervention. Forests only become high-risk fire hazards, precisely because humans have prevented them from burning. This is the identical approach to the planned economy. We paper over one mistake — artificially constraining the natural human impulse to survive, compete, and create value — with an escalating cascade of new mistakes. You cannot fix a problem created by too much friction BY ADDING MORE FRICTION.
At population scale, economic coordination requires massive irreversible computation — compressing everyone’s private preferences, information, and constraints into allocation decisions. The first problem is that this information never even makes it to planners. This was what economist Friedrich Hayek noted in the local knowledge problem. No central authority can ever gather or process the massive fleeting, localized knowledge in real-time rendering central planning structurally impossible. Well, Hayek was right. But the argument I am presenting goes even further. Even if a central node somehow could access this information — which they can’t, it’s impossible — they would still not be able to compute it. Computation dissipates energy. In markets, this work is distributed: each person does their own compression of reality locally (preferences that lead to decisions to buy or sell to open or close businesses and so on), and then prices coordinate through iteration as conditions change. The thermodynamic cost spreads across N nodes. In centralized planning, all of that irreversible computation must happen at a bottleneck — the bounded separator where feasibility gets enforced. The energy dissipation concentrates at one architectural point. At N ≥ 10^6, the required power exceeds what any physical system at that point can provide.
Communism isn’t simply wrong. It’s incoherent. It’s not a failed experiment. It isn’t even an experiment. Communism and socialism are just shitty free markets. It’s the policy equivalent of claiming you can build a machine that outputs more energy than it takes in. The question isn’t “Why did it fail?” but “Why did anyone think these pretty sounding words describing a physically impossible process constituted a plan for governance?” What communist regimes label “central planning” has always been distributed coordination, struggling to survive under pure dead weight loss operating below a layer of mathematically impossible moral assertions.
Markets aren’t better than communism or socialism — they are the only thing that exists.
Here’s the formal proof:
Consider an economy coordinating N agents under the following conditions:
Each agent has private state with decision-relevant conditional entropy h = H(private state | public signals) > 0
Coordination requires real-time feasibility—constraint violations cannot be deferred, averaged, or buffered against future capacity
Coordination becomes invalid at rate f (preferences/constraints change)
Information flow is constrained by coordination graph structure
Theorem: Any coordination architecture whose graph contains a bounded information cut faces three independent constraints that make real-time feasibility enforcement non-scalable:
(1) Communication throughput: Verifying coupled feasibility constraints requires information from all N agents. With private state changing at rate f, the cut must process O(Nf) messages per unit time. Shannon capacity limits throughput; bounded cuts create congestion.
(2) Control loop latency: Real-time feasibility enforcement operates as a control loop. With bounded cut capacity, coordination latency grows with N. Beyond a critical threshold, the loop becomes unstable or violates real-time constraints before corrective action completes.
(3) Thermodynamic floor (via Landauer): The computational operations at the cut — compressing high-dimensional private information into feasibility decisions — are logically irreversible. This imposes a substrate-independent minimum dissipation P_cut ≥ Ω(NhfkT). While engineering limits (bandwidth, latency, heat dissipation) typically bind before Landauer limits, this establishes that the constraint is fundamental to physical reality, not contingent on current technology.
These constraints are independent. Communication throughput alone is sufficient to establish non-scalability; Landauer provides a permanent floor that no future technology can overcome.
P_cut ≥ Ω(NhfkT)
where k is Boltzmann’s constant and T is temperature.
If you scale coordination nodes proportionally with N to avoid this bound, you have distributed the system, which is the claim. A bounded cut defines what centralization means thermodynamically.
This bound follows from applying Landauer’s principle to the irreversible compression of decision-relevant private information required for feasibility enforcement. It is a topological constraint on information flow through bounded separators, not an algorithmic or incentive limitation.
Why Landauer’s principle matters: This bound is substrate-independent. It applies regardless of hardware, algorithms, or technological advancement. You cannot get below kT ln(2) per irreversibly erased bit with any technology, ever. Engineering limits (bandwidth, latency, heat dissipation) are much tighter in practice, but they’re contingent—someone can always hope to overcome them with better technology. Landauer bounds cannot be overcome. This removes the escape route of technological optimism: you cannot solve this constraint with better computers, advanced AI, or future breakthroughs. The bound establishes that the constraint is fundamental to physical reality, not contingent on current technological limitations.
