“Keeping perspective” is Michael Pascoe’s mantra. He writes that AI mania has not just warped perspective, it risks blowing up reality.
Over more decades than I care to count of market watching and reporting, I haven’t seen a time when there’s been more wide-spread conviction that we’re experiencing a dangerous bubble that’s sure to pop, yet the money keeps pouring in to inflate it.
The lead up to the “Crash of 87”, now viewed in the rearview mirror as a minor hiccup, was relatively muted. Not even the “dot bomb” bubble at the turn of the century, when all a company had to do to get a share price boost was to add “e” or “dotcom” to its name, was as widely perceived as over-cooked as the present outlook.
“This time it’s different”
This time round there is vastly more money chasing itself with a circular investment boom at its core built on promises of revolution and, inevitably, that “this time it’s different”. This time round there are trillions of dollars being splurged on a heady mix of momentum riding, FOMO (fear of missing out), and massive bets that the first movers will win all.
There are the market telltales of sky-high valuations for inexperienced startups merely intending to build data centres or AI-somethings, but they are sideshows, mere flotsom that will be blown away when the reckoning occurs, as the shells and frauds were when the dot bomb burst.
As a general rule, things tend to turn out to be not as bad as you fear or as good as you hope.
This time round, the hope is about the level of bad.
The serious game is the impact both the reckoning and the AI promises will have on the real economy, on employment and wealth. That’s where the potential is for a GFC-scale event, not a piddling ’87 or dot bomb.
And this time round, still carrying the cost of all their COVID efforts, governments and central banks will be less able to ameliorate the pain. We’re already reaping the result of a global easing cycle with global government deficit spending fuelling asset inflation with its subsequent wealth effect. After what was, with the benefit of hindsight, overcompensation during COVID, the ammo isn’t there to fight another shock.
Two quite different articles this month have highlighted the impossibility of the nirvana being promised by the AI investment promoters. Former colleague Alan Kohler writing for the ABC was the bleakest, seeing a future where a GFC-size bust is the least-worst option.
Returns not there
The other is a note to clients by independent economist Gerard Minack showing the AI spend simply can’t generate the returns for investors to justify it.
I’ll come back to that. Kohler first, adding the AI and crypto bubbles together for frightening numbers:
Somewhere between $3 trillion and $6 trillion has been invested in building AI infrastructure and software, and that has been responsible for almost all US economic growth over the past year.
The top 10 American AI companies have provided most of the US stock market’s gains over the past two years and are now valued at $35 trillion, almost half the total market.
Meanwhile there are 20,000 cryptocurrencies worth $5.8 trillion, of which Bitcoin represents more than half.
The total cash in the AI and crypto bets is more than a quarter of global GDP; it’s probably the greatest technology investment boom/bubble in history.
No probably about it, in my opinion. It would not take a total crash to send shocks through the financial and real economies.
Jobs armageddon the quid pro quo
But Kohler’s Doomsday outcome is that it would be worse if the investment ends up being justified by profits created as it would create massive longterm unemployment for the many while the very few at the top continue on their present path of becoming even more unimaginably rich.
Gerard Minack’s note has a smaller focus and thus more concrete outcome: there’s a lot of capital being burned.
Minack limits his numbers to the “AI8” listed entities, Alphabet, Oracle, Microsoft, Amazon, Meta, Broadcom, Nvidia and Palantir. In this bubble phase, investors keep rewarding companies that increase planned AI-related investment spending. But:
“The larger the investment spend, the greater the revenue that will need to be generated to ensure an adequate return on that investment. In my view that revenue hurdle is already implausibly high. Whatever the technical wonders that AI will generate, the investment returns will disappoint. That will inevitably lead to significant market losses.”
Greatly simplyfying Minack’s analysis, the AI8 will conservatively have investment stock of more than US$1 trillion by the end of next year. That’s allowing for 20 per cent depreciation on an investment spend heading above US$2 trillion in 2027. (And a reminder that this is ignoring the investment spending of unlisted companies such as Open AI.)
$1T for a 10% return
So how much revenue will their AI businesses need to generate to get a reasonable return on this stock of invested capital? I’ll skip the details of Minack’s figuring but AI would need revenue of some US$925 billion a year to achieve a modest 10 per cent return on invested capital. And that return compares with the hyper-scalers’ current 25 per cent ROIC.
By comparison, Minack quotes Praetorian Capital’s Harry Kupperman’s observation that the incredibly successful Microsoft Office 365 subscription services had revenue of US$94 billion last year.
“In other words, to achieve an average ROIC, the AI industry will need to support 7-10 firms with businesses as widely deployed, and widely subscribed to, as Office 365.”
And then it gets harder. If the hyper-scalers’ current 50 per cent gross margin falls closer to the S&P500 average, they would require additional revenue of US$1.2-$1.6 trillion. As Minack concludes, “good luck with that”.
A further complication is that AI will cannibalise much of the hyper-scalers’ existing businesses.
I would further speculate about what competition between that many players would do to margins. There’s also the well-reported phenomenon of much of the AI splurge being circular – the major players are hanging out and taking in a lot of each other’s washing.
Show me the money
It’s all good fun until investors reach the imminent “show me the money” stage and the music stops.
That’s when the big boys take a hit and the fringe players, the bubble startups, lose their shirts.
As for the concurrent crypto-bubble, RBA Governor Bullock last week pointed to the challenge for the financial system’s security if quantum computing ever effectively works.
“If you believe what they say on the tin of quantum computing, what takes 200 years to decrypt now, to break, will take a matter of minutes. So it is a big threat,” she said.
Decrypted crypto is no crypto at all.
And the Toddler King
Then there is the little matter of the world’s biggest economy being run by a febrile toddler king and a mob of self-enriching accomplices.
It all adds up to the biggest mystery: how markets are so willingly charging higher to yet more records regardless of risk with the biggest gains at the riskiest end.
On one hand there’s the view that the reckoning is always a little further off. On the other, the second law of “old bond dog” Anthony Peters comes to mind:
“Nobody gets fired for being long a falling market, but woe betide anyone short a rising market.”
Good luck with that, too.
Michael Pascoe is an independent journalist and commentator with five decades of experience here and abroad in print, broadcast and online journalism. His book, The Summertime of Our Dreams, is published by Ultimo Press.