The Sovereign Function

A metric stays honest only if whoever reports it also eats the loss when it's wrong. China ran that rule twice, in opposite directions, and got the same answer both times

On the morning of 3 November 2020, a banker working on the Ant Group IPO gave Al Jazeera the only line he had. A "curveball," he called it. Nobody had a better one. Forty-eight hours on, Ant was due to close the largest initial public offering in human history with over $37 billion raised and a valuation north of $300 billion, bigger than any bank on earth. CICC, the lead underwriter, had just jumped from fifth to second in the global league tables on the strength of this single deal, its best year since 2009.

Then the Shanghai Stock Exchange suspended the listing. The official reason was "changes in the fintech regulatory environment," a phrase with the peculiar quality of sounding bureaucratic and faintly amused at its own vagueness at the same time — the kind of sentence a state writes when it would rather not write the true one. CICC's shares fell 6.5 percent before lunch. The fees, the bragging rights, and roughly a decade of underwriting momentum evaporated in a single morning, and the world's largest IPO became, overnight, the world's most conspicuous non-event.

Everyone reached for the same explanation, the moment in late October when Jack Ma had stood in front of the country's most senior bankers and regulators and told them, in substance, that they were running a pawnshop dressed up as a financial system. There is no systemic risk in Chinese finance, he said, because there is barely a financial system to speak of, only the habits of one.

A sentence like that costs a man something before he has finished saying it. But temperament, on its own, is not a sufficient explanation, and the commentary that settled on Ma's hubris — a man who mistook his balance sheet for political cover and picked a spectacularly bad morning to say so — is true as far as it goes but stops far short of where the real story lies. Explaining the Ant episode through Jack Ma's mouth is like explaining a constitutional crisis through the mood of the judge who happened to trigger it: accurate about the spark, silent about the tinder.

II. What Ant Saw

Between 2017 and 2018 the anthropologist Li Dan spent time with young migrant workers at a factory on Shenzhen's outskirts. One of them, given the pseudonym Jiwei to protect his identity, was twenty-two, newly arrived from the countryside, and belonged to the exact category of person China's state banking system was never built to notice. He had no salary history any bank would recognise, no collateral, and no branch manager who'd vouch for him over tea.

To the institutions that ran Chinese finance, Jiwei did not really exist.

He set about fixing that himself, watching his Sesame Credit score the way a trader watches a ticker, and running small experiments to see what moved it: covering a colleague's lunch on his own account, topping up his parents' phone credit through Alipay instead of cash, routing a workplace reimbursement through Huabei rather than his wages. The score went up.

He kept doing it, manufacturing transaction-by-transaction, a credit history the traditional banking system had no drawer to file even if he'd walked in and asked. A second worker in Li Dan's fieldwork, Zhenghan, opened his first Alipay credit line at twenty-five with the goal of building a track record that would let him borrow for a business of his own.

Neither man was legible to the state. Both were legible to Ant. James Scott spent a career describing how states, governing from above, flatten the messy, local knowledge of lived circumstance into a grid simple enough for a distant office to administer. Jiwei was running the operation in reverse, building his own grid, from underneath, because it was the only one that would let him through the door.

By the time of Ant's IPO filing, this arrangement had scaled to an impressive size. Huabei and Jiebei together served roughly 500 million people in the year to June 2020 — about half of Alipay's entire user base — carrying a consumer lending balance of 1.7 trillion yuan, some $254 billion.

That balance amounted to roughly 21 percent of all short-term consumer loans issued by China's deposit-taking institutions. Of the $254 billion Ant had originated, only 2 percent sat on its own balance sheet. The rest was originated on Ant's data and distributed through banks and trust companies with no tool of their own for making that judgment. Most of the borrowers looked like Jiwei and Zhenghan and Ant could see them because it had built the only machine capable of it.

This was what Beijing brought to heel. The IPO was only the lever. What came under supervision was an institutional accumulation of knowledge — a slow-built ledger of who deserved credit and which economic actors were viable — generated entirely outside the architecture through which the Party had always governed access to capital. In the one domain Beijing has always treated as irreducibly political, a private company had begun, quietly and at staggering scale, performing a sovereign function.

