Sohn Hearts and Minds conference: Why AI’s big short is not done yet
In the midst of a super cycle with some of the market’s sharpest stock pickers assembled, few were prepared to jump into cutting-edge tech.
Instead, the picks were decidedly old world. A US steelmaker (Steel Dynamics), a Finnish hardware chain (Puuillo), a sports promotion company (TKO), and even tried-and-trusted gold. There was one twist: Texas-based century-old energy infrastructure play SLB got a look-in on data-centre thematics.
The debate about AI’s sustainability has intensified in recent weeks. The Nasdaq is having a rare negative November, down almost 5 per cent from its record high. Combined with the prospect of US interest rates staying higher for longer, during the past week there has been some of the heaviest selling since Trump’s “liberation day” roiled markets in April.
In many ways, both camps are right. AI does represent an economic inflection point, but no one is quite sure how deep the J-curve runs before the technology starts delivering meaningful returns to owners – or end users.
Added to this are real doubts about sustaining the sheer volumes of capital needed to fund AI’s computing infrastructure. Current estimates now run into the trillions of dollars annually. And no one has answered whether there’s global capacity to do so without sparking a new round of tech inflation.
Speaking via video link at the Sohn Hearts & Minds conference, activist investor Daniel Loeb, who founded New York-based hedge fund Third Point, is conditionally optimistic on AI’s outlook.
Yes, there is a bubble in AI, he says. But he doesn’t believe it’s around the likes of Microsoft, Amazon or even the data-centre hyperscalers.
“We do see some real signs of excess in individual companies – this isn’t AI, but some of the quantum computing valuations got very speculative,” he said in conversation with Anthony Scaramucci.
Some unnamed IT infrastructure businesses have seen “crazy valuations” relative to the investments going in. Loeb nominated quantum computing play CoreWeave as having an “inexplicable valuation”. Before this month’s falls, the crypto-miner-turned-cloud-infrastructure play was up more than 220 per cent this year. It’s now down 45 per cent from its October peak.
“There’s a correction in that,” Loeb says. “The more mainstream companies that participate in that (AI bubble) neither went up as much, nor do I think they’ll correct as much”.
But still “it doesn’t mean that we won’t have some ugly days and weeks along the way”.
Amazon and Nvidia are among Loeb’s top holdings, valued at $US594m ($909m) and $US442m respectively. His single biggest holding is utility Pacific Gas & Electric ($US712m).
Loeb’s stock pick at Sohn was global music entertainment play LiveNation.
Late-Cycle Speculation?
As AI-related investment accelerates, questions mount about whether this represents late-cycle speculative activity.
There’s no doubt investment is flooding into digital infrastructure. Much like the telco fibre-cable boom that preceded the dotcom bust, the infrastructure left a lasting legacy. The current build-out is being compared with railroads and electrification of centuries past – even the shale oil boom of the early 2000s which tilted the US towards becoming a net oil exporter.
On Morgan Stanley numbers, 2025 cash capex for the 11 largest global hyperscalers is now tracking to $US469bn – or 68 per cent year-on-year growth. This includes upgrades from the most recent third-quarter reporting season.
Significantly, the 2026 forecast for capex on cloud computing is now tracking at around $US620bn, representing another 33 per cent jump.
The planned capex cycle is eye watering. Next calendar year, Amazon plans to spend $US146bn, Microsoft $US103bn, Google $US123bn and Meta $US107bn. Nvidia reports third-quarter numbers next Thursday.
For perspective, the combined capex bill for Australia’s BHP and Rio Tinto next year is just over $20bn.
Globally, Morgan Stanley is tipping $US1.09 trillion spending on cloud computing and data centres this coming year, making AI the most important investment story for the next several years.
It believes most of the spending is still ahead. But this will require trillions of dollars in revenue to cover the outlay. Consulting firm Bain & Co believes that, even under optimistic assumptions, the AI industry still faces a $US800bn revenue deficit
Why This Time Is Different
“AI is seen as the most important technology of the next decade by the biggest, most profitable companies on the planet. This should increase the willingness to invest, even if near-term returns on that investment are uncertain,” Morgan Stanley says.
And, unlike major recent capex bubbles – like the dotcom boom – much of this spending is backed by megacompanies with strong balance sheets and significant debt capacity.
Finally, there’s little evidence of overcapacity. Indeed, it’s the shortage that’s spurring the need to overbuild.
On Friday, Australia’s new data-centre hopeful, Firmus, secured commitments for a $500m private raising to fund its national rollout of cloud-based infrastructure. This gives the Oliver Curtis start-up an implied valuation of $6bn although construction (and income) from its flagship data centres are still years away. Just two months earlier, a raising put Firmus’ valuation at $1.9bn, with much of its blue sky built on its relationship with AI chip enabler Nvidia.
Nasdaq-listed shares in another Australian data centre start-up, the $20bn Iren, have slid more than 36 per cent so far this month. However, even after the falls, its shares are up more than 360 per cent this year.
In a note caution, billionaire warehouse developer Greg Goodman, who is a survivor of the global financial crisis debt implosion, has steadily increased his global bet on data centres. But this week he raised warnings that the top of the cycle could be approaching.
The amount of debt flowing into the sector, particularly from private credit funders Blackstone, KKR and Apollo, is “extreme”.
“I don’t think it’s sustainable” Goodman says. At some point the hype behind valuations will fade away.
Burry’s Big Short – again
Underscoring the conflicting signals: Michael Burry’s firm, Scion, famously shorted the US housing market in the lead-up to the global financial crisis, making hundreds of millions of dollars along the way.
However, it’s worth remembering that it was a painfully long bet and Burry very nearly lost everything by holding on to his conviction.
Many backers gave up and yanked their funds long before his big short paid off. This was dramatised in the 2015 film, The Big Short.
Burry has done it again with a near $1bn short bet on two tech-related darlings. Delayed filings emerged last week of Burry’s put options (the right to sell at a strike price) on Palantir and Nvidia for the period to the end of September, with a nominal value of nearly $US1.1bn. The bulk, or more than $US900m, was against Palantir – the data analytics play co-founded by Peter Thiel.
While the filings don’t show the acquisition date, since 30 September both stocks had increased more than 6 per cent before Thursday’s tech hit. As the puts near expiry, they suffer from time decay, making it a painful position to hold.
Although we don’t know the expiry date, we now have a pretty good idea why.
Burry is now reportedly liquidating his Scion Asset Management fund, and according to the Financial Times has warned that his estimates of share valuations have detached from reality. This doesn’t mean Burry was wrong, but rather he simply ran out of runway.
S&P downgrades ASX outlook over risk failures
For any sign the mood is swinging against AI, look no further than the Sohn conference in Sydney.