The Closed Circuit

1. The Receding Target

An entrepreneur is staring at the dashboard at three in the morning. The revenue graph is climbing steadily, but net profit is melting into statistical noise. The volume of work has tripled, the headcount has swollen, the number of operating hypotheses has passed critical mass — yet the end goal, financial independence, resilience, autonomy, has moved even further away than it stood at the start.

The modern subject is objectively busy. They work at the limit of their cognitive and physical capacity, systematically optimize processes, and adopt the newest tools. Yet the sense of approaching the result keeps being replaced by the acceleration of the run. This is not personal foolishness, not an individual burnout-prone personality type, not a lack of discipline. It is the physical experience of landing inside a system whose circuit is closed — and deeply asymmetric.

For twenty years the environment has been stubbornly washing out external friction — and the freed-up space was immediately occupied by the system. It built a race in which participation is paid for while the result belongs to the organizer. The mechanics are simple: the circuit is closed but not symmetric — for some it spins up rent, for others it burns resource.

This article will not offer a way out. It will offer a map of the trap — a precise optics for seeing how the race is built, who monetizes participation in it, and why the speeding-up of the treadmill is not a bug but a business model.

This race has three positions. From here on they run through the entire text — as a frame and as a diagnostic map.

Layer	Who this is	What they do	What they get
Actor	A subject with goal-setting	Keeps a direct link to reality, uses algorithms as an exoskeleton	Autonomy, managed risk, adaptability
Object	The mass participant of the race	Copies proxy metrics, optimizes dashboards, burns resources	A faster treadmill, margin zeroed out (P = MC)
Observer	The platform, infrastructure capital	Sells tickets, equipment, capital, and “images” of success	Guaranteed rent for every turn of effort

---

2. The “Snapshot” and the “Image”

The mass market participant never copies real success. They copy a local statistical outlier, packaged and sold to them as a mandatory standard. Inside this substitution of concepts lies the primary cause of the overproduction of participants in any race.

Two concepts must be separated:

• A “snapshot” is a single local slice. A screenshot of revenue in an ad account, a report on a shipped batch of goods, a specific working workshop, a traffic funnel that happened to fire. It is a raw, isolated fact that contains no information about the cost of capital, net margin, returns, or risk.
• An “image” is an aggregated outlier, stripped of context, passed through a marketing filter, and returned to the audience as the normative standard of entrepreneurial existence.

The mechanics of this process were described by René Girard in the theory of mimetic desire: a person desires not the object itself but the object-as-the-model-desires-it [1]. The revenue screenshot becomes a “model” at the moment it is packaged into a teaching playbook.

The scheme reproduces itself with the precision of a stamp. One successful Amazon store shows an abnormal return in a short window of opportunity. The creator of the “snapshot” scales not the business itself but the playbook — “do as I did.” Ten thousand copyists enter the same market simultaneously, with absolutely identical tools. The market saturates instantly, customer acquisition cost (CAC) soars, the price of the product collapses. As a result, 9,500 participants close at a loss, the remaining 500 operate on the edge of break-even, and the seller of the “image” has booked a guaranteed margin.

What looks like entrepreneurial ambition is, in structure, an imitation of a desire the imitator never actually had.

The desire has been copied. But what exactly does the crowd copy — a concrete action, or a metric of success? This is where the circuit breaks for the first time.

---

3. The Open Loop and Goodhart’s Law

The ground-level actor does not lose for lack of intelligence. They lose because the original market signal inside their control system has been replaced by a proxy metric.

In the normal economic model described by Friedrich Hayek, the free price and net profit serve as an undistorted signal [2]. They accumulate “dispersed knowledge” about all the physical constraints of the environment: from real demand to the cost of logistics and the wear of equipment. The signal directly ties the operator’s actions to reality.

The substitution happens at the moment Goodhart’s law switches on: as soon as a metric becomes a target, it ceases to be a good metric [3]. In the digital and platform economy, the roles of such proxy targets are played by ROAS (return on ad spend), the number of clicks, subscribers, gross revenue, or the virtual indicator MAU.