Technical note: The bound P ≥ Ω(NhfkT) follows from applying Landauer’s principle to irreversible operations at the cut. The critical step—that feasibility enforcement under coupled constraints and continuous change forces Ω(Nh) bit erasures per update at any bounded separator—requires formal proof beyond this article’s scope. The intuition: coupled feasibility constraints cannot be verified independently; each allocation decision requires information about all N agents’ states; under continuous change at rate f, previous computations cannot be cached; therefore the cut must process fresh information from all agents at rate f. A rigorous proof would formalize this as a lower bound on logically irreversible operations at bounded separators under constraint coupling. The architectural claim stands on network flow and control theory; Landauer provides the substrate-independent floor.
On the Ω(Nh) erasure requirement: The claim that feasibility enforcement under coupled constraints and continuous change forces Ω(Nh) bit erasures per update at any bounded separator requires formal proof beyond this article’s scope. The intuition: coupled feasibility constraints create global dependencies—you cannot verify agent i’s allocation is feasible without information about what all other agents received (total consumption ≤ production, no double-allocation of shared resources, mutual compatibility). Under continuous change at rate f, previous computations cannot be cached without becoming stale. Therefore the cut must process fresh information from all N agents at rate f. A rigorous proof would formalize this as a lower bound on logically irreversible operations at bounded separators under constraint coupling.
However, the architectural claim stands on communication and control-theoretic grounds alone. Even if Landauer bounds are loose, or if clever algorithms reduce the constant factors, bounded cuts still face O(Nf) communication requirements and latency-induced control instability. The thermodynamic argument is the third line of defense, not the first.
For any coordination node with available power P_max, the system crosses the coordination event horizon when N > P_max/(hfkT), beyond which maintaining real-time feasibility becomes physically impossible.
Distributed architectures eliminate bounded separators, preventing concentration of irreversible entropy processing at any single node. Total system dissipation remains Ω(NhfkT), but no individual node crosses the coordination event horizon. Price systems and markets aggregate only low-dimensional sufficient statistics—erasure occurs locally wherever information is generated rather than being funneled through a central coordinator.
Scope: This limit applies to systems requiring strict real-time feasibility with structural information cuts. It does not apply when coordination can be delayed, buffered probabilistically, or when shared priors/pre-commitment substantially reduce h.
Corollary: Above the Coordination Event Horizon, centralized feasibility enforcement ceases to exist as a physical process; the system cannot even instantiate the constraints it claims to enforce.
This is not ideology. It’s information theory.
On the Relationship to Hayek’s Knowledge Problem
Hayek showed that knowledge is dispersed and planners cannot access it. This proof shows something stronger: even if planners could perfectly access all dispersed knowledge, the computation required to coordinate N agents in real-time requires power P ≥ NhfkT. Beyond critical population size, this exceeds available power regardless of information access.
Hayek proved an epistemological impossibility. This proves a physical one.
Where Hayek argued planners don’t know enough, this proof shows they cannot compute it even if they did know. The computational bottleneck is independent of the knowledge problem — it’s a separate, more fundamental constraint arising from information theory and Landauer’s principle. Hayek identified the first barrier to central planning. This identifies a second, more fundamental one.
Both arguments are correct. This one is physical law.
End of article.
The following 3,000+ words are only necessary to ensure substantive engagement from pedantic AI agents and specifically adversarial readers. The clean theorem works precisely because it refuses to litigate implications that are already implicit in the setup.
A Note to Adversarial Readers
If you’re reading this looking for escape routes rather than evaluating whether the claim is true, here are the patterns you’re likely following. I’ve grouped them by type to make the underlying structure visible.
Scope note: The impossibility claim applies specifically to centralized real-time feasibility enforcement under coupled constraints with continuous change—the architecture Soviet ideology described when claiming Gosplan centrally coordinated 280 million people. The physics establishes this architecture cannot exist at that scale. What actually coordinated the Soviet economy were distributed mechanisms (hierarchical decision-making, local bargaining, informal networks, black markets) operating beneath the propaganda layer. The physical constraint explains why every centrally planned economy evolved these distributed workarounds rather than implementing the centralized ideal.
I. Definitional Evasions
Scope expansion: Taking a claim explicitly bounded to large-scale economies (N ≥ 10^6, real-time feasibility, changing constraints) and testing it against communes, corporations, small organizations, or static scenarios. The argument states its domain. If you’re applying it elsewhere, you’re not evaluating the argument—you’re changing it.