Ant possessed knowledge without commensurate liability. The banks and trust companies underneath it possessed liability without knowledge. The obvious objection deserves to be made before the interesting one, or the omission will look like evasion. Huabei and Jiebei carried that $254 billion balance while Ant held capital against only 2 percent of it — taking a bank's underwriting upside at consumer-lending scale while banks and trust companies absorbed the downside.

The systemic-risk case for intervention was therefore obvious. Raise the co-funding ratio. Make Ant hold capital in proportion to what it originates. Fold the lending arm into a bank-style holding company under bank-style rules.

That is, roughly, what the restructuring did. It also folded Sesame Credit's scoring architecture into the same apparatus. A prudential regulator had every reason to inspect and validate Ant's models. It did not need, for that purpose alone, to absorb the architecture that made them useful. The intervention went beyond making Ant's lending safer. It brought the thing that made Ant's lending accurate inside the state's own field of authority.

III. The Windows in the Wall

Six years later, money leaves China through doors that barely existed when Ant's listing was stopped. Southbound flows through the Stock Connect scheme reached HK$1.4 trillion — about $179 billion — in 2025, up 74 percent on the year before. Average daily turnover through the same channel more than doubled, to HK$121.1 billion. Portfolio outflows for the year came to roughly $426 billion in total, much of it in funds whose job is being right about returns rather than obedient to instruction.

These are not grey-market flows. They're policy, and the policy permits the same species of independent financial judgment that made Ant politically dangerous.

The immediate reasons are ordinary enough. A weak renminbi, a Federal Reserve moving downward, and a revived Hong Kong IPO pipeline reward capital seeking returns outside a slowing domestic market. That explains the flows but not the state's forbearance.

Capital seeking yield outside a slowing domestic market and a currency under strain also shaped 2015 and 2016. Beijing tightened controls hard enough to burn through nearly a trillion dollars of reserves defending the currency. Today's flows travel largely through channels the state built for them. The difference lies in permission.

The scale involved has stopped being a rounding error. China's current-account surplus reached $735 billion in 2025, more than twice Germany's roughly $294 billion in 2018, and stood roughly 73 percent above China's own figure a year earlier. Twenty years ago, money on this scale would have landed directly on the People's Bank of China's balance sheet, managed inside institutions the state fully controlled.

Today a growing share of it travels through Hong Kong into portfolios chosen for expected return rather than official instruction. The surplus remains Chinese, but the authority to decide what it funds is steadily slipping outside Beijing's hands.

An emerging policy language gives this movement a larger purpose. Shi Zhan, a professor at Shanghai International Studies University, has argued that China's next step in globalization will move the 'front end' of production abroad while keeping the dense supply-chain 'middle platform' at home. Renminbi appreciation would accelerate outward investment, overseas factories would keep buying Chinese intermediate goods, and the result would be a geographically dispersed but still Sinocentric production system, which he calls a 'shadow China.'

The sequence is ambitious and it's certainly not a settled policy blueprint. Its value, however, lies in making clear that outward capital need not represent liberalisation in the older sense. The same channels can finance a production system extending beyond China's borders while preserving its role as an industrial hub.

That too is a sovereign function, delegated because the state needs what independent actors can learn and bounded because it knows what such learning can become. The boundary is the point. From a distance, outward investment, Stock Connect, and Chinese companies bargaining with foreign governments can look like liberalisation. Up close, they amount to a managed distribution of judgment inside a political perimeter Beijing still controls.

Capital controls remain extensive, the renminbi remains managed, and state banks remain dominant. Regulators intervene as soon as outflows begin to resemble flight. The walls are not coming down. Beijing has concluded that the people inside the citadel can no longer see well enough on their own and has started cracking a few windows, hoping the light gets in without anyone having to move the furniture.

Ant's knowledge came first and its political weight followed from it; Stock Connect and the outward industrial system start from the opposite end, with Beijing fixing the perimeter before it lets anyone inside make a call.

