Will the huge amounts of money being spent on AI really help internet companies make more money, or are those big spenders getting ahead of themselves?
Ross Sandler, an analyst at Barclays, said Tuesday that this is a very important question for investors right now. He said that the math doesn’t seem to add up. Wall Street thinks that AI will cost about $60 billion more than planned, but it will only bring in $20 billion more in cloud revenue in 2026.
This is one of the biggest debates in the investment community right now: will all the AI capital expenditures going into data centers turn out to be a 2000s-style overbuild of the telecommunications bubble, or is there enough AI revenue to justify it? That’s what Sandler wrote in a note to clients.
How you answer that question depends on which scenario investors think is most likely when they are trying to figure out why projected AI spending and projected AI revenue are not matching up. If investors think that revenue estimates are too low, it means that the company is spending too much on capital, or if they think that revenue estimates are too high and don’t take into account how AI will affect finances in the future.
Sandler believes that the huge amount of money being spent on AI—$167 billion since the AI craze began—may be due to FOMO, or fear of missing out, rather than a “Field of Dreams scenario” in which AI will soon bring in a lot of money.
That question is worth $167 billion, but Sandler calls it “the gazillion-dollar question.”
“From our early work here, the agreed upon amount for hyperscaler AI capital expenditure in 2026 is enough to support the current internet plus 12,000 new ChatGPT-scale AI products,” he wrote. In reality, though, Wall Street isn’t sure if there will be even that many AI applications that can make money, according to Sandler.
From his point of view, “lots of new services that will bring some of this bull case to light, but probably not 12,000 of them” are possible. In general, there are about 50 consumer apps outside of China that have more than 50 users. In the software market, a few dozen large companies sell about 1 billion seat licenses each.
To use that math, Sandler said, “there is no way the revenue from those competing AI products would justify the capex planned for 2026, even if all of these products were replaced by them.”
He agrees that AI services might have much better engagement metrics in the long run than their human counterparts. In this case, the possible effect of AI on revenue might not be overstated at this point. He says that Character.ai only gets a fifth of the daily inference requests that Alphabet Inc.’s GOOG, +0.80% GOOGL, +0.83% Google does, even though it has a lot fewer users.
“Just five characters.””Is ai the same as Google?” He asked Sandler. “That’s crazy to think about.”
But he wrote that he and his team are “leaning FOMO and expect someone to flinch next year, but it’s still very early for AI” when it comes to the big investment deal.
But he also said that the trend “doesn’t likely mean [Nvidia] NVDA, -1.91% is in any kind of trouble in the medium term, as we expect the AI capex to keep going strong for a couple years before someone blinks and cuts back.” He is mainly interested in internet stocks.
