Xi Jinping removed term limits by 2,958 votes to two. The margin is the story, not the result

Two votes against

On 11 March 2018 the National People's Congress amended the state constitution to remove the two-term limit on the presidency. Of 2,963 delegates, 2,958 voted in favour, three abstained and two voted against. The change itself has been analysed exhaustively. The margin has not, and it is the more informative number: 99.83 per cent agreement on a reversal of the succession norms established under Deng Xiaoping.

Every governing body produces near-unanimous votes on procedural matters, and that means nothing. The signal is what happens on the hardest question available. Removing a term limit that had structured leadership succession for decades is about as contested as a constitutional question gets, and it produced two dissenting votes out of nearly three thousand. Either the delegates genuinely held one view to within a rounding error, or the institution no longer generates the disagreement it was built to register.

What 2026 has contained

2026 marks the first year of the 15th Five-Year Plan, with economic output projected to reach 140 trillion yuan. In July, speaking at the World Artificial Intelligence Conference in Shanghai, Xi called on countries to take up the opportunity of open-source AI and announced an alliance extending cooperation across Africa, Latin America and Asia. Donald Trump visited China in May and Xi is scheduled to make a state visit to Washington in September, making 2026 the most active year of direct US-China leader contact in some time.

Unanimity is an information problem

A body that returns near-total agreement has stopped functioning as a check and started functioning as a ratifier.

The mechanism is not sinister and it is not unique to any one country. Any hierarchy in which disagreement is costly and agreement is free will converge on agreement, and the convergence happens without anyone ordering it. Delegates read the room, and the room has already been read for them. The consequence is specific: the institution stops producing the one output it exists to produce, which is a reliable signal about where support is thin. Once that signal is gone, the centre is left estimating its own support rather than measuring it.

Read the margin, not the outcome

The margin on contested questions, not the outcome, tells you whether an institution can still surface disagreement.

A useful comparison is what a functioning contested vote looks like. Constitutional amendments in most systems clear with meaningful minorities against — twenty, thirty, forty per cent — because the question is genuinely hard and constituencies genuinely differ. A 99.83 per cent result on a hard question is not a stronger version of that; it is a different kind of event. The right reading of the two dissenting votes is not that opposition was negligible but that the observable channel for expressing it had narrowed to two people willing to use it.

Where a system with no dissent breaks

Systems that cannot register dissent cannot transmit bad news, and bad news is what correction requires.

The failure mode is informational rather than dramatic. A centralised system executes rapidly, which is its genuine advantage, but rapid execution of a wrong decision is worse than slow execution of a right one. Correction depends on accurate information reaching the top, and every layer that has learned agreement is free and disagreement is costly filters that information on the way up. The risk is not that the centre ignores warnings; it is that the warnings arrive already softened, and nobody involved experiences themselves as concealing anything.

What the same concentration buys

The same concentration produces capabilities that fragmented systems cannot match, and the 15th Five-Year Plan is the clearest example. A democracy cannot commit capital on a five-year industrial schedule because the schedule outlives the mandate; the plan is renegotiated every time the government changes. China can hold an industrial direction across a decade, which is why its position in batteries, solar and now AI infrastructure was built while comparable economies were still debating the first phase. The open-source AI alliance announced in July is the same instrument applied to a newer sector.

The strongest argument against this

The reasonable objection is that vote margins in the NPC were never a meaningful signal, so their disappearance costs nothing. The body has always been a ratifying chamber rather than a deliberative one, real debate occurs inside the Party before anything reaches a vote, and treating a public tally as an information channel misunderstands where Chinese decision-making happens.

That objection is largely right about where deliberation occurs and it does not dispose of the point. If the internal channel is the real one, then the question simply moves: does that channel still produce dissent, and is there any external way to verify it. The published vote was at least a weak observable. What replaced it is an unobservable, and an unobservable check cannot be assessed by anyone outside — including, at the margin, by the centre itself, which is the part that matters.

What a five-year plan can do that a manifesto cannot

The practical difference between a five-year plan and an election manifesto is not ambition but survivability. A manifesto is a bid for a mandate and is renegotiated the moment the mandate changes hands; a plan issued by a system with no scheduled handover is a commitment that capital can be deployed against.

