A regulatory action read as failure was a stone placed to preserve every option.
§1
Manus was Chinese-founded—its core personnel and technology traceable to China. By mid-2025 it had relocated key operations to Singapore. In December 2025, Meta announced a multibillion-dollar acquisition. The company was preparing to disappear into an American hyperscaler.
China’s National Development and Reform Commission, acting through the Office of the Working Mechanism for Security Review of Foreign Investment, issued a prohibition. The deal was to be unwound. A senior researcher at the Ministry of Commerce’s institute followed with a public commentary. Anyone attempting to evade regulation through corporate transmigration, he said, would in the end fail to achieve his purpose.
Western analysts read the event in a familiar genre. A regulatory action with imperfect extraterritorial reach. The questions followed: was the prohibition credible, did Beijing have the legal arm, would Singapore comply, would Meta walk. Both pessimistic and optimistic readings shared a single assumption. The prohibition’s purpose was to stop the deal. Its success or failure was to be measured against that outcome.
The assumption is wrong.
§2
Western analysts brought specific tools. Roman law on jurisdiction. Common law on enforceability. Westphalian theory on sovereign reach. The administrative-law doctrine of extraterritoriality. Each of these tools rests on a single foundational claim about what regulation is for. Regulation exists to produce a determinate outcome in the matter at hand. If it cannot produce that outcome, it has failed.
The claim holds where the tools were forged. It does not hold here.
The system that issued the prohibition is not a less developed version of the systems Western analysts were trained in. It is a different system. Different objectives, different mechanisms, different criteria for success. To see what the prohibition actually did—rather than what it would have done had it issued from Brussels or Washington—the system itself must be described.
It has three layers. The clearest single image, drawn from the system’s own cultural inheritance and not from any imported governance vocabulary, is the gu jar.
§3
In premodern Chinese poison craft, the gu jar (蛊罐) was a sealed vessel. The cultivator placed inside it a great number of venomous creatures—scorpions, centipedes, vipers, spiders. The vessel was closed. Time passed. Denied food, the creatures consumed one another. After many cycles of predation, a single survivor remained. The survivor had absorbed the venoms of every creature it consumed. It was the gu (蛊): the most virulent product of a process the cultivator did not directly intervene in.
The cultivator acted at two points. The selection of inputs at the start. The extraction of the survivor at the end. In between he fed, sealed, waited. The active practice was gu cultivation (养蛊).
The image is a model, not an essence. It claims nothing about Chinese culture. It claims something about a governance structure, and it names what the standard vocabulary drops. Industrial policy loses the predation. State capitalism loses the cultivator’s deliberate non-intervention in the central process. Internal competition loses the sealed environment and the deliberate selection of who is allowed to emerge. Each available substitute drops a load-bearing element.
Henceforward in this essay: gu, gu jar, gu cultivation. Technical terms. Carry them forward intact.
§4 — Layer One: the basket
The first layer is legislative.
In the Anglo-American tradition, law is accretive. Specific harms produce specific cases. Cases produce precedent. Precedent, over time, produces doctrine. Statutes follow the same logic at a faster tempo. A problem appears, a statute responds. The system is patch-based. Slow, conservative, bound to realities already encountered. It cannot reach forward into realities not yet manifested.
The Chinese legislative tradition operates on the inverse principle.
It is anticipatory. The legislator does not wait for cases. The legislator imagines the entire space of possible cases and constructs a framework wide enough to contain them all. The form is older than any contemporary regime. The Qin codes. The Tang Code. The Great Qing Legal Code. Each shares the same structure. Modern Chinese administrative law inherits it. This is not an immature gesture toward Western legality. It is a different theory of what law is for.
The functional consequence is severe and persistently misread.
Under a basket-style framework, almost any actor in almost any commercial activity can, at some moment, be discovered to be in violation of some provision. This is not a flaw. This is the system’s central feature.
The function of basket-style law is not to constrain behavior. The function is to preserve the dispositive authority of the executor. The framework’s breadth ensures that grounds for action against any chosen actor always exist. Whether action is taken, when, against whom, to what extent—these are not determined by the law. They are determined by the executor, on his judgment, at his chosen moment.
This is the legal foundation. Nothing that follows operates without it.
§5 — Layer Two: the closed loop
The second layer is executive. Its question is operational. Given that legal grounds exist for action against nearly any actor, how is the actual machinery made cheap enough to operate at scale?