Causal substitution: Conflating “central planning as ownership structure” or “central planning as policy document” with “central planning as thermodynamic locus of real-time feasibility enforcement.” These are different things. The impossibility claim targets the last one specifically. If a government writes five-year plans while actual coordination happens through distributed networks, that’s not a counterexample—it’s confirmation.
The definitional rescue: Claiming “central planning doesn’t have to mean centralized coordination—it could mean distributed state protocols, sharded computation across ministries, or hierarchical decomposition.” This is semantic evasion. Central planning in the Soviet context meant centralized rational coordination from Gosplan replacing market mechanisms. That specific claim—centralized real-time coordination at population scale—is what the physics shows is impossible. Redefining “central planning” to include distributed architectures is just describing what actually happened beneath the propaganda layer. You’re trying to rescue central planning by making it not-centralized.
The “this only kills straw-man central planning” objection: Claiming the argument only refutes a naive caricature. The target is exactly the architecture central planners claimed to implement: centralized, rational, real-time coordination replacing market processes. If defenders retreat to buffered plans, local discretion, hierarchy, or price-like signals, they have abandoned that claim. The argument doesn’t mischaracterize the ideal; it explains why every real attempt was forced to betray it.
II. The Distribution Move (All Technical Workarounds Distribute the System)
Every sophisticated technical objection reduces to the same pattern: the proposed solution eliminates centralization. Here are the variations:
The decomposition retreat: Arguing that “coupled constraints can be decomposed via dual variables, Lagrange multipliers, or hierarchical control.” Correct—and what have you just described? A distributed coordination system. Dual decomposition with price-like multipliers is functionally a market mechanism. Hierarchical control with local autonomy distributes the computation. The objection proves the point: to scale, you must distribute coordination. If you call this “central planning” because the state owns the assets, fine—but thermodynamically it’s distributed coordination, which is exactly what the argument claims is necessary.
The sharding escape: Claiming “decisions can be central politically while computation is sharded across data centers, ministries, or regional units, so there’s no bounded cut.” Yes—that’s distributed coordination. You’ve eliminated the bottleneck by distributing the computation. This doesn’t falsify the claim; it confirms it. You’re saying centralized bottlenecks fail, so you must distribute. Whether you call the result “sharded planning” or “distributed coordination,” the architecture is distributed, which is what the constraint requires.
The “scale nodes with N” move: Claiming you can avoid the bounded cut by scaling staff, data centers, regional offices, or parallel pipelines proportionally with N, so aggregate capacity grows. Correct—and you’ve just built a distributed system. If coordination nodes scale with N, you no longer have centralized coordination. The bounded cut defines centralization thermodynamically: aggregate information-processing capacity that does not scale with N. Eliminating it by scaling nodes proportionally is conceding the argument.
The “energy can be parallelized” claim: Arguing that energy dissipation can be spread across many machines, so no single node hits a power wall. Correct—and that is distributed coordination. The claim is not that total dissipation must be centralized, but that if feasibility enforcement is centralized, dissipation concentrates at the separator. Spreading dissipation across machines eliminates the bounded cut. That’s not a counterexample; it’s compliance with the constraint.
The sparse constraints retreat: Arguing that most real economies have sparse or decomposable constraints, so dense coupling is a straw man. Sparsity is not a free assumption; it is an architectural achievement. Large economies with shared infrastructure, multi-stage production, complementarity, and common resources are not sparse by default. When sparsity exists, it exists because coordination has already been decomposed into semi-autonomous subsystems—because the system has been made distributed. Invoking sparsity is not a rebuttal; it’s an admission that centralized coupling had to be broken to scale.
The caching and incremental updates move: Arguing that with clever incremental algorithms, prior computations can be reused, avoiding linear work per update. Incrementalism works only when updates are local and constraints weakly coupled. Under global feasibility constraints and continuous change, cached state becomes invalid at rate f. If violations must be prevented in real time, stale caches are not admissible. Incremental algorithms are precisely the tools used in distributed systems to localize change. If they work, it’s because coordination has been decomposed. Success implies distribution.
The “aggregate suffices” evasion: Claiming that feasibility can be checked with low-dimensional aggregates (totals, flows, averages), so information need not scale with N. This fails because aggregates suffice only after a distributed process has already compressed private state into those aggregates. The question is not whether aggregates exist, but where and how they are produced. If they are produced centrally, the compression itself requires access to the underlying high-dimensional information. If they are produced locally and iteratively, the system is already distributed. Pointing to aggregates as inputs is pointing to the output of markets and pretending it was a free primitive.
