The artificial intelligence bubble, the debt crisis and the threat of war...

A reflection on the state of capitalism and what to do about it

There are currently two strong blips on the radar of economic instability: the massive and irrational concentration of capital into the AI sector, leading to fears of a financial meltdown; and – as per The Economist’s front cover this – growing fears of a generalised government debt crisis.

In this edition of Conflict & Democracy I am going to explore their common root – the moribund nature of post-2008 capitalism - and talk about the one obvious conclusion we can draw in advance of any meltdown: the state isn’t going to be bailing out Wall Street, Silicon Valley or any investor who loses their shirt in the coming crash. The demands of rearmament, and the impossibility of a further round of debt-driven accumulation will see to that.

First off, let’s survey the state of the AI bubble through two charts. During the past year ten lossmaking startups, most of them producing LLMs, have gained $1 trillion in value.

Among the tech companies that make real stuff and generate real profits, there is now a circular flow of capital, emanating primarily from the chip maker Nvidia and cash rich countries like Saudi Arabia.

Nvidia makes the chips, sells them at a price that – as the FT points out – is unlikely to be covered during their brief life before obsolescence. It then recycles its profits by buying services or hardware from established giants like AMD, Intel and Oracle, and investing in OpenAI.

This merry go round can, of course, go on recycling the same capital for some time until (a) it becomes clear that the thing the Nvidia chips are supposed to achieve – artificial general intelligence (AGI) – isn’t happening or (b) we achieve something like AGI but its much-vaunted productivity effects don’t happen.

Either way, money is flowing like crazy into the so-called Magnificent Seven tech companies. Among S&P 500 companies, Profits, capital expenditure, share price gains and market capitalisation are all heavily concentrated into the Mag7, as this report from Apollo Global Management shows. With Nvidia’s price/earnings ratio above 50, and the other big tech firms close to that, we are at a much higher level of peril and concentration than in either the 2000 or 2008 scenarios:

And as the FT’s Bryce Elder shows, even some of the revenues of the so-called hyperscalers (Google, Meta, Amazon and Microsoft) are based on monopoly rent seeking, in a market where startup competitors are massively undercutting prices.

From these facts alone we could build a model of how the AI bubble bursts: either lack of progress in AI models, or the simple dry-up of money from the Saudis and NVIDIA, or a black swan event (nearly all crashes begin because someone didn’t calibrate their risks properly) trigger the insolvency of the LLM vendors; this crashes the revenues of the hyperscalers, leading to a stock market crash and a flight to safety. Then if the state bails out the small investors and pension funds who got burned, there is a government debt crisis, austerity followed by a Depression. As this tanks tax revenues, any state on the hook faces potential bankruptcy.

© Paul Mason 2025

For me – having lived through the dotcom and structured finance bubbles, and observed the same rationalisations for exuberance over and over again – the only question is when this happens and how bad does it get?

One obvious circuit breaker would be not only for central banks to issue warnings – as the Bank of England have done – but for governments to start pressuring pension funds to stress test their positions against crash scenarios.

Because if/when this does happen, the public finances of the Western world are already so strained as to make the bailout option impossible.

The coming debt crisis

There are still people who believe that, because they possess a printing press, states can always money-create their way out of the debt positions they’ve amassed due to the 2008 liabilities, the resulting austerity, Covid and the need for defence spending hikes due to Russian aggression and Chinese rearmament.

I don’t. But the bond markets are already betting that some governments are going to try.

Lending to stable, democratic governments with tight fiscal rules is supposed to be the safest place to put your money on earth. Yet bond yields, which were close to zero at the start of the decade, are now between 3% and 5% for a swathe of developed countries, including both the USA and UK (see column highlighted in red below from this site.

As Henry Curr in The Economist points out: those yields – the effective interest rate paid by governments when they borrow – reflect investor fears that the value of the debt is about to be wiped out through inflation or some form of “financial repression”, whereby yields to savers are suppressed using policy measures and capital controls.

IMF deputy director Gita Gopinath pointed out last year that the condition that underpinned both the previous boom-bust cycles – a global savings glut arising from the development spurt of populous mid-ranking countries (India, China, SE Asia) – is over, to be replaced by a global bond glut.

The following two charts illustrate this starkly: global savings ratios are down compared to the boom years, while global debt to GDP was 93% in 2003 and is projected to reach 100% by 2030. The USA and several West European countries are already running wartime levels of debt – even before the pressures of climate, ageing and increased defence spending kick in.

IMF Map Showing countries with >100% debt/gdp ratios

So it is at least imaginable that if the AI bubble bursts, states will for the first time in modern history choose not to bail out those who lose their money – which will not only be big tech firms and the financial intermediaries but the whole of the global bourgeoisie, which is (as with the crypto scam) heavily invested in keeping the plates spinning until they, themselves, can exit to safer assets. Hardest hit of course will be Americans, who are forced to invest their savings in these highly concentrated and risky markets, and who have no social safety net to fall back on if they lose their jobs or homes.

When the flight to safety begins, there are only two exits: to gold (and physical assets) and to bonds themselves. At this point, the bond market – the proverbial 600lb gorilla – will have to engage its primate brain and figure out whether it trusts governments to use their newly lowered borrowing costs to manage the inevitable recession away from Depression territory.

What does this tell us about capitalism?

What I think this twin dilemma – AI bubble and potential bond market strike – tells us about capitalism is that it is running into what the historian David Harvey once called “limits”.

Industrial capitalism was, above all, based on free energy from carbon. Now that is no longer sustainable without creating climate chaos, the system is facing one-off transition costs that require either consumers (through energy bills) or governments (through tax) to fund new investments. That’s the first limit.

The second limit is demographic ageing, which is – for the first time in the history of capitalism - creating an unsustainable dependency ratio of the elderly on the young.

The third is the fragmentation of the rules-based international order: this might seem like an exogenous source of instability but if you consider the history of capitalism as a whole it is not. Rules-based systems have always underpinned periods of innovation and growth – starting with Germany’s move from silver to gold after the Franco Prussian war, and gaining rocket boosters after 1945.

We simply do not know the economic effects of a rules-based system collapsing when economies have become so interconnected by trade and digital communications.

But there is a final “limit” I want to explore – which is highly relevant to the AI mania that has gripped investors. It’s the one I explored in my book Postcapitalism: the idea that information technology itself alters the dynamics of capitalism in a way that makes it long-term unsustainable.

Artificial Intelligence and Postcapitalism

The Postcapitalism Thesis never rested on the achievement of AGI. Rather it states that – as Paul Romer outlined in 1990 - basic, human-designed processing power plus digital networks creates goods whose reproduction value can be reduced close to zero.

In 2015 I argued that this would either produce a new form of ultra-monopolised capitalism, sucking the dynamism out of the non-monopolised private sector, or – where rent seeking and monopolisation could be defeated – we would start to see islands of non-profit business models based on collaboration and sharing. I argued that if the left could promote the latter it could design a transition path beyond economies dominated by market forces – a path entirely different to the failed centralised planning model of the USSR and pre-Deng China.

At the core of this proposition is that information technology, by automating both physical and mental tasks, disrupts the normal process in capitalism, whereby labour is displaced by automation but “reinstated” through the creation of new, human-centric tasks.

Since then, numerous economists have used the possibility of achieving AGI as a mental framework to discuss the ultimate fate of capitalism.