For Big Tech, a penny invested in AI is a penny earned… Maybe. After an indeterminate amount of time. Investors hope.
On earnings calls last week, Amazon, Google, Microsoft, and Meta reported more than $350 billion this year on capital expenditures, or longer-tail investments in a company’s future. All four told investors to expect the number to skyrocket even further next year: Microsoft said “higher,” Amazon an “increase,” Google a “significant increase,” and Meta “notably larger.”
That probably translates to more than $400 billion total for the four companies next year, according to Joe Fath, partner and head of growth at Eclipse VC.
The return on investments for these companies so far is opaque. Dedicated AI companies are burning through cash in the meantime: OpenAI reportedly hit $12 billion in annualized revenue this summer — while reportedly being on track to burn through $115 billion through 2029.
Tension over this mismatch, Fath said, is ratcheting up. There’s a “push and pull between those companies and investors,” he added. “Investors are saying, ‘Am I going to get a return on this spend?’” It’s one of the increasingly clear indicators that some parts of the AI industry are a bubble — but it doesn’t yet tell us what happens after it pops.
AI hype has remained extremely high for several years, and startup valuations have hit eye-popping numbers. OpenAI, for instance, is reportedly hoping for a $1 trillion IPO in 2026 or 2027 and planning to raise $60 billion or more.
But AI companies insist there’s still not enough money for chips, data centers, and other resources. In a Q&A with reporters at OpenAI’s annual DevDay event last month, executives repeatedly emphasized their concern over lack of compute to expand services like Sora’s video-generation AI and ChatGPT’s daily Pulse feature, and discussed the need to eventually turn a profit from such services. Amazon, Google, and Microsoft — which provide cloud services on a quickly growing scale — have “all called out being pretty capacity-constrained,” Molly Alter, a partner at Northzone VC, told The Verge.
If these claims are accurate, they indicate that simply coming up with good products won’t be enough to make AI companies profitable — because they can’t afford to scale those products to support a huge user base. Even if they’re exaggerated, the systems are incredibly costly to operate. OpenAI is still thought to be losing money on even the $200 monthly subscription tier of its ChatGPT service, thanks to the cost of running queries.
OpenAI’s rumored IPO is a perfect example of the conundrum, Alter added. The company wants to secure about 26 gigawatts of computing capacity for data centers (which translates to about $1.5 trillion at current costs, per Alter) — meaning that even with the company’s current revenue, an up to $100 billion investment from Nvidia, and other “circular deals,” Alter says she still hasn’t been able to understand how the company’s clear funding gap gets solved.
Some investors in the company are asking the same questions. Brad Gerstner, an OpenAI investor and CEO of Altimeter Capital, asked OpenAI CEO Sam Altman on his podcast Friday about how a company with $13 billion in revenue can make $1.4 trillion in spending commitments.
“First of all, we’re doing well more revenue than that; second of all, Brad, if you want to sell your shares, I’ll find you a buyer,” Altman replied. “I just… Enough.”
In past quarters, Big Tech executives have presented customizable models and AI agents as a saving grace that will surely, eventually, turn a profit — reiterating that they need to spend money to make money, including by cutting costs elsewhere and diverting the resources to AI.
Now, though, agents from OpenAI, Google, and others are in users’ hands. And though companies promise they’ll steadily improve at automating “tedious tasks,” in their current state, they’re not taking the world by storm.
Investors seemed concerned with the details of Meta’s projected expansion, and their demands for specifics weren’t always met with clear answers. “There are lots of moving pieces in the budget. It’s not baked yet. It’s still sort of in the process of coming together,” responded CFO Susan Li to one question. “We don’t have, you know, specific targets to share.”
Some investors seemed wary about whether there’s a coherent plan at all. Meta made headlines in 2025 for spending billions to lure AI engineers and researchers away from competitors for its brand-new Superintelligence team, then announced internal restructuring and layoffs soon after. Meta’s AI initiative comes on the heels of a quixotic quest for the virtual reality “metaverse,” in which it’s so far spent and lost tens of billions through its Reality Labs division. “I don’t think they’re getting any results there that would lead you to believe that that’s good spend,” Fath told The Verge, speaking about Reality Labs.
Some of the same concerns were present on other company earnings calls, with investors asking about the AI industry’s hype levels, capacity constraints, and feature adoption. On Microsoft’s earnings call, one investor asked, “Frankly, are we in a bubble?”
Even tech executives have admitted some aspects of the industry may be overblown. OpenAI’s Altman told reporters last month that there are “many parts of AI that I think are kind of bubble-y right now.” And on Microsoft’s latest earnings call, CEO Satya Nadella told investors, “I don’t think AGI as defined, at least by us in our contract, is ever going to be achieved anytime soon.” But bubbles are largely driven by sentiment and behavior, as well as fear of missing out and expertly marketed corporate narratives.
If it is a bubble, the consensus seems to be that it’s one that won’t explode the industry; rather, it’ll just lead to fewer players and more consolidation. Alter said the funding gaps within the industry keep her up at night, especially since a company’s investment in its future growth has to ideally lead to, well, real growth and profit in the end. The companies that succeed may not be the most glamorous or consumer-facing — think coding agents, customer service AI, and potentially creative content generation, rather than solely AI social networks and all-purpose chatbots.
But there’s no fighting AI FOMO, and so any bubble-y parts of the industry have no sign of slowing down just yet — but Fath said he’s watching for if or when OpenAI slows down for any reason, and the same goes for Nvidia’s data center business.
“My sense is, when a board is sitting there, they’re asking the CEO, ‘What are you doing about AI?’” Fath said. “That’s the question they’re getting. And they need to come in with an answer prepared to say how they’re going to spend on that. And if the business starts to deteriorate, and they’re not spending in AI, there’s going to be a lot of criticism of those executives… even if we really don’t know what the returns will look like right now.”
“This gets into that whole ‘FOMO’ that’s building across industries and across companies to make sure that you’re not on the wrong side of change.”
And if over-investing in AI becomes the wrong side of change? Well, at least everyone was doing it.