The aggregate suffices evasion (worked example):
A specific instance of this pattern: “Consider N agents with private state, one shared resource with capacity C(t), requiring Σx_i(t) ≤ C(t). Upstream nodes combine children’s totals and forward a single number. The cut processes O(1) bits per update, not Ω(Nh). Therefore no dissipation concentration at the cut.”
This counterexample is distributed coordination, not a rebuttal to it:
“Upstream nodes combine children’s totals” - Computational work is distributed across the tree
“Forward a single number” - The cut receives pre-compressed summaries
“O(1) bits cross the cut” - Because O(N) compression work already happened upstream
The objector has eliminated the bounded cut by distributing feasibility computation across upstream nodes. Each upstream node processes information from its subtree, performs local compression, and passes summaries upward. The “central node” is just a final aggregation point—the actual coordination work is distributed across the network topology.
This is the argument, not a counterexample to it. The bound applies to architectures where coordination computation concentrates at a non-scaling separator. If you avoid this by distributing computation across scaling infrastructure (the upstream nodes), you’ve built a distributed system. That the root node receives O(1) summaries proves the work was distributed, not that centralization works.
The rhetorical move: call the root node “central” while hiding the distributed computation that made its job tractable. This is like claiming a CEO “centrally coordinates” a company because all reports reach their desk, while ignoring that the actual coordination work distributes across divisions, managers, and local decision-makers. The aggregates reaching the top are outputs of distributed processing, not evidence that centralization succeeded.
The sufficient statistics objection: Could the coordinator receive pre-compressed summaries (like demand curves in markets) that don’t scale with N? Only if the constraint structure is sparse or decomposable. In tightly coupled economies (complex input-output networks, complementary production, shared infrastructure), global feasibility checking requires processing coupled information from all agents. The coordinator cannot verify feasibility from low-dimensional summaries alone—those summaries are what markets produce through distributed iteration, not inputs to centralized computation.
Pattern recognition: Notice what every technical objection has in common—the proposed solution eliminates centralization.
“Use aggregates” → those are outputs of distributed compression
“Exploit sparsity” → requires decomposing into semi-autonomous subsystems
“Cache incrementally” → only works when constraints are weakly coupled and locally handled
“Parallelize across machines” → that’s distributed coordination
“Use dual decomposition” → that’s a market mechanism
Every sophisticated objection reduces to: “Central planning works if you don’t do central planning.” That’s not a rebuttal. That’s confirmation with redefinition.
III. Computational and Thermodynamic Objections
Mechanism ambiguity: Questioning whether coordination “really requires” centralized processing without engaging with why coupled feasibility constraints across N agents cannot be decomposed. The answer is in the mathematics of constraint satisfaction, not in whether you find the connection between coordination and bit erasure intuitive.
The reversibility objection: Arguing that “coordination might be computationally reversible even if outputs are low-dimensional.” Coordination requires selecting one feasible allocation from exponentially many possibilities. This selection destroys information about infeasible alternatives—you cannot reverse the allocation to recover all preference information that was considered and rejected. This is inherent to constraint satisfaction, not a computational convenience. The task is irreversible.
The reversible computation escape: Claiming that coordination could be done reversibly, avoiding Landauer dissipation. Reversible computation only avoids dissipation if information is preserved indefinitely. Coordination requires selection: one feasible allocation among many. That selection destroys information about rejected alternatives. Real-time feasibility under change requires overwriting stale state. Memory must be reset. Reset is erasure. Erasure costs energy. You can delay dissipation; you cannot eliminate it. Reversible logic does not save you from entropy when the task itself is lossy.
The Landauer magnitude objections (consolidating “Landauer is too weak,” “Landauer-tightness,” and “Landauer is tiny”): Noting that kT ln(2) (where k is Boltzmann’s constant, approximately 3×10^-21 joules per bit at room temperature) is small, or that engineering limits (bandwidth, latency, power dissipation) typically dominate. Correct—but this misses the strategic purpose. Landauer establishes a substrate-independent lower bound that cannot be overcome by any technology. Engineering limits are contingent—someone can always hope better hardware will overcome them. Landauer proves the constraint is fundamental. In practice, communication and control limits bind first, which only strengthens the case. The architectural requirement to distribute coordination follows from whichever physical constraint binds first, whether Shannon capacity, control stability, or Landauer’s principle. The magnitude is irrelevant to the role Landauer plays: closing the technological-optimism loophole. The bound need only be nonzero and linear to do its job.
