Markets

Is something tipping here?

The slump in AI stocks after the 2026 US midterms, tested against the hallmarks of a tipping point.

7 min read

Translated from the German original.

IThe Thursday

On 3 November 2026 the United States elected a new Congress. Two days later, on Thursday, the tech-heavy Nasdaq index fell by around five and a half per cent, its sharpest drop in a year and a half. The semiconductor index lost twelve per cent. Nvidia, the company whose chips sit in almost every AI data centre, shed around nine per cent in a single day, AMD and Micron around seventeen.

The next morning the movement travelled around the globe. In Tokyo the Nikkei fell by six per cent; in Seoul the Kospi plunged almost nine per cent at the open, and trading was halted for twenty minutes. The headlines spoke of a worldwide aftershock.

The reports cited fears that the new majorities could slow the construction of AI data centres: through regulation, through conditions, in some states through a building freeze. On Friday chip stocks partly recovered.

The word was quickly at hand, as it always is at such moments: is this the tipping point? Is the AI bubble bursting? The question is legitimate. But it can be asked more precisely than the headlines do.

IIGrain and avalanche

In 1987 the physicists Per Bak, Chao Tang and Kurt Wiesenfeld described a simple model: let grains of sand trickle onto a pile one at a time. Mostly nothing happens. Sometimes a few grains slide off. And occasionally a single grain triggers an avalanche that sweeps away half the pile. The grain that triggers the big avalanche is in no way different from the thousand grains before it.

Bak called this self-organised criticality. Through its own dynamics, the pile brings itself into a state in which events of every size are possible. The cause of the avalanche then lies not in the last grain but in the state of the pile.

Applied here, this means: the election was probably the grain, not the cause. The pile was already steep. As early as October, chip stocks and data-centre operators had given ground, because of rising interest rates and fears of over-investment. Several of the big technology companies now spend more on AI infrastructure than their ongoing business brings in, and finance the difference with debt. The autumn quarterly results had tended to disappoint. A market in this state does not need a big push. It needs any push.

IIIThe loop

The investor George Soros coined a term for such markets: reflexivity. Expectations about the future change the future they expect. The page on self-reference on emergenz.art describes this as a loop that can tip.

In the build-out of artificial intelligence this loop is particularly tight. High market valuations make it easy to raise money for new data centres. New data centres promise growth that justifies the high valuations. As long as the loop runs upwards, it reinforces itself. If it runs downwards, it reinforces itself just the same: falling prices make financing more expensive, the build-out slows, and the slower build-out confirms the falling prices.

An analyst at Barclays described the shift in game-theoretical terms. The AI race, he said, had turned from a prisoner's dilemma into a war of attrition. In a prisoner's dilemma everyone invests because they fear the others will. In a war of attrition, whoever holds out longest wins. Electricity and land have become scarce, and scarce resources favour the largest balance sheets. Such a change in the rules of the game is itself a kind of phase transition: the same players behave differently because the structure in which they play has changed.

IVOne tipping point pushes the next

The truly remarkable part happened before the election, and it happened slowly. Data centres need enormous amounts of electricity, often water too, and they change the places where they are built. For a long time resistance to them was a local matter: a town meeting here, a lawsuit there. According to an industry association's count, between the beginning of 2024 and May 2026 at least 46 projects in the US with announced investments of over 170 billion dollars were halted, withdrawn or delayed after protests. The resistance came from both political camps.

In 2018 the sociologist Damon Centola showed in experiments how social norms tip. Below a certain size, a committed minority remains ineffective. Once it reaches about a quarter of the group, the whole group tips. What was a fringe position becomes consensus overnight. Whether resistance to data centres has crossed such a threshold cannot be measured. But the pattern is recognisable: in this election campaign, many scattered local concerns became a political platform.

The crash would then be not a single tipping point but a cascade: a slow social change pushes a fast financial one. It is the same logic climate research uses when it speaks of coupled tipping points – the Greenland ice weakening the Atlantic current. Except that here no oceans are involved, but opinions and expectations.