Optimization of the numbers inside the dashboard proceeds simultaneously with the slow death of the real enterprise:

• The dropshipper optimizes the ad account for a ROAS of 4.0, not noticing how hidden payment-processor fees, cash-flow gaps, logistics leverage, and returns of defective goods eat the entire actual margin.
• The founder of a SaaS service reports explosive MAU growth to investors, ignoring the fact that churn makes the cost of acquisition permanently unpayable.

Let us take churn slowly — on it, the substitution is visible as if under a microscope. A subscriber brought in by paid traffic costs more than they bring in over the first months: the economics only close on retention. But retention is a slow variable. It shows up a quarter later, when the money has been spent and the report has been sent. In the dashboard — growth. In the till — a hole with a lag. The founder optimizes what they see: not the business — its reflection on the screen.

The dashboard looks healthy exactly until the moment the bank account stops being healthy. The circuit is closed — but not for the runner: the feedback is open precisely at the point where the Observer collects rent.

In the digital world the substitution of the signal happens instantly. In the real sector it comes with a delay — and therefore seems safer. That is an illusion.

---

4. The Three-Tier Race: P = MC and Forced Disposal

In the real sector, the economic mechanics of signal substitution work harsher than in the digital one: the signal lags, and the gap between the decision and the reckoning grows with the weight of the capital. This mechanics has a single finale — the fall of the price to the level of marginal cost (P = MC).

The race unfolds across three consecutive levels:

1. The digital layer (fast cycle): dropshipping, Amazon FBA, traffic arbitrage. The low entry threshold leads to an instant overload of the auctions. Official statistics supply the mortality frame: according to the BLS, only about half of new American businesses reach their fifth year [4] — and the digital layer lives by the same rules, only faster and harsher. The rising cost of a click eats the margin before the accountants have had time to compute it.
2. The traditional CapEx sector (medium cycle): the entrepreneur sees someone else’s working production shop or logistics hub (a “snapshot”). Procuring equipment, installation, and launch take from 6 to 18 months. During this pause another two hundred operators, blinded by the same signal, make an analogous decision. By the time the capacity enters operation, supply exceeds demand, and the price packs itself into the formula P = MC — price equals marginal cost [5].
3. The AI infrastructure race of 2024–2026 (macro cycle): thousands of companies — from hyperscalers to shell startups — sent hundreds of billions of dollars into buying GPUs and reserving compute capacity, counting on a linear payback from token sales [17]. The CapEx lag came to 12 to 24 months. By the time the servers came online en masse at full capacity, the overproduction of compute crashed the price of rent: the hourly price of a flagship accelerator fell from $7–10 to $2–4 [17], reproducing the classic P = MC trap at the scale of an entire industry.

When the margin zeros out (P = MC), the runner’s only way to survive is to forcibly accelerate turnover. To sustain gross income on a falling rate of profit, the system demands a continuous acceleration of consumption. A product can no longer be made “to last”: its life cycle is artificially shortened, landing squarely in forced obsolescence and rapid disposal.

This race for volume has an inevitable price — systemic averaging and quality degradation. A high-quality, durable thing becomes direct economic vandalism for the closed circuit. As a result, the system slides into mindless, predatory exploitation of productive forces and physical resources. The planet’s reserves of raw materials, energy, and human labor are burned not to create new value, but to sustain the very rhythm of the endless restart of sales.

In the AI infrastructure case, this predatory exploitation instantly ran into a hard physical limit — energy. Unlike software metrics, data centers require gigawatts: the share of American electricity going to data centers grows in a single cycle from 4.4% to 6.7–12%, by the estimates of the national laboratory LBNL [17]. Grid connections are handed out in years — the queue for bringing in power is longer than any startup cycle. Water for cooling is already the subject of municipal hearings. Physics proved uncompromising toward proxy metrics: a gigawatt cannot be printed as a report.