The Western answer, rendered in Foucault, is the panopticon. A surveillance architecture in which the observed are aware that observation is possible at any moment, and discipline themselves accordingly. The solution is partial. It still requires the architecture. It still requires that the possibility of observation remain visible. It still depends on the disciplined subject’s belief in the apparatus.
The Chinese system operates one layer deeper. It does not require continuous architecture. It runs on a closed loop with four moments.
Exemplary selection. A specific case is selected for visibility and demonstrative power, not for case-specific severity. The actual subject is not the named entity. The actual subject is every other entity in the named entity’s class.
Material dispossession, with collateral reach. The selected target is not criticized. The target is stripped. Ant Group: the IPO halted hours before pricing, the founder pushed into prolonged silence, the corporate structure restructured under regulatory pressure. Didi: pressured into delisting from New York, subjected to a cybersecurity review and app-store removal, and pushed toward state-supervised handling of sensitive data. The for-profit tutoring sector: a hundred-billion-dollar industry abruptly restructured by directive, listed providers stripped of most of their market value, hiring demand collapsed across the sector. The dispossession is real. It is visible. It is rarely reversed. And it does not stop at the named entity. Family members, business partners, downstream suppliers, employees—each is exposed. The radius of pain is the mechanism’s defining feature. Lever-points held against the actor are activated together, not sequentially.
Generalized deterrence. The class of similarly situated actors observes the dispossession. The observation is unmediated. There is no need for official messaging. The exemplary case carries its own meaning. Every actor in the class performs the cost-benefit calculation in advance and arrives at the same answer. The next prohibition need not be issued. The behavior pre-adjusts.
Self-adaptation. This is the loop’s closure. The actor, having observed the cost of non-compliance, restructures his own conduct, his own assets, his own family arrangements, his own cross-border exposure—in advance of any direct interaction with the executor. The disciplinary work is outsourced from the executor to the actor himself. The system runs cold. It does not require continuous force. It does not require an ideological apparatus that every subject must affirm. It requires only the periodic, precise demonstration that the levers exist and that the executor will, at chosen moments, pull them.
This is more economical than any disciplinary architecture in the Western record. The energy cost is what permits the system to scale across a population of one and a half billion.
§6 — Layer Three: gu cultivation
The first two layers are control mechanisms. By themselves they produce nothing. They do not generate Huawei. They do not generate CATL. They do not generate the pipeline of national champions through which Beijing now contests the technological leadership of the United States.
The third layer produces. Its mechanism is gu cultivation.
The cultivator opens a domain—electric vehicles, batteries, drones, large-scale AI, semiconductors, commercial space—and admits a large number of competitors. The admission is uneven. Some receive more nourishment than others: state-media attention, sovereign-fund stakes, local-government equity, preferential listing channels, talent-program flows, government procurement. The cultivator does not pre-select a winner. He selects a population of plausible survivors, gives each enough resources to persist for the duration of the cycle, and permits them to engage one another.
The environment is sealed. This is non-negotiable. Capital cannot freely exit. Senior personnel cannot freely emigrate. Core technologies cannot be transferred to foreign acquirers. Data cannot cross borders. The Manus prohibition belongs to this layer. It is not a transaction-specific act. It is the periodic adjustment of the seal on the artificial-intelligence cultivation jar. Without periodic adjustment, the seal degrades. With it, the seal holds long enough for the cycle to complete.
Within the sealed jar, the competitors devour one another. The strongest acquire market share, talent, capital, and relational depth at the expense of the weaker. The cultivator does not interfere in the predation. He watches.
After a period—five, ten, fifteen years, depending on the domain—the cycle approaches resolution. A small number of survivors emerge. They are not the entities the cultivator predicted at the cycle’s opening. They are the entities the process selected. The cultivator extracts. The survivors are recognized as national champions, integrated into the strategic apparatus, and permitted to operate outside the jar—on the international stage—carrying the venom they absorbed from everything they consumed in their formative environment.
Huawei is the proof of concept. The new-energy-vehicle and battery sectors are the demonstration that the model replicates. Artificial intelligence is the cultivation cycle now in progress.
What Western analysts have read for two decades as Chinese industrial-policy inefficiency—the persistence of dozens of competing firms, the failure to consolidate, the apparently uncoordinated investment, the survival of zombie entities, the parallel pursuit of mutually substituting technological pathways—is none of these things. It is gu cultivation, observed by observers whose vocabulary cannot contain it.
§7 — the system as one thing
The three layers are not independent. Each requires the others.
