Azure Reaccelerates to 35% — Demand Still Exceeds Supply, and the Capex Question Just Got Quieter

Results vs. Consensus

Quality of the Beat

• Revenue: Almost entirely organic. Constant-currency growth of 15% versus 13% reported implies a ~2-point FX headwind. The $70.1B print is a clean broad-based beat — all three segments were ahead of expectations.
• Azure decomposition: Of the +35% cc growth, 16 percentage points came from AI services and 19 from non-AI. The decisive surprise was the non-AI side: Hood explicitly said "the real outperformance in Azure this quarter was in our non-AI business" — enterprise-segment acceleration plus improvement in scale motions. AI services were "ahead of where we had guided to" but only modestly, with the upside coming from delivering supply earlier than planned, not from new demand.
• Margins: Operating margin of 45.7% (+1pt YoY) was driven by Q4-shifted opex and continued discipline on layers/managers. Microsoft Cloud GM of 69% held against a 3pt YoY drag from AI infra — the AI margin headwind is real but contained. Capex came in at $21.4B versus management's H2 framing implying slightly higher; Hood attributed the variance to data-center lease delivery timing rather than reduced spend.
• EPS: $3.46 GAAP (+18% YoY) versus $3.22 consensus on a roughly 18% effective tax rate. Share count is steady; the EPS beat is operational, not financial-engineering.
• Cash flow: Operating cash flow of $37B (+16%) easily funded $16.7B of cash capex, leaving FCF at $20.3B and supporting $9.7B of returns to shareholders. The cash-flow math comfortably accommodates the ramp.

Segment Performance

Productivity & Business Processes ($29.9B, +10% / +13% cc)

P&BP delivered a clean broad-based beat with M365 Commercial Cloud +12% reported / +15% cc, in line with expectations but with the right composition: paid M365 commercial seats grew 7% YoY to 430M+, with growth led by SMB and frontline-worker offerings. ARPU expansion from E5 and M365 Copilot is doing the rest of the work. The disclosure that "hundreds of thousands of customers" now use M365 Copilot, with Copilot deal sizes "continuing to grow" and a "record number of customers returning to buy more seats," is the clearest indication yet that Copilot is moving past the pilot phase.

"Microsoft 365 Copilot is built to facilitate human agent collaboration, hundreds of thousands of customers across geographies and industries now use Copilot, up 3 times year-over-year. Our overall deal size continues to grow. In this quarter, we saw a record number of customers returning to buy more seats." — Satya Nadella, CEO

LinkedIn revenue growth of 7% / 8% cc was ahead of expectations across all businesses but Hood flagged Talent Solutions as still pressured by hiring-market weakness. Dynamics 365 +16% / +18% cc was in line, with Verizon called out as a recent migration win from a legacy provider. Segment operating income +15% / +18% cc with margin essentially flat YoY — impressive given the AI infra drag.

Assessment: P&BP is the quiet compounder of the print. Copilot seat growth at "3x YoY" is the metric to track into Q4; if it converts into ARPU re-acceleration in M365 Commercial Cloud the segment has another leg of growth.

Intelligent Cloud ($26.8B, +21% / +22% cc)

This is the segment that defines the print. Azure and other cloud services +33% reported / +35% cc beat the StreetAccount consensus of ~31.5% by ~350bps, and the 16-point AI contribution disclosure is the highest the line has been to date. The decomposition matters more than the headline: Hood's framing was that the real upside was on the non-AI side. Enterprise acceleration plus improved scale-motion execution drove the beat; AI services performed "in line with what we had guided to," with the marginal benefit coming from delivering capacity earlier than planned rather than from new bookings.

"In Azure and other cloud services, revenue grew 33% and 35% in constant currency, including 16 points from AI services. Focused execution drove non-AI services results where we saw accelerated growth in our enterprise customer segment, as well as some improvement in our scale motions. And in Azure AI services, we brought capacity online faster than expected." — Amy Hood, CFO

On-prem server revenue declined 6% reported / 4% cc, a touch below expectations on contract-mix in-period revenue recognition. Enterprise & Partner Services +5% / +6% cc was modestly ahead. Segment operating income +17% / +18% cc with gross margin down 4pts YoY on AI infra scaling — the unit-economics drag is visible but not punitive at this scale.

Assessment: The non-AI Azure reacceleration is the most strategically important data point in the report. It says Microsoft's hyperscaler is winning workloads that are not directly AI-related — migrations, data, cloud-native consumption — which is exactly what funds the AI infra build without forcing every dollar of capex to clear an AI-only ROI bar.