Meanwhile the Observers — equipment suppliers, power-generating companies, landowners, and data-center operators — extract guaranteed rent from every attempt by the runners, regardless of the finale of any particular startup. Ground-level operators make decisions based on a distorted signal; their resources are burned not because they are stupid, but because the environment is designed so that this happens.

The Observer has an upper floor. BlackRock manages more than $15 trillion in assets against equity of about $56 billion — a gap of roughly 270:1 [19]. Blackstone has $1.35 trillion against $8.7 billion of its own: 155:1. This is not a player’s leverage: an Actor stakes their own and can die; the Observer stakes nothing — the responsibility is declared by them, the risk is carried by other people’s money, and their platform Aladdin keeps the score of another ~$21 trillion of other people’s assets [19]. In the top lines of the Observers’ own registries, the same few names recur: the three largest managers are the top shareholder in roughly nine out of ten S&P 500 companies [20]; one of them was even born inside the other — they went separate ways in ninety-four [20]. Both are already standing at the new floor as well: Blackstone bought the data-center operator QTS, and a consortium led by BlackRock bought Aligned Data Centers — $40 billion for more than six gigawatts [21]. The Observers try to loop everything back through themselves — but the sandpile does not obey the one who recounts it; all one can do is add grains to it, one grain at a time.

By the time the first balance sheet shows a loss, the next cohort of runners has already signed the lease.

The error seems unique to the novice. But the same error reproduces at a level where no one should be allowed to make it.

---

5. The Bridge: Model vs. Reality

The error of the beginning marketplace seller and the error of managers with world-famous names are identical in nature. It is the blind belief that a mathematical model takes priority over physical reality.

“Optimizing for the model” is the same operation at every level of complexity. Here George Soros’s principle of reflexivity comes into force: the mass replication of a successful model changes the very conditions that made the model successful [6]. A thousand identical playbooks destroy the environment of their own application.

Level of complexity	Action	Change in the environment	Final result
Ground-level operator	Optimizes ROAS by the playbook	The ad platform’s auction becomes oversaturated	CAC rises, the strategy turns unprofitable
Quant fund	Optimizes the historical correlation of assets	The strategy turns into a crowded trade	Correlation breakdown, the strategy is liquidated instantly

The model works only as long as a limited number of participants use it. It stops working precisely at the moment it becomes popular. The success of the recipe is its own coup de grâce.

If the bridge between the dropshipper and the laureate has been crossed, it remains to check how the same logic looks at the highest level of the elite — where the error costs billions.

---

6. Elite Blindness: LTCM, 1998

Even the most credentialed market participants possess a fatal blindness to catastrophes, because their own previous success trained this blindness into them, systematically.

In 1998 the fund Long-Term Capital Management (LTCM), whose board included the Nobel laureates Myron Scholes and Robert Merton — the 1997 prize [7], — had pushed its leverage to a ratio of 167:1 by September 1998 [8]. Against equity of roughly $600 million, the fund held positions in assets above $100 billion. The fund’s models priced risk by historical volatilities — simultaneous liquidation was considered negligible. One line on the invoice: the face of the case was John Meriwether; the notional of open derivative positions exceeded $1 trillion [8]. When the construction began to come apart, a rescue of $3.6 billion was assembled by 14 banks under the aegis of the New York Fed [9].

As Nassim Taleb pointed out, the model did not account, in principle, for the “fat tails” of the distribution [10]. The Russian default of August 1998 became the very event that instantly destroyed the construction. Thorstein Veblen called this the phenomenon of “trained incapacity” [11]: the academic triumph of the authors had atrophied their ability to entertain the thought that their own massive presence on the market had completely destroyed the statistical correlations of previous years.

In terms of Per Bak’s sandpile model, the system had been accumulating criticality for years [12]. The particular trigger of the collapse could have been anything, but the very class of catastrophic events was mathematically inevitable.

The difference from the upper floor is fundamental. For the Observers at the top, the gap is 270:1, and no crash threatens them: their leverage is of a different kind — it stands on the fee-stream from other people’s assets, the risk is locked into the funds’ clients, such a player cannot be liquidated. LTCM, by contrast, staked real balance-sheet debt — and believed its mathematics stronger than market friction.

