Nearly four years into the AI boom, we still know little about what returns the tech giants could be making on their enormous bets on AI infrastructure.
Why it matters: The data center buildout — the engine driving markets and economic growth — hinges on the expectation that it will eventually enable the tech giants to generate big profits. But earnings reports have yet to provide any confirmation.
How it works: Amazon, Alphabet, Microsoft and Meta don't break out their sales and profits directly attributable to AI data center investments. Instead, results are embedded into parts of their respective cloud computing units: Amazon Web Services, Microsoft's Intelligent Cloud — home to Azure — and Alphabet's Google Cloud. Meta doesn't have a cloud business and largely uses its AI investments to goose its existing businesses.
Follow the money: Operating profit margins from those cloud businesses could offer hints about data center profitability. It's difficult to say, however, because these cloud units also contain other highly profitable, non-AI computing businesses.
The latest: Profit margins at Amazon Web Services did rise a lot in the second quarter, about 6.5 percentage points, to 39%, from the same time last year. Still, Amazon executives demurred from tying the uptick to AI and seemed to play down the durability of those margins. Alphabet's operating margin at its Cloud division, likewise, jumped to 35.6% in the second quarter from 20.7% a year ago. Executives there also warned that those margins might not last, as they will be adding costly capacity in coming quarters that "will put some pressure on operating margins for cloud." Meanwhile, Meta's overall operating margin collapsed by more than 12 percentage points, to 31% from the same quarter a year ago, which the company largely blamed on ballooning legal expenses and severance costs. Without those costs, its operating margin would have been 36.8%, but even that is below the 43% seen during the same quarter last year. Then there's Microsoft, whose Intelligent Cloud division — which includes Azure — saw its operating profit margins stay flat at 41%, despite huge expenses associated with AI investment. That implies that the company could see an AI profit surge if those investment costs slow down.
Reality check: It makes sense that no AI business is crazy profitable yet. These companies are spending roughly $800 billion combined this year on the AI buildout, if you include spending from Oracle. "These companies are spending this much on capex, and so the margins that Microsoft and Amazon and Google and Oracle are getting off AI are meaningfully less than traditional cloud," said Rishi Jaluria, an analyst with RBC Capital Markets.
What they're saying: Analysts say neither they, nor investors, are particularly worried about the lack of near-term profitability for AI. But when that lack of profits comes alongside plans to keep on building out data centers at a breakneck pace, they do get a bit twitchy. "What people are concerned about is: Are you building too much capacity?" said Jason Helfstein, head of internet research at Oppenheimer & Co. "If the world builds too much of it, the price is going to go down."
Between the lines: There is another worry. An uncomfortably large amount of the hyperscalers' AI sales appear to come from just two companies: Anthropic and OpenAI, which are reliant on significant continuing investments to keep spending. Stephen Bersey, head of U.S. technology research at HSBC, estimates that around 50% of AI-related backlogs disclosed by Oracle, Amazon, Microsoft and Alphabet represent orders from the two. Others cite even higher percentages. "Considering both OpenAI and Anthropic are currently private, there is significant uncertainty in assessing customer concentration risk for hyperscalers," Bersey wrote.