Microsoft Corporation (MSFT)

News Summary

• Azure grew 40% YoY in Q3 and is guided at 39-40% for Q4 with 'modest acceleration' in H2 CY2026; management explicitly stated demand continues to exceed available capacity and expects to remain capacity-constrained at least through 2026, with over 300 customers on track to process more than 1 trillion tokens on Azure AI Foundry this year, up 30% QoQ — concrete evidence that inference workload scale is outpacing hyperscaler supply.
• Microsoft's first-party Maia 200 AI accelerator is now live in Iowa and Arizona data centers, delivering over 30% improved tokens per dollar versus the latest third-party silicon in their fleet, and Cobalt CPUs are deployed in nearly half of all DC regions — a direct signal of accelerating vertical integration in the AI chip stack that intensifies competitive pressure on third-party silicon vendors.
• Microsoft's AI business hit $37 billion ARR growing 123% YoY; Microsoft 365 Copilot commercial seats grew from 15 million in January to over 20 million with weekly engagement now at Outlook-level, and CEO Nadella restructured the OpenAI relationship to secure royalty-free frontier model IP rights through 2032 while opening Azure exclusivity to all cloud providers.
• Revenue of $82.9B (+18% YoY) and EPS of $4.27 both beat consensus; however, full-year 2026 capex guidance of ~$190B — up 61% from 2025 and $35B above Visible Alpha consensus — reflects a $25B component cost inflation impact, while gross margin compressed to 67.6%, the narrowest since 2022, as depreciation from the infrastructure buildout accelerated.
• MSFT stock fell 4.87% post-earnings despite the beat, as below-consensus Q4 revenue guidance ($87.25B midpoint vs. $87.53B expected) and margin compression weighed on sentiment; Scotiabank cut its price target citing a valuation reset, and the stock entered the print already down 12% YTD following its worst quarterly performance since 2008.

Financial Highlights

Earnings Call Highlights

Microsoft's Q3 FY2026 call was defined by a tone of confident acceleration, with management projecting near-zero anxiety about AI ROI despite massive capital commitments, and analysts broadly accepting the narrative rather than pressing hard on risk. The dominant theme was the transition from seat-based to seat-plus-consumption business models across every AI product layer, with Satya Nadella and Amy Hood repeatedly framing this as a structural TAM expansion rather than a cannibalization risk. Management posture was unusually bullish on infrastructure returns, citing demand that continues to exceed supply and a $627 billion RPO as validation. Analyst sentiment was constructive, with questions focused on modeling the transition rather than challenging the underlying demand thesis. The call reinforced that Microsoft views its AI infrastructure buildout as a compounding competitive moat, not a speculative bet.

• Microsoft's AI business surpassed $37 billion ARR, growing 123% year-over-year, and Azure grew 40% in constant currency with management guiding 39-40% for Q4 and 'modest acceleration' in the second half of calendar 2026, explicitly stating demand continues to exceed available capacity.
• CapEx is guided to over $40 billion in Q4 and approximately $190 billion for full calendar year 2026, with Amy Hood noting roughly $25 billion of that reflects higher component pricing — a direct signal of GPU and CPU cost inflation pressuring the supply chain that Cerebras competes within.
• Microsoft's first-party Maia 200 AI accelerator is now live in Iowa and Arizona data centers, delivering over 30% improved tokens per dollar versus the latest third-party silicon in their fleet, and Cobalt CPUs are deployed in nearly half of all DC regions — signaling aggressive vertical integration in the AI chip stack.
• Over 300 customers are on track to process more than 1 trillion tokens on Azure AI Foundry this year, accelerating 30% quarter-over-quarter, with the number using Anthropic and OpenAI models together doubling sequentially — a concrete adoption metric illustrating the scale of inference workloads now running on hyperscaler infrastructure.
• Amy Hood disclosed that Microsoft expects to remain capacity-constrained 'at least through 2026,' and that the mix of short-lived assets (primarily GPUs and CPUs) represented roughly two-thirds of Q4 CapEx — underscoring that the bottleneck is compute hardware availability, not customer demand.

Sources

• Microsoft (MSFT) Q3 earnings report 2026 - CNBC
• FY26 Q3 - Press Releases - Investor Relations - Microsoft
• As AI war rages, Google reports faster growth in Cloud business than Microsoft and Amazon