That distinction shows up in industries with long build times. Battery, solar and semiconductor capacity require capital commitments that outlast any single government, and suppliers price the risk that the policy disappears halfway through.

This is the strongest available case for the arrangement, and it should be stated at full strength rather than dismissed. Whatever one thinks of the politics, the industrial results in several sectors are not an accident of subsidy alone — they are a product of policy continuity that fragmented systems structurally cannot offer.

Open source as an industrial policy instrument

The July announcement is worth separating from its framing. Advocating open-source AI and building an alliance around it across Africa, Latin America and Asia is not primarily a technical position; it is a distribution strategy.

Open weights travel into markets that cannot afford proprietary licensing and into governments unwilling to depend on a foreign commercial vendor. A model that becomes the default in fifty developing markets sets the standards, the tooling and the talent pipeline in all of them, at a fraction of what buying that position would cost.

It is the same logic this publication has traced in other contexts: reach purchased at low margin is worth more than yield extracted from a small base, provided there is something downstream to monetise. The open-source route is the cheapest available way to establish an installed base at continental scale.

The comparison the vote invites and does not survive

It is tempting to compare the 2018 amendment directly with Russia's 2020 term reset, and the surface parallel is real: both removed a scheduled limit on tenure and both passed with negligible resistance.

The mechanisms differ in a way that matters for forecasting. Russia's change reset a count while leaving the office and its election intact; China removed the limit from the state presidency, which is in any case the less important of the leader's titles, with the General Secretaryship of the Party never having carried a term limit at all.

The practical consequence is that the Chinese change was more symbolic and less structural than it is usually described, and that the succession question there was already open before 2018. Analysts who date the problem to the amendment are dating it about two decades late.

A caveat on reading closed systems

Everything above rests on published figures: the vote tally, the plan's start year, the output projection, the dated visits and the conference announcement. None of it depends on claims about internal deliberation.

Reporting that purports to describe factional disputes, personal rivalries or the health of senior officials in a closed system should be treated as unverified regardless of the outlet carrying it. That category of claim has an extremely poor accuracy record and none of it is used here.

The narrow claim being made is that a published vote distribution is a real observable, that this particular one is extreme, and that extreme distributions have known informational consequences. That argument stands or falls on the arithmetic, which anyone can check.

What transfers outside politics

Any organisation can measure this on itself, and most will not like the answer. Take the hardest decision your board, committee or leadership team made this year and look at how the room actually split.

If genuinely difficult questions are producing near-unanimous outcomes, the likely explanation is not that everyone agrees but that disagreeing has become expensive relative to going along. That is rarely the result of an instruction; it is the accumulated effect of what happened to the last few people who objected.

The practical repair is structural rather than cultural: assign someone the explicit job of arguing the other side, record dissent in the minutes by name, and make sure the person who raised the objection that turned out to be right is visibly rewarded. Cheap dissent is not a virtue, it is an information system.

How to read any governing body's margin

For any board or legislature, look at the vote distribution on the hardest question of the year.

Find the hardest question the body voted on in the last cycle and look at the distribution rather than the result. Compare it against the same body five and ten years earlier, because the trend carries more information than the level. Ask what happens to the career of someone who votes against, since that answer explains the distribution better than any statement about consensus. Then look for where dissent has relocated to, because it does not disappear, it moves somewhere less visible. Finally, judge the system on how it handles a decision that turns out to be wrong, which is the only real test of a correction mechanism.

Watch the plan's first revision

The 15th Five-Year Plan will encounter conditions its drafters did not anticipate, as every such plan does. The informative moment will be the first public revision: whether targets are formally adjusted, quietly restated, or held while reality diverges. A system that can revise a flagship plan in public retains a correction mechanism. One that cannot has substituted commitment for feedback.

Reassess at the plan's mid-term review rather than on annual output figures, and treat single-year GDP prints as too noisy to judge a decade-long industrial strategy. Reassess separately after the September Washington visit, since a concrete bilateral outcome would say more than either side's framing of it.

Editorial conclusion

Two delegates out of 2,963 voted against the most consequential constitutional change of the era. That number is usually reported as a measure of Xi's authority, which it is, but it is also a measure of what the institution can no longer do. The concentration it records is what allows China to hold an industrial direction for a decade while its competitors relitigate theirs every election. It is also why, when the direction is wrong, there is no longer a body positioned to say so early enough to matter.