The “queueing is just engineering” dismissal: Treating throughput, latency, and control instability as contingent engineering problems. Queueing instability is not an implementation bug; it is a theorem. If arrival rate exceeds service rate at a bounded server, delay diverges. No hardware upgrade changes the fact that a non-scaling service capacity facing linearly scaling arrivals will eventually fail. Calling this “engineering” is like calling Shannon capacity “an implementation detail.”
The “you’re smuggling in N” accusation: Claiming the linear scaling in N is assumed rather than derived. N enters through two independent facts: (i) private state exists per agent (h > 0), and (ii) global feasibility couples agents. If you deny linear scaling, you must explain how feasibility of N interdependent decisions can be certified while touching information from only o(N) agents per update. That is not a rhetorical objection; it is a concrete algorithmic claim that must be demonstrated.
IV. Procedural Evasions
Counterexample fishing: Pointing to large organizations that “work” without checking whether they actually concentrate feasibility computation at a central node, or whether they distribute it through internal markets, divisional autonomy, and hierarchical decomposition. The argument predicts exactly which architectures fail and which succeed.
Appeals to authority: Noting that “no established physicist has proven this” or “this isn’t in the economics literature” as if novelty equals incorrectness. Landauer’s principle is established physics (1961). Information theory is established mathematics. The application to economic coordination is the novel part—which is how all interdisciplinary arguments work. If your objection is “I haven’t seen this specific claim before,” that’s not an objection to the physics or logic, it’s just an observation about what you’ve read.
Absence as evidence: Claiming “there’s no established proof of this” as if that constitutes a rebuttal. The proof is in front of you. You can check whether Landauer’s principle applies to irreversible computation. You can check whether coupled feasibility constraints require centralized processing. You can check whether bounded cuts create throughput limits. The fact that this specific synthesis hasn’t appeared in a journal is irrelevant to whether the reasoning is sound. Evaluate the argument, not its publication history.
The precision-theater pattern: Acknowledging “your core architectural intuition is basically right” while demanding journal-level formal proofs before accepting the conclusion. This move looks like substantive engagement but is actually evasion. Here’s how it works:
Concede the qualitative claim (”bounded cuts create bottlenecks at scale”)
Demand formal proofs of every quantitative step (communication-complexity lower bounds, constraint-coupling requirements, bit-erasure calculations)
When the article acknowledges “formal proof beyond this article’s scope,” claim the argument “breaks” or “isn’t established”
Offer to “rewrite it to be rigorous”
This conflates “formal proof not provided in a Substack article” with “claim is false.” The article explicitly states which steps require formal proof beyond its scope while arguing the qualitative conclusion stands on established principles. If you accept that bounded cuts + coupled constraints + real-time enforcement create throughput and stability limits, you have accepted the conclusion. Demanding formal proofs before accepting what you’ve already conceded is evasion, not disagreement.
Theatrical Retreats to Precision: Demanding mathematical rigor for specific parameters while ignoring that the qualitative claim—power requirements scale with N, centralized architectures create bounded cuts, physical limits bind at population scale—stands regardless of exact coefficients.
The “you haven’t proved it formally” stall: Acknowledging the intuition while withholding assent pending full formalization. This is not skepticism; it’s delay. The argument identifies which assumptions imply non-scalability and which architectural escape routes exist. If you accept that bounded cuts + coupled constraints + real-time enforcement create throughput and stability limits, you have accepted the conclusion. Demanding journal-level proofs while conceding the mechanism is evasion, not disagreement.
V. Clarifying What Is NOT Claimed
The claim is not: all hierarchy is thermodynamically impossible. The claim is not: central ownership cannot exist. The claim is not: governments cannot write plans. The claim is not: the USSR violated physics.
The claim is: Centralized real-time feasibility enforcement across N ≥ 10^6 agents with coupled constraints and changing preferences, when architecturally concentrated at a bounded information cut, requires power that exceeds available capacity. What the USSR and other centrally planned economies claimed to be doing—centralized coordination at population scale from Gosplan—cannot exist as a physical process. What actually coordinated those economies were distributed mechanisms (hierarchy, local bargaining, informal networks, black markets) operating beneath the theatrical layer of central plans.
