- Google, Meta, and Microsoft all reported earnings and showed no sign of slowing their AI investments.
- Each spent more on AI infrastructure this quarter and warned of even higher costs in 2026.
- Their escalating bets on AI come as investors debate whether the boom has become a bubble.
Talk of an AI bubble isn't slowing down Big Tech. Google, Meta, and Microsoft are spending more than ever to build the backbone of the AI boom.
All three reported earnings on Wednesday, and all three told investors they'll spend even more on the data centers and chips at the heart of their massive AI bets — raising guidance and saying 2026 will bring even higher spending than 2025.
Google said it now expects to spend between $91 billion and $93 billion this year on capital expenditures. That's an increase from the $85 billion figure it gave to analysts and investors in July, which was itself an increase over the guidance it gave in April.
"We are investing to meet customer demand and capitalize on the growing opportunities across the company," CEO Sundar Pichai said in the company's earnings release. Google reported a record $102.3 billion in revenue for the quarter.
Microsoft said it spent $34.9 billion in capital expenditures in the quarter, up from $24.2 billion in the previous quarter, as "demand again exceeded supply" for its cloud computing capacity. Amy Hood, the company's CFO, told investors it will spend more in 2026 than it spent in 2025, "increasing our spend on GPUs and CPUs." Its revenue was up 18% to nearly $78 billion.
Meta updated its 2025 capex guidance to $70 to $72 billion, up from its previous guidance of $66 billion to $72 billion, and CFO Susan Li said 2026 will bring even more AI spending, "with growth primarily driven by infrastructure costs." It reported revenue for the quarter of $51.2 billion, beating Wall Street's estimates of $49.5 billion.
With Big Tech collectively spending hundreds of billions of dollars on AI infrastructure, analysts and investors have begun sounding the alarm about an AI bubble.
The earnings results may calm some of those fears, proving that Big Tech is generating significant revenue from their AI products and services, making it possible that they might someday be able to recoup their investments.
Gil Luria, an equity analyst at DA Davidson, takes the middle view on whether a bubble is forming. "These companies represent real demand," he said. "So if they are buying more chips and building more data centers, that's healthy."
"We are also seeing elements that are bubbalicious," he said. Companies borrowing tens of billions of dollars for speculative investment, and circular spending, such as Nvidia investing in cloud computing provider CoreWeave, is worrying. "That's unhealthy behavior."
"Both things are happening," he said. "It depends on what side of the elephant you are looking at."
Coming into this quarter, Business Insider estimated these three companies, along with Amazon, would spend $320 billion on capex in 2025, primarily for AI infrastructure. That's more than the GDP of Finland and just shy of the total revenue ExxonMobil earned in 2024 — and it's now destined to be even more with the recently announced spending.
On Tuesday, the CEO of chipmaker Nvidia, Jensen Huang, provided a preview of the companies' capex spend when he said his company had $500 billion in orders for the chips that power artificial intelligence.
Jacob Sonnenberg, a portfolio manager at Denver-based Irving Investors, said Wednesday's results were in line with that announcement.
"The capex spending trends tonight were confirmatory to what Jensen said," he said. "People expected big numbers and they got big numbers."
Even if it's not a bubble, the spending on AI can't continue indefinitely, said Sonnenberg. It leaves investors struggling to predict when a slowdown will come. Earlier this month, Apptopia, which measures app usage, reported that OpenAI's active users had started to flatten.
Luria says even if there is a bubble, the pain when it pops likely won't be felt by Big Tech, which have millions of customers and can use compute capacity regardless. It will be felt by other companies downstream.
"The problem is CoreWeave, Oracle, and those companies," he said. "If they get stuck with this capacity, they won't have anything to do with it. They don't have customers, they don't have an internal use of this. That's where the problems will be."
