Microsoft on Wednesday reported strong earnings, but the percentage increase in its intelligent cloud operating income has dropped off sharply.
It reported that, for Q2 of its 2025 fiscal year, the revenue increased $1.3 billion, a 14% boost. But that is the lowest percentage increase in quite some time; in the same quarter of 2024, intelligent cloud operating income grew by $3.6 billion, a whopping 40%.
Drop caused by AI investments: Microsoft
The large percentage drops are primarily due to the fact that Microsoft, like all of the major hyperscalers, invested massively in generative AI. Math also plays a key role, in that percentage increases are always going to drop as the raw numbers become increasingly large.
The numbers on Microsoft’s site, incidentally, are not current, according to a Microsoft official familiar with the investor relations (IR) group. The financials are routinely recast in the summer, the source said, adding that the numbers on the IR site had not been updated. But they were fundamentally the same, and the percentages didn’t change materially, the Microsoft official said.
“There are minor differences, but the thrust that the percentages are decreasing over time is not wrong,” the Microsoft source said.
“The primary reason [for the percentage drop] is that we are building out our AI infrastructure and building at a higher pace than our competitors,” the official said, adding, “we’re spending much more than what Google or AWS appears to be spending [on AI].”
The fact that the AI space is growing so astronomically and so fast also plays a role. Nvidia is struggling to keep up with the demand for its chips.
Microsoft is also capacity-constrained. “We have more customers wanting to buy [AI] services from us than we have the capacity to sell,” said the source, noting that the company has $298 billion in signed contracts that have not yet been fulfilled.
Forrester Principal Analyst Lee Sustar said that, although “Microsoft’s topline earnings continue to rack up big gains as the result of its AI offerings, the company’s Intelligent Cloud segment, which includes Azure and various other services, saw operating expenses spike 10%, a sign that maintaining AI momentum is getting expensive.”
“Investors and customers are watching closely to see if Microsoft can continue to deliver, after the company noted a 70% decline in gross margin for Microsoft Cloud as the result of scaling out AI infrastructure,” Sustar said.
Jason Anderson, a principal analyst with Moor Insights & Strategy, said he is not surprised by Microsoft’s falling growth percentages in the intelligent cloud.
“I cannot comment on specific numbers, but the downward trend makes total sense. Expect to see more of it,” Anderson said. “Why? It comes down to many factors, [including] fewer humans per system due to business scale and software automation, investing in and a pretty successful launch of lower power technologies such as Cobalt chips, which are ARM-based, versus Intel.”
However, Scott Bickley, advisory fellow at Info-Tech Research Group, said he didn’t see much of interest in Microsoft’s latest earnings announcement.
“To me, this is a business as usual quarterly announcement. They are still forecasting 30% growth,” Bickley said. “Microsoft is totally OK with investing now to maintain its leadership in the AI infrastructure race, as they view scale as a key deciding factor as to who will emerge as a long-term winner. To the hyperscalers, this is an arms race akin to the nuclear race during the Cold War. DeepSeek-like innovations notwithstanding, nothing stops this train right now.”
Hard pricing questions coming from CIOs
The numbers on both sides of the equation are massive, and Sustar anticipates that enterprise CIOs are soon going to be demanding better answers on why they are being charged so much.
“Commodity cloud services are enormously expensive to scale out globally,” and Google, for example, lost $1 billion in a single quarter, Sustar said. “If DeepSeek holds up, there might be a much lower price point. That gives (CIOs) more options.”
And that’s when CIOs will start asking hard questions, Sustar said. “Are we overpaying for these kinds of services? Did we end up overpaying for commodity services?”
The conversation will then move to distinguishing between routine “mundane cloud compute” services and the premium capabilities. The convenience of pouring many services into the same hyperscaler may be questioned, Sustar said. “You don’t want to end up paying the bill for an overly aggressive plan from your cloud provider.”