More Personal Computing ($13.4B, +6% / +7% cc)

MPC was the quiet upside. Windows OEM and Devices +3% reported was ahead of expectations on tariff-driven inventory builds — Hood flagged that elevated OEM channel inventory is expected to come down in Q4, which is the source of the wider-than-usual MPC guidance range. Search and news advertising ex-TAC +21% / +23% cc was the standout, driven by a third-party partnership (almost certainly meaningful revenue from the OpenAI/Bing relationship plus better Edge/Bing rate and volume). Xbox content and services +8% / +9% cc beat on third-party and first-party content; PC Game Pass +45% YoY is doing real work.

Operating income +21% / +23% cc with gross margin up 2pts YoY — Hood credited "strong execution on margin improvement in search and gaming." MPC has quietly become a profitability story as Microsoft prunes lower-margin device businesses.

Assessment: MPC continues to outpunch its weight. The Q4 guide implies a step-down in revenue ($12.35-12.85B vs. $13.4B printed) on Windows OEM channel-inventory normalization; the segment will look weaker on the surface in Q4 but the margin story should hold.

Key KPIs

The single most important number on this page is the $315B commercial RPO with 47% YoY growth in the beyond-12-month tranche. That is contracted multi-year backlog — it is not "demand signal," it is signed customer commitment. The 12-month-recognizable tranche grew 17% YoY, which is the better Azure-revenue-growth leading indicator than any single quarter's print.

Key Topics & Management Commentary

Overall Management Tone: Confident but disciplined. Nadella's prepared remarks were unusually heavy on engineering specifics (dock-to-live times down 20%, AI performance per ISO power up 30%, cost-per-token "more than halved"), which reads as a deliberate response to the spring's "Microsoft is walking away from data centers" press narrative. Hood was crisp on the numbers and refused to extrapolate FY26 capex beyond reaffirming "lower rate than FY25." The throughline of the call was "demand still exceeds supply" — which after a quarter of bear narratives about AI overbuild is the headline.

The "Walking Away From Data Centers" Narrative — Refuted on the Call

The first analyst question, from Morgan Stanley's Keith Weiss, was directly about media reports of changing data-center commitments and Nadella's prior comments on potential GPU oversupply. Nadella's answer was emphatic: Microsoft has always tuned its build/lease pace to forward workload growth, and the calibration is being made tighter — not pulled back. The decisive disclosure was Hood's: Microsoft will be "a little tight" on AI capacity exiting Q4, having previously expected to be in supply/demand balance by then. That is a longer demand runway, not a shorter one.

"In fact, Amy just mentioned, we will be short power. And so therefore — but it's not power, but it's not a blanket statement. I need power in specific places so that we can either lease or build at the pace at which we want." — Satya Nadella, CEO

"I did talk about in my comments, we had hoped to be in balance by the end of Q4. We did see some increased demand as you saw through the quarter. So we are going to be a little short, a little tight as we exit the year but are encouraged by that." — Amy Hood, CFO

Assessment: This was the most important exchange of the call. The "AI capacity constraints beyond June" disclosure directly contradicts the bear narrative that hyperscaler AI capex was overshooting demand. It is the single piece of evidence that should give pause to anyone who has been short the AI-infra leaders into the spring. The operative question is no longer "will the AI demand show up" but "how long does the supply-constrained period last."

Capex Trajectory and the FY26 Setup

Q3 capex of $21.4B (cash $16.7B) came in slightly under guide on data-center lease delivery timing. Hood reiterated that H2 capex in total is unchanged from January's framing, with Q4 stepping up sequentially. The forward-year framing was unchanged: FY26 capex will grow but at a slower rate than FY25, with a higher mix of "short-lived assets" — code for GPUs and CPUs that more directly tie to revenue. Roughly half of Q3's cloud-and-AI spend went to long-lived assets supporting 15+ years of monetization.

"We expect CapEx to grow. It will grow at a lower rate than FY 2025 and will include a greater mix of short-lived assets, which are more directly correlated to revenue than long-lived assets." — Amy Hood, CFO

On the "DeepSeek moment" software-efficiency thesis, Nadella's framing was the most concise on the call: the difference between hosting and hyperscale is software, and every cloud generation has been driven by software efficiency. He explicitly cited cost-per-token having "more than halved" and AI performance per ISO power up nearly 30%. Hood's complementary point — Microsoft's AI-side margins "are better than they were at this point by far" than during the server-to-cloud transition — is the cleanest defense of the capex-to-margin trajectory.