You see the grain. You only see the pile it lands on once it starts to slide.

VIn step

That a slump in New York reaches Tokyo and Seoul the next morning is no accident of time zones. Markets that follow their own rhythms in calm times suddenly move in lockstep during crises. The literature knows this as one of the signals that precede or accompany a shift: the parts of a system lose their independence, and correlations rise. Korea and Japan, with their chip and memory manufacturers, are closely coupled to the data-centre build-out. What began as a political question in the US took effect in Asia as a question of order books.

The countermeasure is also worth noting. The twenty-minute trading halt in Seoul is a deliberately built-in interruption of the loop. It is meant to prevent sales from triggering further sales before anyone has had time to think. It is an attempt to cut the feedback for a moment. The page on finance on emergenz.art tells of the 2010 flash crash and of why such systems can only be steered through their rules, not through their outcomes.

VIThe test

Whether a system has tipped or merely swung violently can be checked with four questions.

Is there a reinforcing feedback loop? Yes. The loop of market value, financing and build-out is well documented, and it runs in both directions.

Are there two stable states? Conceivably. One would be a market that keeps carrying the AI build-out with ever new capital. The other a market that wants to see profits from the companies first before financing more. Whether there is a smooth transition between the two or a threshold remains open.

Were there early warning signals? In hindsight, yes: the weakness in October, the growing debt, the disappointing quarterly figures. But finding early warning signals in hindsight is easy. The hard question is whether they could have been recognised as such beforehand, and with markets there is rarely an honest answer to that.

Is there hysteresis? This is the decisive question, and it has not yet been answered. A system that has tipped does not return when the trigger disappears. Chip stocks partly recovered on Friday. That points more to a violent jolt than to a tipping point. Whether they regain their old level, or whether the market settles into a new, lower state, will only become clear over weeks and months.

VIICoda: Proximity, not location

Whoever says “tipping point” claims irreversibility. That is a strong statement, and in most headlines it is not backed up. What can be said about complex systems is more modest: one can recognise that a system is approaching a threshold. Where exactly it lies and when it has been crossed is almost always seen only afterwards.

The Thursday after the election showed how steep the pile is. Whether it has slid has not yet been decided. That is not an evasion but the most precise answer that can be given today. And it holds for markets just as it does for glaciers: the question is not whether there is a crash, but whether things go back to the way they were afterwards.

As of 8 November 2026. An update will follow once it becomes clear whether the market has recovered. This text describes a system dynamic. It does not judge any party and is not investment advice.

An interpretation from the perspective of a thesis – whether the build-out of AI follows a necessity – is attempted in the essay “Die Probe” (German) on mkoenighaus.de.

Sources and further reading

  • Bak, P., Tang, C. & Wiesenfeld, K. (1987): Self-organized criticality: An explanation of the 1/f noise. Physical Review Letters 59, 381. DOI: 10.1103/PhysRevLett.59.381
  • Bak, P. (1996): How Nature Works. The Science of Self-Organized Criticality. Copernicus. — The sandpile model, also applied to markets.
  • Soros, G. (1987): The Alchemy of Finance. Simon & Schuster. — Reflexivity in financial markets.
  • Sornette, D. (2003): Why Stock Markets Crash. Critical Events in Complex Financial Systems. Princeton University Press.
  • Scheffer, M. et al. (2009): Early-warning signals for critical transitions. Nature 461, 53–59. DOI: 10.1038/nature08227
  • Centola, D. et al. (2018): Experimental evidence for tipping points in social convention. Science 360, 1116–1119. DOI: 10.1126/science.aas8827
  • CNBC (6 Nov 2026): AI stocks hammered by midterm shock as data center buildout becomes a ‘war of attrition’. cnbc.com
  • NPR (5 Nov 2026): Fears of an AI crackdown after midterms spark $2 trillion rout in tech stocks. npr.org
  • ZDFheute (German): KI-Aktien im Sinkflug: Warum die Börsen nach der US-Wahl zittern. zdfheute.de