The same mechanics keeps reproducing itself decades later [18]:

• 2021: the collapse of Archegos Capital (Total Return Swaps, hidden leverage).
• 2024: the unwinding of the Yen Carry Trade (mass copying of the yen strategy).
• 2024–2026: the AI infrastructure boom (boards of directors buy GPUs en masse on linear payback expectations, which leads to a crash in the price of rent).

To reproduce this error, Nobel laureates are no longer needed — the resolutions of standard boards of directors are enough.

The model did not break because of a defect in the mathematics. It broke because the people who built it lost the ability to imagine a world in which that mathematics does not work.

LTCM is a separate failure. But there have already been enough such failures in history to see in them not an anomaly but a rhythm.

---

7. Sand, Spotlight, a New Round

The crash is not a malfunction of the system — it is a structurally necessary phase of it.

According to Carlota Perez’s theory of technological cycles, every wave passes through inevitable stages [13]:

1. Installation frenzy (Frenzy) — a mass inflow of capital and of runners building infrastructure.
2. Financial crash (Crash) — the inevitable reckoning for overproduction and false signals.
3. Deployment phase (Deployment) — consolidation.

Perez reads the entire industrial history of the modern era through this rhythm, and every time the finale looks the same: the infrastructure outlives its builders [13]. In the 1840s Britain went through the cycle literally. Parliament was stamping acts for new trunk lines, capital flowed into railway shares, companies built parallel lines in a race against one another. The mid-decade crash buried dozens of companies — but the tracks laid down became the backbone of the transport network the country still uses today [14]. A century and a half later the scenario repeated itself with fiber optics: the fever of internet cabling at the end of the nineties, the crash of 2000 — NASDAQ lost about 78% by the autumn of 2002 [15] — and the dark fiber that the survivors bought up at residual price and that later became the physical layer of the cloud era.

Wave	Frenzy	Crash	Consolidation	Who buys up
Railways, Britain of the 1840s	Acts for trunk lines, parallel routes, capital in the builders’ shares	The banking crisis of 1847, the collapse of railway shares	The surviving networks buy up lines at the price of rails	The incumbent magnates and the merged companies
Dot-coms, 1995–2002	IPOs without profit, the laying of trunk fiber	NASDAQ loses about 78% by October 2002	Trunk fiber goes to the survivors at residual price	The telecom giants and the future cloud platforms
AI infrastructure, 2024–2026	GPU over-buying, gigawatt reservations	The crash of compute rental prices (P = MC)	Absorption of capacity into corporations and the state sector	Observers (Blackstone, BlackRock, the hyperscalers)

The mechanics of cleansing are cruel: almost all the Objects burn in the collapse phase. The surviving Actors and Observers buy up the devalued capacity — the railway tracks of the nineteenth century, the fiber optics of the 2000s, or the data centers and GPU servers of 2026 — for cents on the dollar. The infrastructure of the new round always reaches the economy on the bones of the participants of the previous cycle.

Survivorship bias serves as the permanent marketing engine of this process. The system hides the perished Objects in the sand while spotlighting the single winner. The survivor immediately turns into a new “snapshot,” restarting the mimetic cycle for the next cohort.

Here macroeconomics closes with micro-psychology. Carlota Perez’s cycle explains why the system needs periodic crashes. The hedonic treadmill (the concept of Brickman and Campbell) [16] explains why a particular person keeps running even after a local collapse: at the individual level, the norm of success shifts faster than the runner manages to register and appropriate the result.

The system does not break. It metabolizes.

Metabolism explains why crashes do not kill the system. The diagnosis goes deeper: the running itself leads nowhere.

---

8. Brownian Chaos and the Physics of Horizontal Order

The system sells the run to its participants as meritocratic competition, where every step brings the goal closer. On the cross-section, Brownian diffusion is visible: particles collide endlessly, burn energy, and produce noise — but the net displacement vector equals zero. The winner is selected by survivorship bias: the particle that survives is the one the chaotic impulse happened to carry slightly farther than the rest.