Assessment: The FY26 capex setup is the cleanest it has been in 18 months. "Growth, but at a lower rate" with a shift toward shorter-life assets translates to lower long-term-asset depreciation pressure on margins and tighter coupling between capex and revenue. The bear case that capex compounds without margin discipline is harder to make after this print.

Azure AI vs. Non-AI Decomposition — Hood's Most Important Clarification

Goldman's Karl Keirstead asked whether the 16-point AI contribution to Azure could rise into Q4. Hood's answer was almost surgical in its precision: the upside in Q3 came primarily from non-AI, with AI services performing close to plan plus a benefit from delivering capacity early. This is a meaningful framing — it pushes back against the implicit assumption that Azure's growth is now AI-driven. The platform is still winning conventional cloud workloads, with AI services as a separately tracked line that is supply-constrained.

"First, the real outperformance in Azure this quarter was in our non-AI business... The only real upside we saw on the AI side of the business was that we were able to deliver supply early to a number of customers." — Amy Hood, CFO

Hood reinforced the point in the answer to Wolfe Research's Alex Zukin: "It's getting harder and harder to separate what an AI workload is from a non-AI workload" — workloads use shared GPU/CPU/storage/network infrastructure, and digital natives often run AI and non-AI workloads on the same cloud. The implication is that Microsoft is consciously moving away from segmenting Azure into AI vs. non-AI on the call. That is a strategic disclosure choice — it reduces visibility but accurately reflects the converged nature of the workload mix.

Assessment: The non-AI Azure reacceleration is what makes this an Outperform print rather than a Hold. Single-stack hyperscalers that are gaining share in conventional cloud AND running supply-constrained AI services have a wider moat than the bear case has been pricing.

Copilot Adoption and the Move to Agents

The agent-platform narrative is the second strategic theme of the call. Discrete data points: 230,000 organizations (90% of Fortune 500) have used Copilot Studio; 1M custom agents created across SharePoint and Copilot Studio in Q3 alone (+130% QoQ); 70,000 enterprises using Foundry; over 10,000 organizations using the new agent service in its first four months; 100T tokens processed in Q3 (+5x YoY); 50T tokens last month alone. Dragon Copilot documented 9.5M physician-patient encounters in Q3 (+50% QoQ). GitHub Copilot at 15M users (+4x YoY) with the new agent mode in VS Code is the developer-side equivalent.

"We're evolving GitHub Copilot from pair to peer programmer with agent mode in VS Code, Copilot can now iterate on code, recognize errors and fix them automatically." — Satya Nadella, CEO

Assessment: The agent-platform pivot is real and showing up in usage metrics that compound. The 1M agents in a single quarter from Copilot Studio is the kind of number that is small in revenue today but defines a new monetization surface in 18-24 months. M365 Copilot moving to "record returning customers buying more seats" is the leading indicator that ARPU expansion in P&BP has another leg.

Recession-Resilience Posture

Bernstein's Mark Moerdler asked about recession sensitivity. Nadella's framing was that Microsoft's value proposition becomes more compelling in macro stress — "software is the most malleable resource we have to fight any type of inflationary pressure" — and the company would expect to gain share rather than retrench. The 98% annuity mix and $315B RPO support that posture mechanically: most of the revenue is contracted, and the longer-dated tranche is growing 47% YoY.

Assessment: Microsoft has the cleanest defensive posture of any large-cap tech franchise. Annuity-heavy revenue plus mission-critical workload concentration make this a natural quality-flight name in a tariff-driven slowdown.

Guidance & Outlook

The Q4 guide is materially above the LSEG consensus on the headline ($73.7B midpoint vs. ~$72.26B Street). The Azure +34-35% cc guide is essentially a flat-to-slight-deceleration off the Q3 +35% print, and notably lands at the high end of what the Street had been pricing — the AI capacity constraint commentary is the structural reason it does not decelerate further.

The forward FY26 commentary deserves attention. Hood reaffirmed: capex will grow but at a lower rate than FY25, with a higher mix of short-lived assets. That is the cleanest forward signal investors have had since the spring's data-center-walkaway narrative began. The fact that operating margins are guided "up slightly YoY" for FY25 despite the largest single-year capex absolute step-up in company history is the single best evidence that the AI cycle is not breaking margins.

Implied Q4 trajectory: Midpoint guidance of $73.7B implies ~5% sequential revenue growth. Azure at 34-35% cc on a much larger base than Q3 is the demand-bridge to the +13-14% total revenue growth.