Without basket-style law, executive action has no grounds. The lever-points cannot be activated under any color of legality. The outsourcing of self-policing collapses.
Without lever-point retention and exemplary instruction, basket-style law is paper. The executive becomes too expensive to operate at scale. Subjects do not self-regulate. The cultivation jar leaks at every seam.
Without gu cultivation, the first two layers control without producing. The state acquires the apparatus of authority without acquiring the productive capacity that authority is meant to project. The shell calcifies. This was the Soviet failure mode.
The Chinese system avoids that failure mode by stacking the three layers into a single integrated operating system. Layer One supplies the legal medium in which Layer Two operates. Layer Two supplies the disciplinary economy in which Layer Three is contained. Layer Three supplies the productive force that justifies the existence of the first two and provides their material base.
The integration is the achievement. There is no precedent for it in Western political-economy. The closest historical antecedents—Legalist statecraft in the Warring States and Qin periods, the Tang and Ming administrative architectures, the late-Qing self-strengthening movement—each contain elements of the structure but never the integrated form. The integrated form is recent. It has been refined over the four decades since the opening of the reform era. It is the principal product of those four decades.
§8 — what the system cannot be reduced to
A description of this kind invites two characteristic misreadings. Both must be refused.
The first treats the system as a developmental stage. China is in transition, the reading goes. The basket framework will narrow into rule-of-law in time. The executive layer will mature into proceduralism. The cultivation layer will dissolve into a normal market. The reading is wishful. There is no observable mechanism by which the transition would occur. The system as integrated produces outcomes its operators have no incentive to surrender. A system that has produced Huawei is not in a hurry to become one that cannot.
The second treats the system as a totalitarian apparatus, indistinguishable from earlier authoritarian forms. This is also wrong. Soviet planning lacked Layer Three. Maoist mobilization lacked it as well. The integrated gu system is more economical, more productive, and more durable than either. To call it totalitarianism is to apply a Cold-War term to a post-Cold-War structure that the term was not designed to name.
The system has its own pathologies. They will produce their own crises in their own time.
§9 — how the system moves
A final difficulty remains. The three-layer architecture describes what the system is. It does not describe how it moves. Western analysts, even when they grasp the structure, persist in misreading specific actions.
The reason is simple. The Western analytic frame treats a regulatory action as a single move with a single determinate purpose. If the move cannot be shown to achieve that purpose, it has failed.
The Chinese frame does not assume single purpose. A familiar Chinese analogy: in go, a single stone can serve as territorial claim, as influence on distant fighting, as material reserved for a future exchange, as a probe, as a forcing move, or as a sacrifice. The same stone, in the same position, can carry several of these at once. Which function becomes the operative one is determined later, by how the opponent responds. The player places the stone to preserve options. The game reveals what the move was for.
This is the cognitive habit—not the metaphor, the habit—inside which Chinese strategic action is conducted. Reserving multiple latent purposes in a single act is not a stylistic preference. It is the default mode.
The Manus prohibition, read this way, decomposes into several simultaneous moves.
It is a probe. The reactions of Meta, of Singapore, of Washington, of the wider Chinese AI industry are each absorbed and integrated. The next move depends on what the probe reveals.
It is reserved material for the upcoming bilateral negotiation between Beijing and Washington. The prohibition can be sustained as a signal of resolve, or it can be withdrawn as a concession in exchange for movement on a separate file. Its value as a bargaining asset rises until the moment it is spent.
It is settled territory in the domestic field. The recognition of the central authority’s dispositive jurisdiction over Chinese AI cross-border transactions is now established. Unlike the international dimensions, this part of the move is irreversible regardless of subsequent developments.
It is influence on future cases. The next firm contemplating the same exit route faces a recalibrated risk function. The cultivator’s authority over the seal of the jar has been generalized.
It is a forcing move. It compels the counter-party to respond, while the player who issued the move is free to act elsewhere. While Meta’s lawyers work the question of whether to walk away, Beijing has its hands free for other dossiers.
A single act. Five simultaneous functions. Possibly more. Which of them will, in retrospect, prove to have been the move’s decisive function—the cultivator does not know yet. The stone was placed to preserve all of them. The unfolding situation will reveal which one mattered.
Western analysts, reading this through their single-purpose frame, repeatedly produce assessments that turn out, six months or two years later, to have missed the point. They are not failing technically. They are looking for the single determinate purpose of an action that was not designed to have one. The reservation of indeterminacy is itself the strategic asset.
The two frames are not in dialogue. They are operating on different assumptions about what a strategic action is.