The prize of this race is built so that it is impossible for everyone to win it. Fred Hirsch showed [22]: the object of the struggle is positional goods — status, rank, the “image of success.” They cannot grow for everyone at once: the rise of one position devalues another, and the race folds into a zero-sum game. The patent wars are a vivid example: in 2014 Elon Musk opened Tesla’s patents, and later called patents “for the weak” [23] — mines that merely slow the common flow in order to defend a paper territory.

Since the price signal is broken by Goodhart’s law, the question arises: who sets the goals if the metrics lie? A centralized Gosplan is not viable for lack of dispersed knowledge (Hayek) [2]. The natural alternative is polycentric goal-setting: defining tasks through autonomous, overlapping groups with their own feedback (Elinor Ostrom) [24]. Digital protocols have for the first time made such horizontal coordination technically possible without loss of complexity. Structurally, this is the long-standing intuition of the anarchist tradition of mutual aid — order without a supreme center — now confirmed empirically. But anarchist polycentrism is not a panacea: Mancur Olson warned that without strict internal rules any horizontal group slides into free-riding (free-rider) and the same struggle for status [25].

Here the entire architecture of the article folds into a single plain criterion: the circuit belongs to whoever controls the loss function (loss function).

A runner optimizing ROAS and MAU is computing the gradient of someone else’s loss function, having taken it for their own vector. An autonomous group survives exactly as long as it formulates its own loss function (cash-flow gaps, real margin, physical turnover). A tool remains an exoskeleton as long as it optimizes your goals, and becomes a prosthesis when it relays external ones.

From this follows the main conclusion about the structure of the modern economy: vertical rent is inevitable where scale, capital, and heavy infrastructure are required. But the anarchist/decentralized assembly of society fundamentally changes the way this rent is distributed.

The architecture of modern transformers explains this mechanism vividly. Base training of AI requires billions in investment and inevitably creates vertical rent at the infrastructure level. Yet the very process of decentralized generation of answers to millions of local requests distributes computational benefit across the entire horizontal network. When a network keeps control over its own loss function and rests on the principles of horizontal coordination, vertical rent ceases to be a mechanism for extracting resources from runners and becomes the infrastructural foundation of shared access.

If the circuit is closed and self-sustaining, is the exit an illusion? Not quite. But the path to subjecthood lies not through an “exit” — it lies through changing one’s position inside the system.

---

9. A Diagnostic Map Instead of a Recipe

There is no universal mass recipe for “escaping the system.” Any attempt to standardize an instruction for getting out is instantly packaged by the Observers into a new playbook and sold to the runners. The system has already done this with meditation, minimalism, the concepts of slow living and downshifting.

The only alternative to the position of the Object is the subject position of the Actor who keeps a direct link to reality. It rests on three practical strategies:

1. Pressure as training, not as punishment. The understanding that a complete absence of external pressure inevitably leads to the degradation and atrophy of the mind. External pressure (hard deadlines, scarcity of resources, competition) is perceived not as an emotional enemy but as an athletic training machine. The function of the race changes: from an external end in itself it becomes an internal instrument of tempering.
2. Technologies as an exoskeleton, not as a prosthesis. Any tool (including artificial intelligence and automation) can work in two modes. If the tool makes decisions and thinks for the person, it is a prosthesis leading to thinking by default. If the tool extends and amplifies the person’s goal-setting, it is an exoskeleton. The worst interface of our time is the polished mirror: it returns to the person their own opinion, neatly polished and without a single scratch of resistance. But the same interface has a second mode: it remains a mirror as long as the person demands convenience, and becomes an exoskeleton precisely when the subject keeps goal-setting for themselves.
3. A complex web of ties instead of algorithmic atomization. The only defense against a distorted proxy signal is horizontal calibration. This is a closed cohort of 5–7 people with whom one regularly discusses not pitch decks and marketing proxies but real physical numbers, cash-flow gaps, and costs. The combination of horizontal trust and vertical access to context creates a filter impenetrable to mass illusions.

