Artificial intelligence spending could rise $60B year-over-year in 2024, bringing the total to roughly $167B since the AI craze started, Barclays estimates. And although the spending frenzy has shown no signs of slowing down, the investment firm wonders when enough is enough.
Spending for the big hyperscalers like Google (GOOG) (GOOGL), Microsoft (NASDAQ:MSFT), Amazon (AMZN) and Meta (META) on AI-related capex in 2026 is likely enough to support the entire internet as it exists today, plus 12,000 new ChatGPT-scale AI products, the investment firm said.
For comparison, there are roughly 1.2B enterprise seat licenses (and fewer actual users) in western markets, after more than 30 years of building out software and over 25 years of building the consumer internet, analysts at the firm said. And there are only around 50 apps with more than 100M users, the figure ChatGPT has been able to achieve in a little more than two years.
And while many still believe the world is still in the early stages of the generative AI era — with examples like the number of ChatGPT users or the number of start-ups building AI products — Barclays said there have been only two “break-out successes” in the consumer and enterprise spaces over the past 20 months: ChatGPT and GitHub Copilot.
As such, the firm gets the sense Wall Street is growing “increasingly skeptical” about cloud capex spending and that some investors believe “capex investments may have to flatten out at some point … before 2026.”
What does all this mean?
Despite laying out some concerns, spending isn’t going to just grind to a halt, Barclays said.
Roughly $70B is expected to be spent on training foundation models and another $95B is likely to be spent on inference, or what Barclays called “the cost of serving up finished products.”
However, there is the chance that next year (or perhaps later), one of the major companies could “blink” and cut back on its spending.
Even if there is a cut in spending, it does not mean that innovation in AI is slowing down, the firm said, citing recent breakthroughs in smaller foundation models that are used for consumer devices such as smartphones or PCs.
“Silicon Valley is buzzing about AI changing the world, AI agents, digital employees, AGI, etc – and we do expect lots of new services that will bring some of this bull case to light, but probably not 12,000 of them,” Barclays wrote.
Implications
The implications of any potential cuts to cloud spending could be wide-ranging, both for the hyperscalers and their suppliers, notably Nvidia (NVDA).
Barclays reiterated it does not believe Nvidia is not in “any of kind of trouble” in the medium-term, as it expects AI-related spending to be strong for the next couple of years. Additionally, the lead times for new chips and the fact that AI is still in the early part of the cycle suggests “the party continues for the foreseeable future.”
However, Nvidia generates roughly 20% to 30% of its revenue from customers in areas that do not overlap with hyperscalers. And if the “final” stage of AI is realized sometime in the next couple of years — what many call general artificial intelligence — then those models could cost as much as $10B to train, Barclays said.
For comparison purposes, the cost-per-model in the current generative AI era is roughly around $200M. And there is the expectation that the majority of the major eight or nine AI companies will be comfortable spending $1B per model and “most” have come to terms with spending the aforementioned $10B spending that could happen as soon as next year or in 2026.
Meta
Of the companies that have boosted their spending related to AI, Meta’s could be seen as the “most egregious,” Barclays said, when looking at it through the lens of whether AI revenue or AI capex are closer to the real trend.
That begs the question: what is Meta planning to do with that much capacity?
It’s possible Zuckerberg and his executive team are working on new business models on the back of AI, which Barclays admitted is nowhere close to being built into estimates.
Zuckerberg hinted at this possibility on Meta’s most recent earnings call, pointing out the volatility in the stock when the company invests to scale new products.
“Historically, investing to build these new scaled experiences in our apps has been a very good long-term investment for us and for investors who have stuck with us and the initial signs are quite positive here too,” Zuckerberg said. “But building the leading AI will also be a larger undertaking than the other experiences we’ve added to our apps, and this is likely going to take several years. On the upside, once our new AI services reach scale, we have a strong track record of monetizing them effectively.”
