IBM’s CEO Just Did the Math on AI Spending — and the Numbers Don’t Add Up
IBM’s CEO Just Did the Math on AI Spending — and the Numbers Don’t Add Up
Rich DupreyTue, August 11, 2026 at 2:01 PM UTC
0
Arvind Krishna on XQuick Read -
Krishna's math shows 100 gigawatts of committed AI buildout demands between $1 trillion and $2 trillion in new annual revenue that IBM's CEO says simply isn't there.
The market prices in somewhere between 6 and 12 large AI model survivors, but Krishna expects only 2 or 3 to make it as models commoditize and switching costs fall.
IBM trades at 18x forward earnings with a 2.85% yield, selling enterprise software and consulting rather than chasing the infrastructure arms race.
It sounds nuts, but SoFi1 is giving new Active Invest users up to $3,000 in stock for a limited time, and all it takes is a $50 deposit to get started.2 See for yourself (Sponsor)
Big Tech is about to spend $725 billion on AI infrastructure in 2026 alone -- Amazon (NASDAQ:AMZN), Alphabet (NASDAQ:GOOG), Microsoft (NASDAQ:MSFT), and Meta Platforms (NASDAQ:META) combined, up 77% from $410 billion just last year. That kind of spending has become the defining feature of this market cycle, and most bubble warnings focus on the usual suspects: stretched valuations, circular vendor financing, or a handful of chatbot apps carrying too much investor hope.
IBM (NYSE:IBM) CEO Arvind Krishna isn't worried about any of that. He appeared on Nicolai Tangen's In Good Company podcast in May and laid out a different argument -- one built on kilowatts, dollars, and simple division. It's worth considering, because Krishna isn't a short-seller. He's a 35-year IBM veteran with a direct stake in how this plays out, and his math points somewhere specific.
SoFi Active Invest is offering a limited-time promotion. Open an account, fund it with $50 or more, and you could receive up to $3,000 in complimentary stock for Active Invest accounts. See for yourself by clicking here now.(Sponsor)
The Gigawatt Math That Doesn't Add Up
Krishna's case starts with AI data center power. He estimates 1 gigawatt costs $60 billion to $80 billion in semiconductors to populate. Companies have already committed to roughly 100 gigawatts of AI buildout globally -- pointing to a total of $6 trillion to $8 trillion in spending. Run that through a five-to-seven-year payback period, and Krishna's math demands an extra $1 trillion to $2 trillion in annual revenue, even assuming high-margin AI services at 20% to 30% margins.
"That much incremental revenue I don't believe is there," he said. That's the entire thesis in one sentence -- not that AI lacks value, but that the buildout has outrun the revenue math required to justify it within a reasonable timeframe.
24/7 Wall St.
A $725 billion bet meets a $2 trillion reality check. See why the math behind the AI boom might leave investors chasing a phantom revenue stream while only a few giants survive. © 24/7 Wall St.
Fewer Winners Than the Market Is Pricing In
Krishna's second point compounds the first. He expects the largest AI models to become commodities, with switching costs low enough that customers hop between them freely. That means the market is pricing for six to 12 large-model companies surviving long-term -- but "maybe two or three" actually make it.
Advertisement
If capital spending had come in at half of today's levels, Krishna believes it "completely makes sense." At double that, however, some of those companies are not going to be able to generate sufficient returns.
Instead, Krishna believes distribution will determine the outcome. Companies with an existing consumer footprint aligned to AI -- think search, cloud, or enterprise software already embedded in daily workflows -- have, in his words, "a pretty good chance" of winning.
Where IBM Sits in Its Own Story
The irony is that Krishna runs a company that's largely sat out the infrastructure arms race he's warning about. IBM carries a trailing P/E of 21 and a forward P/E of 18 -- a discount to Nvidia's (NASDAQ:NVDA) forward multiple of roughly 22 and well below most of the capex-heavy hyperscalers funding that $725 billion buildout.
IBM's dividend yield sits at 2.85%, backed by 31 consecutive years of increases, with free cash flow that rose to $2.5 billion sequentially in the second quarter, with management guiding to another $1 billion of full-year FCF growth. Revenue climbed 6% in the same quarter, with software guided above 10% for the year. IBM isn't chasing gigawatts. It's selling picks and shovels -- consulting, hybrid cloud, and enterprise software -- to companies deciding how much AI infrastructure they actually need. That's a bet on Krishna's own thesis being right.
Key Takeaway
Krishna isn't calling AI a fraud -- he's saying the infrastructure math ahead of the revenue that's supposed to fund it. For investors, that argues against chasing the highest-capex names purely on AI enthusiasm and instead favors companies with real cash generation today: a 18x forward P/E, a 2.85% yield, and free cash flow already growing beats a promise that $2 trillion in new annual revenue shows up on schedule.
Granted, Krishna has an obvious incentive to talk his book -- IBM benefits if enterprises get cautious about hyperscaler lock-in. But the underlying math is his own, verifiable, and worth checking against whatever AI capex updates come out of the second-half 2026 earnings season.
Want Up To $3,000 In Stock? SoFi Is Giving New Active Invest Users Complimentary Stock
Looking to grow your money but unsure where to begin? SoFi Active Invest is offering a limited-time promotion—open a new Active Invest account, fund it with $50 or more, and you could receive up to $3,000 in complimentary stock.
From $0 commission trading3 to fractional shares4 and automated investing, this app is designed to simplify investing for everyone, whether you’re just starting or already experienced. Its easy to sign up and secure your bonus.(Sponsor)
Contact editorial@247wallst.com for any questions or corrections.
Source: “AOL Money”
