Microsoft’s Backlog Grows Beyond OpenAI as Azure Revenue Climbs 48%

Microsoft reported $92.28 billion in quarterly revenue, above the $90.63 billion estimate shown on Bloomberg’s screen. Revenue from 365 Cloud was $27.78 billion, also above the $27.48 billion estimate. Azure revenue increased 48%. The shares rose more than 4% after hours.

For Bloomberg Intelligence analyst ? anurag-rana, the results left two questions for Microsoft’s analyst call. One was how much the company planned to spend on capital expenditures over the next 12 months, and whether it would raise its CapEx plans for 2027. Rana doubted Microsoft would give a full 2027 figure, but said investors would look for hints. Bloomberg’s on-screen figures put first-quarter CapEx at $39.37 billion, below the displayed estimate of $45.93 billion.

The other question was whether Microsoft’s backlog was growing through customers beyond the major AI model companies. Rana called additions from “non-frontier customers” an important metric: he wanted to see how much enterprises were committing to cloud contracts that did not come from OpenAI or Anthropic. A large backlog tied to a few AI companies would tell a different story from one supported by broader enterprise spending.

Microsoft’s total remaining performance obligations, or RPO—a measure of backlog Ed Ludlow cited—reached $277 billion. Ludlow said Microsoft reported that RPO was growing 25% excluding OpenAI. That gives investors a specific measure of growth beyond the OpenAI relationship, though Rana’s question was about additions from non-frontier customers, not only the overall backlog figure.

Microsoft has relationships with both OpenAI and Anthropic, but Rana described its relationship with OpenAI as much larger. He singled out the backlog from hosting the ChatGPT app as especially important, and distinguished it from cloud capacity used to train models.

Rana argued that ChatGPT hosting was a more durable business than training for a model company. A model developer could reduce training expenditures over the next few years, he said; he did not expect use of ChatGPT to decline in the same way. In his view, hosting therefore offered Microsoft a steadier revenue stream than infrastructure spending tied to model training.

Anthropic matters to Microsoft for a different reason. Rana pointed to Microsoft giving customers a choice of models in its products, including the option to use Claude in GitHub Copilot’s coding agent rather than only an OpenAI model. He expected that approach to extend across more applications, including Microsoft 365, which he thought could run on OpenAI, Anthropic and other third-party models.

If that happens, Rana said, Microsoft remains critical to the ecosystem because it distributes those models through its products. The distinction is between earning from a particular model company’s training needs and maintaining a role as the channel through which customers use models from multiple providers.

Microsoft’s OpenAI exposure also appeared in the earnings discussion. A Bloomberg graphic reported a negative $0.03 impact on earnings per share from the OpenAI investment. Ludlow, describing Microsoft’s figures, said the company recorded a $207 million net gain from OpenAI as a contribution to net income, equivalent to a three-cent-per-share boost.

Ludlow noted that the contribution can change from quarter to quarter.

The results arrived amid news that Microsoft was offering buyouts to 60,000 employees. Rana cautioned against reading that figure as the number of people expected to leave. He said that in a previous offer to 8,750 employees, the take rate was 30%, and that Microsoft expected a lower take rate this time.

As an illustration, Rana used a 25% take rate. That would mean about 15,000 departures; against a workforce of 223,000, he put the share at roughly 6.7%, or about 7%. His point was that the 60,000 figure described the number of employees offered buyouts, not the number of expected exits. The eventual scale depends on how many employees accept.