IV. What the State Cannot Know

The problem those windows are meant to solve is an old one. In 1920, the economist Ludwig von Mises argued that rational calculation over capital goods required prices generated through exchange. A planning authority could count physical inputs and still have no reliable means of judging their relative economic value. Hayek pushed the point further. The knowledge required for economic coordination is scattered among people embedded in circumstances no committee could gather, let alone transmit.

Much of it cannot be stockpiled in advance. It is generated only by people making choices under real uncertainty and living with what follows.

Finance is more or less this same problem wearing a suit. A factory turns a plan into an object. A bank or a fund decides which plans receive the means to become objects — which unearned income may be borrowed against, which unbuilt building will be allowed to rise. The circulatory-system metaphor undersells it: finance is really the economy's faculty of imagination, the mechanism by which a society decides which of its possible futures actually gets built.

Controlling finance means controlling which version of tomorrow gets made real.

China's own reform record makes an awkward witness for the political conclusion Hayek drew from this. Four decades of directed savings, state-guided infrastructure, and industrial upgrading don't sit easily next to the claim that central direction is epistemically crippled from the start. Beijing has never had trouble marshalling resources; the harder test was always whether the system could notice, in time, when those resources were already going to the wrong place.

Property was where the distinction became ruinous. The system retained its ability to build and lost its ability to know when building should stop. High-speed rail and semiconductor fabrication, under the same political system, largely avoided that fate, though neither was spotless. China's chip drive has produced its own corruption cases and stalled fabs, proof that waste and political vanity survive even where the state knows exactly what it is trying to build.

Three things stacked together in property, and nowhere else.

Land sales had become the primary revenue source for local governments, giving officials a direct fiscal stake in rising land values on top of the usual career stake in a reported growth number. Pre-sale financing meant homebuyers were, without quite realising it, the developers' largest unsecured creditors — paying in full for units that didn't exist yet and had no claim senior to a bank's if the money ran out. Cadre evaluation had, for a generation, treated construction activity itself as a proxy for competent governance, so the incentive to keep building ran through the promotion system as well as the balance sheet. None of that was true of a rail line or a fab.

Rising prices validated more construction, which required more borrowing. Banks extended the loans while their own health came to depend on those same land values holding up, and developers refinanced on the same assumption.

Evergrande alone eventually built more than 1,300 projects across roughly 280 cities, pre-selling apartments to families for whom the purchase was the single largest financial commitment of their lives.

In September 2021, creditors — construction workers, cleaners, suppliers, ordinary homebuyers — gathered outside Evergrande's Shenzhen HQ chanting for their money back. Police lined the building, and within days were removing protesters who wouldn't disperse. By 2023, Nomura's chief China economist put the number of unfinished, pre-sold units nationwide at roughly 20 million, with a funding gap around $446 billion.

Nobody in this chain behaved irrationally by the logic of their own institution. The trouble was that the signals had become entirely self-referential, detached from any independent read on what those assets could actually generate. Everyone was grading their own homework, and the homework kept coming back marked correct, right up until it very publicly was not.

James Scott gave this failure a name in Seeing Like a State (1998). High-modernist administration does not collapse because officials are stupid. It crumbles because legible metrics necessarily simplify the thing they represent, and past a certain point the simplification quietly takes the place of the reality it was supposed to be tracking. The local knowledge that would have caught the gap, what Scott called mētis, gets treated as noise for the same reason it's valuable: it resists the format the state needs in order to see it at all.

China's property machine wasn't cratered by a shortage of state capacity. It had plenty. It was undone by considerable capacity aimed, with total confidence, at a signal that had stopped corresponding to the ground underneath it. Authority stayed fully intact the entire time. What failed was perception, which is a much harder thing to catch failing, because nothing about the failure looks like weakness from the inside.

V. What Beijing Cannot Say Out Loud

Nobody in Beijing will make the next argument on the record. There is no cable or leaked minute to point toward, only a pattern of interventions since 2020 that is hard to explain any other way.

Hayek's framework was a general theory of complex economies. China tests it at a scale, and under political conditions that he never had occasion to confront. It has 1.4 billion people, a fertility rate below 1.3, and a political memory that makes distributed financial error a live wire. A property collapse doesn't merely dent household portfolios, but threatens the primary store of wealth for hundreds of millions of families at once. Central control of finance, on this reading, is risk management at civilizational scale.

