In an October 4 piece for Bloomberg, Seth Fiegerman and Carmen Reinicke wrote, of the AI frenzy, that “never before has so much money been spent so rapidly on a technology that, for all its potential, remains somewhat unproven as a profit-making business model.”
That’s an incisive assessment and I’d take it a step further: It’s still not entirely clear what we mean when we say “for all its potential” in the context of AI. “For all its potential” — or variants thereof — feels increasingly like a boilerplate disclaimer to preempt jeers from a world thoroughly enthralled and thereby predisposed to castigating skeptics as Luddites.
I’m not suggesting AI’s akin to the metaverse in being, um, an undefined hypothetical. And it’s worth noting that even technologies which perennially fail the use-case test, like blockchain, can produce vast fortunes and manifest in something marginally useful, like Bitcoin. (And please, spare me your “All the ways blockchain has changed the world” lists, or if you insist on them, don’t expect a response.)
Still, concerns are growing in some whispery corners that “capex bubble” isn’t just a buzz term good for selling articles, but rather an apt description of ~$1 trillion in AI spending by the hyper-scalers over just three years.
The figure above’s simple: It plots the share of Mag7 cash flow dedicated to buybacks and the share dedicated to capex, using estimates from one large Wall Street bank for this year and next.
Plainly, not all capex is AI capex, even among the hyper-scalers, but it’s safe to say most of it is. And if BofA’s projections are correct for 2026, those outlays will mean capex spending triples (and then some) versus 2021 for the vaunted septet which comprises 35% of S&P 500 market cap.
Regular readers are fully apprised: I’m generally a proponent of more capex and fewer buybacks assuming, of course, management’s capable of identifying spending-for-growth opportunities that’ll clear the hurdle rate and accrue to the business over the long-term.
The problem — or prospective problem — is that it feels like Mag7 management’s every bit as caught up in the AI hype as investors, and probably even more so. That, in turn, raises questions about whether the C-suite’s actually being realistic about the payoff from AI outlays.
It’s true that shrinking the float to boost the bottom-line is a kind of financial engineering and thereby doesn’t represent “real” profit growth. But I’ll take “fake” EPS growth funded through buybacks over no EPS growth “funded” with speculative capex any day.
Put as a question: What happens if, five years from now, a standing market joke asks, “Remember Nvidia?”
Mark Zuckerberg firmy believes in it so I, of course, do as well. Remember the success of the metaverse?
It would be an interesting exercise to calculate the depreciation schedule of the Mags for all this past capex plus the trillions estimated for the next couple of years. Given the lightning fast product cycles, even a one year writeoff seems too long. If the excessive growth cycles aren’t met, positive cash flows might be another AI hallucination.
Every software developer in the know is spending $100 a month on AI coding assistants. I will keep spending, my spending will increase, I got my intern a subscription, etc etc
‘From what I can see, software as a whole accounts for 5% of GDP. Not sure what perecentage of that is writing software. So even if it is makes coding remarkable faster, the TAM is pretty low, no?
Is it another example of “where’s the beef?” As in ROI to the LLM providers?
Fair point. I am just giving my example. In my world, we are also using it to draft basic consulting agreements with third parties.
Further, there will be a whole new class of applications that use AI to provide a superior product. For example, ingesting your emails to help you run your small business by suggesting next steps, reuters is injecting ai into their lawyer tools to generate the first pass at a legal document.
I’m generally not one to jump on the bandwagon. Never bought into crypto, ask h, never the metaverse or ar gaming, but I think llms are a pre big deal and will just become a thing we use everywhere provided the climate and political situation holds, which…
For sure it will be useful, just as its forebearers have been. But the payback load simply seems to enourmous.
A recent study indicating that AI is giving a boost to the accuracy and throughput of radiology readings has not led to a dop in demand for radiologists. That astonished me.
But perhaps that did not include demand for radiology image readers in India? And the same for junior coders?
It strikes me that AI may be a poisoning India’s future.
Add drafting of preliminary legal documents and preparing tax returns as well as, ohmygod, stock analysis.
So AI may be replacing lower cost Indian labor rather than (fat, lazy & overpaid) US professionals. If that’s the case, how much will US payroll costs be reduced?
Interesting WSJ article last week about AI investment. Author thought AI spending would need to get to about 5X current total software spend to cover the costs… A lot of the expense is debt he/she said.
Another interesting aspect to this is that tech is now capital intensive. Old investment selling point was that it was not capital intensive.
Time will tell.
H-Man, Yahoo!
Nvidia is gonna be knee deep in robots in 5 years…. AI is a big part of that. I wouldn’t start chiseling the gravestone yet.
Human nature, people get fired for being an “outlier” whether it is in the “next big thing” in a business (AI investment, etc) or in the investment world where “everyone has to own” XYZ (so safer to closet index and if you go down with the ship you are with everyone else so everyone can’t lose their jobs – the old saying “No one gets fired for owning IBM” – ofc that was 50 years ago lol). Herd mentality will always be alive and well.