Street at: Pre-print LSEG consensus was ~$72.26B for Q4; Microsoft guided to a $73.7B midpoint, ~2% above. Azure consensus was tracking to ~32-33% cc; the 34-35% cc guide is ~2-3pts above.

Guidance style: Microsoft historically guides modestly conservative, with Hood's commentary that "if the environment changes, our results may be impacted" reading as standard tariff-environment caveat. The Azure guide is unusually aggressive relative to the prior-quarter base — this is a confidence print.

Analyst Q&A Highlights

Data Center Strategy & Capacity

• Keith Weiss, Morgan Stanley: Asked about reported changes to data-center commitments and Nadella's prior comments on GPU-oversupply risk. Nadella said adjustments to build/lease pace are routine and tied to forward workload demand; the "short power" and "tight" exit-Q4 capacity framing was the headline.
  Assessment: Direct refutation of the spring's "Microsoft walking away" press narrative. The most important exchange of the call.

Cloud Migration & Workload Mix

• Brent Thill, Jefferies: Asked Nadella to elaborate on accelerating cloud-migration demand. Nadella outlined four trend lines: classic SQL/Windows Server migrations, data growth (PostgreSQL, Cosmos, Fabric), cloud-native compute consumption, and AI workloads — with explicit ratios between AI and non-AI consumption per workload.
  Assessment: Useful framing. The "AI ratio" point — that AI workloads also drive Cosmos/Postgres/storage consumption — is the mechanical reason Azure's non-AI growth is reaccelerating alongside AI services.

Recession Sensitivity

• Mark Moerdler, Bernstein: Asked about recession sensitivity given how much Microsoft's business has changed. Nadella argued software is the most malleable cost-out resource and Microsoft would lean into customer support during stress, with share gains as a likely byproduct.
  Assessment: The 98% annuity mix and $315B RPO support this posture mechanically. Defensive positioning is real.

Azure AI vs. Non-AI

• Karl Keirstead, UBS: Asked about the 16-point AI contribution to Azure and whether it could rise. Hood clarified that Q3 outperformance was on the non-AI side; AI was in line with guide plus an early-supply benefit.
  Assessment: The single most important clarification on the call. Reframes Azure's growth as broader-based than the AI-services line alone.
• Kirk Materne, Evercore: Asked about the non-AI Azure outperformance and how this cycle differs from the optimization-pressure period a few years ago. Hood pointed to enterprise-segment acceleration and improved scale-motion execution.
  Assessment: The "scale motions" reference is to SMB and mid-market — Microsoft has been investing here and Q3 is the first quarter where it is showing up in the numbers.

Capital Efficiency & Useful-Life Assumptions

• Kash Rangan, Goldman Sachs: Asked whether Microsoft can slow capex growth while accelerating Azure including AI. Hood and Nadella both pointed to compounding software/hardware/build-out efficiencies. Hood said AI margins are "better than they were at this point by far" relative to the server-to-cloud transition.
  Assessment: Hood's comparison to the prior cloud transition is the cleanest defense of unit economics. This is a credible answer.
• Mark Murphy, JPMorgan: Asked whether software efficiencies could extend GPU useful-life assumptions. Hood was firm — long history needed before changing depreciable-life assumptions; current focus is on getting full useful life from existing assets.
  Notable: This is the right disclosure posture. A useful-life extension would be a discrete EPS tailwind, but Hood is unwilling to pre-commit.

Workload Convergence

• Alex Zukin, Wolfe Research: Asked how non-AI Azure growth is being pulled by AI workloads, and how test-time-compute changes the AI growth curve. Hood emphasized that AI/non-AI workloads share underlying compute/storage/network and that the segmentation is increasingly artificial.
  Notable: Hood is signaling that future disclosures may move away from the 16-point AI-contribution metric — workloads are converging.

What They're NOT Saying

1. FY26 capex absolute number: Hood reaffirmed "growth at a lower rate than FY25" but declined to quantify. With FY25 cash-paid capex tracking to ~$80B+, "lower rate" still implies a meaningfully larger absolute number — the market will price the actual figure when guided in late July.
2. Specific OpenAI consumption details: The "Azure commitment from OpenAI" was disclosed as a driver of bookings strength but not quantified. The strategic interdependence remains the single largest concentration risk in Azure's growth profile and management is deliberately opaque on it.
3. Future Azure disclosure granularity: Hood twice indicated AI vs. non-AI workload separation is "getting harder." The implicit signal is that the 16-point AI-contribution disclosure may be discontinued or de-emphasized — this would reduce visibility into a key growth driver.
4. Copilot revenue contribution: Nadella disclosed adoption metrics ("hundreds of thousands of customers," "+3x YoY") but not revenue. The ARPU contribution to M365 Commercial Cloud growth is opaque, and management has chosen not to quantify it. This is appropriate while pricing is still being optimized but means modeling Copilot revenue requires assumptions.
5. GPU useful-life extension: Mark Murphy's question was a direct prompt and Hood declined to commit. A 1-year extension on the GPU useful-life assumption would be a multi-billion-dollar EPS tailwind; management is keeping it as discretionary upside.
6. Tariff-impact magnitude: The MPC guide was widened due to "elevated OEM channel inventory" — the unspoken cause is tariff-driven inventory pull-forward in Q3. Management did not quantify the Q4 channel-correction risk.