None of the three strategies takes one out of the circuit — all three change the position inside it. The criterion for each is the same: ownership of one’s own loss function. The table from the beginning of the article now reads differently: the layers are not swapped by an act of will, but the precision of calibration inside a layer is a question of discipline, not luck.

The map of the trap is not the way out of the trap. But it is the only thing that returns the runner to themselves.

---

Sources

1. Girard R. Deceit, Desire and the Novel. Baltimore: Johns Hopkins University Press, 1961.
2. Hayek F. A. The Use of Knowledge in Society // American Economic Review, September 1945.
3. Goodhart C. A. E. Problems of Monetary Management: The UK Experience. Papers in Monetary Economics, Reserve Bank of Australia, 1975. The commonly quoted phrasing — Strathern M. “Improving ratings”: Audit in the British University System // European Review, 1997.
4. U.S. Bureau of Labor Statistics. Business Employment Dynamics: Establishment Age and Survival Data — long-run cohort averages.
5. Varian H. R. Intermediate Microeconomics: A Modern Approach. New York: W. W. Norton — the chapter on perfect competition.
6. Soros G. The Alchemy of Finance. New York: Wiley, 1987.
7. NobelPrize.org. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1997.
8. Lowenstein R. When Genius Failed: The Rise and Fall of Long-Term Capital Management. New York: Random House, 2000.
9. Federal Reserve Bank of New York, statement on the consortium, September 23, 1998; New York Times archive.
10. Taleb N. N. The Black Swan: The Impact of the Highly Improbable. New York: Random House, 2007.
11. Veblen T. The Instinct of Workmanship. New York: Macmillan, 1914; carried into rhetoric by Burke K. Permanence and Change, 1935.
12. Bak P., Tang C., Wiesenfeld K. Self-Organized Criticality // Physical Review Letters, 1987.
13. Perez C. Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages. Cheltenham: Edward Elgar, 2002.
14. Chronology of the British railway mania of the 1840s (Railway Mania); economic-history surveys; see also Perez C., 2002.
15. Market data: NASDAQ Composite, 2000–2002.
16. Brickman P., Campbell D. T. Hedonic Relativism and Planning the Good Society // Appley M. H. (ed.). Adaptation-Level Theory. New York: Academic Press, 1971.
17. Capital expenditure reports of the hyperscalers (Microsoft, Alphabet, Meta, Amazon), 2024–2026; Silicon Data — GPU rental indices, 2023–2025; IEA. Energy and AI, 2025; LBNL / U.S. DOE. 2024 United States Data Center Energy Usage Report, December 2024.
18. Market chronology 2021–2026: the collapse of Archegos Capital, March 2021 (NYT, FT archives); the unwinding of the Yen Carry Trade, August 2024 (Bank of Japan); Silicon Valley Bank — FDIC, 2023.
19. BlackRock — Q2 2026 earnings press release; BlackRock 10-K for 2025 (balance-sheet equity); Blackstone — Q2 2026 earnings press release; industry reviews of the Aladdin platform, 2025.
20. Azar J., Schmalz M., Tecu I. Common Ownership and Competition in Product Markets // Journal of Financial Economics, 2018; aggregated 13F registries, 2026; CNBC, June 22, 2017 — the genealogy of BlackRock and Blackstone.
21. Blackstone — press release on the acquisition of QTS, 2021; The Wall Street Journal — the consortium deal for Aligned Data Centers, 2026.
22. Hirsch F. Social Limits to Growth. Cambridge, MA: Harvard University Press, 1976.
23. Tesla Motors. All Our Patent Are Belong To You — blog post, June 12, 2014; Business Insider — Elon Musk’s CNBC interview about SpaceX, September 2022.
24. Ostrom E. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge: Cambridge University Press, 1990.
25. Olson M. The Logic of Collective Action: Public Goods and the Theory of Groups. Cambridge, MA: Harvard University Press, 1965.