The global bond rout that has roiled financial markets provides a sharp and threatening edge to the layers of pre-existing risks within the markets.
Last weekend’s letter by Bank of England governor, Andrew Bailey, to the G20 finance ministers and central bankers attending the summit in North Carolina detailed some of those threats to the global economy and financial system, with special mentions for AI-related security and market risks.
Bailey said that markets, while absorbing the supply shock of the war in the Middle East and the energy-driven inflation it has generated, remain vulnerable to a “potentially disorderly” correction that could spread across borders.
He referred to the “fragilities” in sovereign debt markets, the increased use of leverage by market participants, vulnerabilities in private credit and the interconnectedness of private credit with other parts of the financial system and “stretched” asset valuations, particularly for artificial intelligence-related investments.
The starting point for any assessment of the levels of risk in the financial system is the amount of leverage within the system. It’s awash with debt and leverage.
Government debt in the post-pandemic era is at record and arguably unsustainable levels, with government debt alone nearly 100 per cent of global GDP. Total debt exceeds 300 per cent of global GDP.
While household balance sheets, globally, appear to be in reasonable shape, overall private sector debt, according to the International Monetary Fund, is above 140 per cent of global GDP, with non-financial corporate debt around 150 per cent of GDP.
The threat posed by overly leveraged governments and companies is a conventional one, albeit still threatening and with the level of that threat being exacerbated by the rising in bond yields and interest rates more generally flowing from Donald Trump’s trade war and real war-driven inflation.
The AI threat, to which Bailey devoted a large proportion of his letter, is less conventional.
He noted the increase use of leverage in equity markets, including the growth of leveraged exchange-traded funds and correlated momentum-driven investment strategies and a growing presence of hedge funds, who are also exposed to sovereign debt strategies, in equity markets.
“The issue is not simply that investors are borrowing more, but that leverage is interacting with high valuations and market concentration, in particular the increasing cross-investment between artificial intelligence companies and hyper-scalers, in a way that could amplify a future market correction.
“I remain concerned therefore that a large shock or combination of shocks could concurrently trigger multiple vulnerabilities,” he wrote.
“Circular” financing that has been a feature of the boom in AI – a tangled web of relationships between those, particularly Nvidia and the hyperscalers who have balance sheets and cash flows and the AI developers that have the technology but lack the financial resources to fund the infrastructure of chips and data centres to support it.
Nvidia, in particular, is acting as the lender of first resort to the sector, providing equity and debt funding support in return for large-scale purchases of its chips.
It’s at the epicentre of a web of incestuous relationships, although there are plenty of other strands to that web as the hyperscalers also race to develop their own AI offerings and the infrastructure to support them.
In effect, the financial entanglements have created acute market concentration and a high level of mutual dependence and mutual vulnerability among the companies in the sector.
There’s also a rapidly increasing level of debt funding within an AI sector that, on a standalone basis, produces nothing but torrents of red ink.
Even the hyperscalers – companies like Google, Amazon, Meta, Microsoft and Oracle – are issuing debt because the scale of their AI investments has exhausted their cashflows.
The combination of debt – whose value, from the borrowers’ perspective, is static, and inflated equity valuations, which are potentially volatile – is a combustible mix.
Hindsight might show that AI valuations aren’t overly inflated, if the companies’ eventual profitability lives up to the extraordinary levels of optimism baked into their valuations.
A valuation of almost $US2 trillion ($2.8 trillion) for Elon Musk’s SpaceX – which lost $US541 million in the June quarter – or the mooted $US2 trillion Anthropic hopes its initial public offering will value it at tend, however, to suggest that valuations in the sector are extremely vulnerable to almost any level of adverse events, with the rising levels of debt within the sector amplifying the consequences of any negative shift in market sentiment towards AI.
The combination of inflated valuations, market concentration and increasing leverage explains why Bailey believes that a large shock, or combination of shocks, could trigger multiple vulnerabilities.
The spiking bond yields and the prospect of a Federal Reserve rate rise in response to continuing above-target inflation rates in the US, which has the world’s most important and influential financial markets and is the epicentre of the AI boom, are an obvious dual threat to financial stability.
They threaten to increase the cost of servicing the debt AI companies have taken on and, perhaps, decreasing the access to it that the AI pioneers increasingly rely on, while also undermining the arithmetic of the extreme valuations being assigned to AI companies.
Higher interest rates mean lower net present values as the hoped-for torrents of future cash flows are discounted back to today’s dollars.
The risks of AI aren’t entirely financial, although the non-financial risks could have financial and market impacts.
As Bailey noted, frontier AI models are showing increasing autonomy and threat capabilities.
The starting point for any assessment of the levels of risk in the financial system is the amount of leverage within the system. It’s awash with debt and leverage.
Late last month there was yet another instance of an experimental AI model escaping its “sandbox,” gaining access to the internet and hacking into another company’s systems to cheat on a test OpenAI had given it, none of which is supposed to happen. Anthropic, Meta, and OpenAI, previously, have had similar incidents of their agents escaping supposedly sealed environments.
In a highly-connected global financial system, based on widely shared infrastructure, those episodes underscore a novel source of risk and vulnerability.
While, so far, a Trump administration that has received significant financial support from the big technology companies (and Elon Musk) has been loath to respond to the obvious threats exposed by the incidents, at some point there will be an administration that will seek to slow the breakneck pace.
Given that the companies are racing to try to generate the cash to cover their ever-escalating AI-related investments, any serious attempt to regulate them could be a threat to their value and financial stability.
Among the multiple vulnerabilities Bailey referred to, the market risk – the risk that the equity market won’t continue to supply new capital at increasing valuations or that access to debt markets dries up – and the risk of regulation probably represent the most obvious specific risks to the AI boom, overlaid by the broader threats lurking with an overly-leveraged global economy and financial system.
The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.
From our partners
About the Author