The property crash happened anyway, which rather undercuts the argument. Beijing's likeliest reply would be that the crash proves the metrics were wrong, not that central administration is wrong in principle. Better dashboards, it might say, would have caught developer insolvency before local governments and state banks did.

But a better metric does not remove the failure Scott described. It relocates the blind spot to wherever the new metric doesn't reach, and it does nothing about the fiscal and promotion incentives that made the old metric worth gaming in the first place. Legibility has a cost, and the bill always seems to arrive at the most expensive moment.

Beijing's advisers would likely reach for a sharper objection and ask why not keep the method and lose only the independence — absorb Ant's architecture into a state fintech utility, run the same models on the same behavioural data, and collect the informational benefit without tolerating a private entity that converts knowledge into leverage?

It is a real option, but the reason it wouldn't deliver is already sitting in Evergrande's ledgers.

Ant's algorithm was accurate because the company had built a commercial system whose value depended on being right about borrowers the banks could not see. A bad loan was usually a bank's loss, not Ant's, and that was the regulatory problem. But every default still altered the data on which Ant's fees, partner appetite and future growth depended. The model learned from failure even when the balance sheet underneath it belonged to someone else.

Move the algorithm into a state utility and the data do not disappear. What changes is the hierarchy of purposes to which the resulting judgment must answer. The model may continue to calculate Jiwei's risk perfectly capably, right up until its conclusion conflicts with a lending target, a local government priority, or a borrower the system has decided cannot be allowed to fail. Evergrande's banks did not lack exposure. They lacked permission to let that exposure dictate the decision.

The algorithm was therefore only half the asset. The other half was permission to let commercial failure embarrass the model and alter the next judgment. A nationalised Ant could preserve the technique while gradually softening the feedback that made the technique useful.

The reverse experiment fares no better. Leave Ant independent, and the independence that keeps its information trustworthy is exactly what turns the company, over time, into something too systemically important to leave alone. Financial knowledge of this kind converts into institutional weight by a route so short that, past a certain point, the line between knowing and governing blurs.

VI. What the Record Shows

Beijing has now run two incomplete arrangements. In property, state banks carried the losses while institutional pressure kept those losses from correcting the signal — liability separated from consequence. In Ant, a private company generated the signal while carrying too little of the loss — knowledge separated from liability. Both let error accumulate somewhere the governing institution preferred not to see it.

Knowledge arrived in rough proportion to the hardness of that feedback. Hayek's wager was never simply that private ownership produces truth. It was that dispersed judgment requires consequences which no central authority can indefinitely postpone. China's experience complicates the wager without overturning it.

Beijing is running a third arrangement, permitting capital to leave through Hong Kong in amounts too large to be a rounding error, while encouraging firms to build a China-centred production system abroad along Shi Zhan's lines.

The informational problem does not disappear at the border. In fact, it becomes more acute. A company operating in Thailand or Indonesia has to know which partner can deliver, which rule will be enforced, and which local grievance can stop a factory. Such knowledge is useful because it is local. The authority accumulated from its possession is political for the same reason.

Midea's overseas platform and the Chinese Chamber's complaint in Indonesia are small signs of the bargain taking shape. Beijing needs companies capable of building systems and influencing rules abroad. It will also need to decide how much independent institutional weight those companies may accumulate before another sovereign function has escaped the perimeter drawn for it.

Whether this forbearance survives the next real test of it (a sharp depreciation, a bad quarter, a reason to feel it's gone too far) remains an open question.

The IPO halted in November 2020 makes a good story about a regulator and a billionaire, but it's more useful as an account of a political system running into the cost of its own success — needing financial intelligence that only an entity it couldn't fully control had produced, and then trying to keep the product while managing the producer.

Jiwei, for his part, is presumably still on the factory floor, still checking a Sesame Credit score he once complained hadn't moved in three months. The machine that once read him most clearly has since been made more legible to the state. Scott would have recognised the bargain. Beijing can now see further into Ant. Whether Ant can still see as far into China is another question.