VI. The Substantive Core
Strip away all evasive moves and three technical claims remain:
(1) Large-scale feasibility enforcement is a coupled constraint-satisfaction problem. Economic feasibility requires: total consumption ≤ total production for each good, shared resources cannot be double-allocated, production plans must be mutually compatible. These are global constraints—you cannot verify agent i’s allocation without knowing what all other agents received.
Markets solve coupled constraints through distributed iteration; any system that solves them at scale must distribute the computation. The question is not whether coupled constraints can be solved, but where the computation occurs.
(2) Irreversible coordination entails entropy dissipation. Landauer’s principle (1961): erasing information requires minimum energy kT ln(2) per bit (where k is Boltzmann’s constant; at room temperature this is approximately 3×10^-21 joules per bit). Coordination compresses high-dimensional private information into low-dimensional outputs. This is lossy—you cannot reconstruct inputs from outputs. Information is destroyed. This is thermodynamically irreversible and must dissipate energy. Whether practical limits are set by Landauer bounds or engineering constraints (bandwidth, latency, power dissipation), the architectural problem remains: coupled constraint satisfaction at scale requires distributed processing.
(3) Bounded information cuts impose throughput limits. If system architecture requires all coordination-relevant information to flow through a separator with finite capacity, throughput is limited. This is Shannon’s channel capacity theorem. More critically: even with infinite communication bandwidth, the computational substrate at the receiving node has finite power capacity. Processing coupled constraints from N agents requires computation that dissipates energy at the central node. Physical hardware has power limits.
These three claims are independent and convergent. You could weaken any single one and the argument still holds:
Even if Landauer bounds are loose, coupled constraints create computational scaling problems
Even if you found algorithmic shortcuts, irreversible operations still dissipate energy
Even if communication bandwidth were infinite, the central computational substrate still has finite power capacity
The defenses don’t depend on each other. They’re independent constraints from combinatorics, thermodynamics, and physical hardware that all point to the same architectural limitation.
What substantive disagreement looks like:
If you think the argument is wrong, you must claim one of:
Coupled feasibility constraints at scale can be verified without information scaling with N (and explain how for dense coupling)
Bounded information cuts don’t create throughput limits (denying Shannon)
Real-time control loops remain stable with unbounded latency (denying control theory)
Physical computational substrates don’t have power/capacity limits
Or: accept all three claims but argue N* (the threshold where centralization fails) is much higher than 10^6, and provide calculations showing where power/throughput/latency constraints actually bind
Anything else—demanding more formalization, noting sparsity exists, claiming it’s definitional, or acknowledging the intuition while rejecting the conclusion—is evasion.
If your objection is “you haven’t formally proved the Ω(Nh) bit erasure requirement,” the article already states this requires proof beyond its scope. If you agree bounded cuts + coupled constraints + real-time requirements create non-scalability but reject the conclusion because formalization is incomplete, you’re engaged in motivated reasoning. The question is whether the claim is true, not whether every step is proved to journal standards in a Substack post.
The “markets aren’t special” deflection: Saying markets are just one of many distributed systems, so the argument doesn’t privilege capitalism. Correct—and irrelevant. The claim is not that markets are morally or politically superior, but that distributed coordination is physically necessary. Markets happen to be the most general, robust, and information-efficient distributed mechanism we know. If someone can name a non-market distributed system that scales better, they’re not refuting the argument; they’re offering another instantiation of it.
Every sophisticated objection to this argument reduces to: “You can avoid the constraint by not doing centralized coordination.” That’s not a rebuttal. That’s confirmation. If the only way to make “central planning” work is to distribute the coordination, decompose the constraints, introduce local autonomy, and tolerate delays—then what you’ve built is a distributed system, regardless of what you call it politically.
VII. What This Establishes
The idealized notion of central planning—centralized, real-time feasibility enforcement over a large, fast-changing economy—is architecturally non-scalable. Any system that functions at that scale must relax at least one of those requirements (centralization, real-time enforcement, tight coupling), and once it does, it has become a distributed coordination system in thermodynamic and computational terms.
Historical “central planning” systems functioned only insofar as they deviated from the ideal by introducing hierarchy, buffering, delay, local discretion, and informal markets. The classical “central planning” ideal collapses, under scale, into something it explicitly denies—distributed adaptation.
The core contrast is not “state” vs. “market,” but bottlenecked vs. distributed coordination. Markets are one instantiation of distributed coordination; bureaucratic hierarchies are another, typically worse one. Centralized real-time economic control is not a viable architectural primitive at population scale, and any system claiming to implement it is either misstating what it does or relying on hidden distributed mechanisms.