Every year, when it is revealed that the AI thematic once again contributed most of their returns, the super fund chief investment officers that who front the media to talk them up have one refrain when asked what they’re doing about the risk that comes with it: “diversifying”.

And for years, that was a good enough answer. It was possible to spread the risk out, to buy toll roads and airports and bonds. But now that’s looking a lot harder, and even less rewarding (not that it was in the first place; for decades, diversification – in this case, daring to invest even small amounts of money outside the S&P500 – has hurt you).

Global equities are, obviously, heavily exposed to AI – including in the US, and emerging markets, which have long been proffered as an alternative to their richly-valued developed market counterparts. But then so are alleged safe haven assets like Treasury bills. That’s without getting into the massive amount of lending, both public and private, to AI firms; private equity, where new capital is chasing AI; and infrastructure, where funds are shopping for data centres and the energy assets that support them, as well as the fibre networks that connect them.

It’s all one risk, and a significant one. As the Bank for International Settlements (BIS) noted in its annual economic report, intense competition between AI firms may push them to over-commit resources to projects with “still uncertain returns” – leaving them all vulnerable to disappointments in the payoff, and turning a capex boom into a protracted investment bust.

And with a shortage of supplies in advanced semiconductor chips, grid equipment and electricity, AI firms are also attempting to lock in future capacity through long-dated contracts, which will only “further expose them to any disappointments in demand”, the BIS said.

To find out what that looks like on the downside, the BIS says it’s “instructive” to look back at historical investment booms. 

“The canal mania of the 1830s, the British railway mania in the 1840s, the electrification exuberance of the late 1920s (roaring 20s) and the dotcom boom of the late 90s all shared one common trait: a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify. These episodes with an eventual reversal in investment, inducing economy-wide recessions.

“The scale and pace of the current AI investment boom accompanied by expectations of large productivity payoffs bear resemblance to these precedents, highlighting potential downside risks in the near term.”

Then there are the potential amplifiers to that risk; rising leverage (from an admittedly low base) in hyperscalers, circular financing, the growing mass of retail investors, and the emergence of hedge funds as intermediaries in government debt markets.      

For super funds, there aren’t many uncorrelated drivers of return left, and they don’t exist at the scale that would allow them to truly hedge out their exposure to AI. They could buy truckloads of gold, take on all the Florida hurricane risk there is or tip billions into Japanese activist strategies, and it wouldn’t move the needle. All of that is compounded by the Your Future Your Super performance test and the peer comparison metrics used by regulators and research houses to judge returns every year, which do not reward deviating from the benchmarks in any substantial way.

So the “diversification” answer doesn’t really cut it. Super funds should be telling people what they plan to do if the bet doesn’t pay off, given it’s one they’re now making across their portfolios, whether they intend to or not.

Super funds themselves admit that they don’t have as clear a picture as they would like of the dependencies and betas lurking within their portfolios, and it could be that their risk exposure is even greater than that highlighted here. It would seem that the first step towards mitigating some of it would be to figure out exactly where it is. Using the total portfolio approach (when it is truly value accretive), with the requisite technological uplift in areas like exposure and risk management systems, is one way of doing that.

The next step would be a rethink of the performance test to discourage herding around benchmarks, though it is increasingly obvious that some of the problems there are behavioural and that funds would much rather fail together – where their underperformance is unspectacular on a relative basis – than alone.

The upside case for the AI boom could be quite real, and the rewards for participating in it massive. But funds must think hard about how they will manage the downside – starting with figuring out how much of the bust they will share in, if or when it comes.