Market Reaction

• After-hours move (April 30, 2025): +9% in extended trading.
• Next-day close (May 1, 2025): +7.6% on the regular session.
• Volume: Elevated, in line with prior earnings prints.
• Drivers cited: Azure +35% cc beat by ~350bps, Q4 Azure guide of 34-35% cc, "demand exceeds supply" capacity framing reassured the AI-overbuild bear case, $315B RPO with 47% growth in long-dated tranche, and the cleanest beat across all three segments.

The +7.6% next-day move is rational: this is a print where the headline beat (Azure +35% cc) cleared expectations by a clean margin, and where the forward guide (Azure 34-35% cc into a much larger base) reframed the AI-capex question from "will demand show up" to "how long does the supply-constrained period last." Sell-side reception was uniformly constructive on the call, with every Q&A opener some variant of "great quarter."

Street Perspective

Debate: Is the Capex Magnitude Sustainable?

Bull view: $315B contracted RPO with 47% growth in the beyond-12-month tranche is the demand anchor. Hood explicitly framed FY26 capex growth as slower than FY25 with more short-lived assets — exactly the discipline framework the bear camp had been demanding. AI-side margins are tracking better than the server-to-cloud transition at the equivalent stage.

Bear view: "Lower rate of growth" still means a larger absolute capex number, with FY25 cash capex tracking to $80B+. The 3-point Microsoft Cloud GM compression is real and AI infra depreciation has not yet fully ramped. OpenAI dependency is unquantified but material.

Our take: The bull case has cleaner near-term evidence. The shift toward short-lived assets in FY26 mechanically improves the capex-to-revenue elasticity. We accept the FY26 capex print will be uncomfortable in absolute terms but expect the deceleration framing to hold.

Debate: Is the AI-Demand Story Real or a Bubble?

Bull view: Hood's "AI capacity constraints beyond June" disclosure walks back the prior plan to be in supply/demand balance by year-end. That is unambiguous evidence that demand is outpacing supply at the largest hyperscaler. The "Microsoft walking away" press narrative was wrong, and the FY26 mix-shift toward short-lived (revenue-tied) assets says management has high conviction in monetization runway.

Bear view: The "DeepSeek moment" software-efficiency thesis is real and could compress the AI compute trade structurally. Hood acknowledged on the call that cost-per-token has "more than halved" in a single year. If model efficiency outpaces model demand, capex returns get squeezed.

Our take: Nadella's framing — that every cloud transition has been driven by software efficiency, and efficiency drives demand rather than caps it — is the right historical analogy. We side with the bull case but acknowledge the risk window is real.

Debate: How Much Should Microsoft Trade Above the Hyperscaler Average?

Bull view: Microsoft is the only hyperscaler with a converged AI infrastructure / SaaS application / developer-tools / agent-platform stack. The cross-sell economics (Foundry into M365 Copilot into Copilot Studio into Power Platform) are unique and compound across surfaces.

Bear view: Multiple is already at the upper end of the historical range. Even with execution this clean, valuation already prices much of the AI optionality.

Our take: Premium is earned on the breadth of stack and the cleanliness of the print. We do not see a re-rating catalyst from here, but we see continued earnings power that supports the valuation.

Model Update Needed

Valuation impact: Modest upward revision to NTM revenue and EPS, with a smaller upward revision to FY26 Azure growth and a notable improvement in the FY26 capex framing. Net intrinsic-value impact is positive; multiple-driven upside is more limited.

Thesis Scorecard Post-Earnings

Overall: Thesis strengthened. Two of three bear points were directly challenged by the print; the bull side has cleaner near-term evidence than at any point in the past 12 months.

Action: Initiating coverage at Outperform. Monitor (1) Q4 print for Azure deceleration vs. the 34-35% cc guide, (2) FY26 capex framing on the July call, (3) Copilot revenue disclosure if and when management chooses to quantify it.