Big investors are changing the question they ask about AI capital spending. After strong results from Microsoft and Amazon reassured markets that demand for AI infrastructure remains robust, some of the world’s largest asset managers are focusing less on whether Big Tech’s spending will pay off and more on which companies will keep delivering profits over the long term, Reuters reported in mid-August.
Capital spending, or capex, is money companies invest in long-lasting assets such as data centers, servers and chips. The largest technology companies are spending hundreds of billions of dollars a year on it to build capacity for AI.
Hyperscalers back in favour
Many fund managers kept large positions in chip stocks even after a sector sell-off in July, when doubts grew about whether AI spending was worth it and about rising Chinese competition. Now they are also adding to the hyperscalers, the biggest cloud providers, whose scale allows them to expand AI infrastructure quickly.
“The hyperscalers are being recognised in this moment as companies that are likely to be very large beneficiaries of this AI paradigm shift,” said Brian Barbetta, co-head of the technology platform at Wellington Management, which manages about $1.3 trillion. “They remain core holdings in our portfolios, and we’ve in fact increased our positioning in many of these companies recently.”
Their shares have had catching up to do. The stocks of the four biggest AI spenders all lagged a 75% rise in the Philadelphia Semiconductor Index, as well as rallies in the “neocloud” companies CoreWeave, up about 50%, and Nebius, up more than 200%. Neoclouds rent out AI computing power to customers ranging from AI labs to ordinary businesses, and have benefited from high prices for scarce capacity.
When does the spending pay back?
The central issue is timing. A Reuters analysis estimates that the hyperscalers will generate about $340 billion more in annual operating cash flow in 2027 than in 2025, while their capex is expected to rise by about $534 billion. In other words, spending is still growing faster than the cash it produces.
Richard Clode, a portfolio manager at Janus Henderson, expects that to reverse. “By later next year into 2028, we think you’re going to start seeing these companies growing profits and cash flow faster than the incremental capex growth,” he said.
John Lamb of Capital Group, which manages about $3.6 trillion, said data centers typically take 12 to 18 months to go from construction to producing revenue. “We’re just starting to see from the latest quarterly earnings this inflection point,” he said. “It’s not about whether chips are better investments than hyperscalers. It’s about having both in your portfolio.”
Risks for the neoclouds
Not every AI infrastructure company is seen as a long-term winner. Noah Weisenberger, chief US equity strategist at BCA Research, said neocloud providers could be vulnerable if new computing capacity comes online and prices return to normal, because they rely more on debt and on today’s high prices. He recommends buying hyperscalers and betting against neoclouds.
Clode made a similar argument: Amazon, Microsoft and Google have more lasting advantages because of their scale and customer relationships, and because they control both the computing capacity and the software that helps customers use AI efficiently.
Valuations are not extreme by recent standards. Hyperscaler valuations have fallen this year and remain below their post-pandemic peaks. Microsoft traded at about 24.6 times expected earnings, the highest of the group, and Meta at 17.6, the lowest.
The monetisation gap
Even optimists admit that AI revenue has a long way to go. Alberto Conca, chief investment officer at the Swiss wealth manager LGF+ZEST, estimates that AI revenue would need to grow fivefold to thirteenfold to justify current spending plans. Barbetta expects competition to reduce the number of winners as the market matures. “There are absolutely going to be fewer winners in the future than there likely are players today,” he said.
Investor appetite for AI-related stories is still strong elsewhere too, as the recent stock market debut of the robot maker Unitree showed in the humanoid robot industry.
Why it matters for businesses
For companies that buy cloud and AI services, the investor view has practical consequences. If computing capacity becomes less scarce, prices should fall, which would benefit customers. Businesses relying on smaller neocloud providers for cheap capacity should also consider the financial strength of their supplier, since a provider with heavy debt is more exposed if prices drop.
Our take
The shift in investor thinking is sensible. The debate about whether AI spending is too high cannot be settled yet, but it is clear that companies with real customers, broad product lines and their own infrastructure are better placed than those living on a temporary shortage. The Reuters numbers are the key point: capex is still rising faster than cash flow. Until that reverses, AI spending remains a bet on future demand, and investors should keep asking when it will pay off.