MSFT Spot and Perp Total Returns

Quarter in one view

• Revenue $90.0B (+18%, +17% cc); EPS $4.74 (+23% adjusted for OpenAI); operating income +18%; operating margin 45%, up slightly YoY — "exceeded expectations across revenue, operating income, and earnings per share" after adjusting for discrete items. FY26 full year: revenue >$331B (+18%), operating income >$155B (+21%), Microsoft Cloud >$214B (+27%), Azure surpassed $100B (+41%).
• Discrete items added $0.27 to EPS: a $3.2B gain on the Anthropic investment and lower-than-expected Voluntary Retirement Program costs, partially offset by severance and Xbox impairment charges (third consecutive quarter with gaming impairments). Adjusted other income +$2.8B ex-OpenAI.
• Azure +43% (vs 39–40% cc guide) — a large beat, driven by fleet efficiency gains, faster capacity delivery (dock-to-live times down ~50% over the fiscal year), and stronger-than-expected GitHub Copilot consumption after the June usage-based pricing change. Q1 FY27 guided to ~45% cc with H1 acceleration still expected. Demand still exceeds supply.
• RPO $678B (+84% incl. OpenAI; +25% ex-OpenAI); duration 2.3 years; ~30% recognized in next 12 months (+37%); beyond-12-months +112%. Notably: all sequential RPO growth came from customers outside frontier model companies. Bookings +10% (+11% cc) incl. OpenAI; +18% ex-OpenAI — a sharp re-acceleration from +7% last quarter.
• Capex $41B (in line with >$40B guide); FCF $19.6B on OCF $55.4B (+30%) — second consecutive quarter of FCF recovery. Finance leases $5.6B; cash PP&E $35.8B. $10.2B returned to shareholders.
• Accounting change with real disclosure impact: useful life of data centers/buildings extended 15→25 years starting FY27, shifting leases from finance to operating — CY2026 capex expectation restated from ~$190B to ~$175B with underlying investment "unchanged." Q1 FY27 capex guided to >$50B.
• Cloud GM 65%, better than expected (vs ~64% guide) — the step-down sequence paused. Company GM 67%, down YoY.
• M365 Copilot crossed 30M paid seats (from 20M), net adds more than doubled QoQ; E7 already at "millions of seats" two months post-launch with EY at 400,000 seats. GitHub Copilot revenue accelerated >60% QoQ on usage-based billing.

What management is focused on

• Model independence as architecture and ideology: Nadella's dominant theme — harness/context/memory separated from any model family so "every model is substitutable"; firms must build "their own learning machine" and not outsource core IP. 11,000+ models in the catalog; 5x increase in multi-provider customers since the start of the year; Mistral added to Sovereign Cloud. This is both a product stance and a hedge against OpenAI concentration.
• First-party models as COGS weapons, now with quantified proof points: MAI-Thinking-1 (first reasoning model) plus a dozen others; MAI-Code-1-Flash in GitHub Copilot (higher acceptance, 10% lower median tokens) and Excel (comparable to GPT-5.6 on common tasks at significantly lower cost); 89% GPU cost reduction in D365 (MAI-Voice-2-Flash), up to 84% in PowerPoint (MAI-Image-2.5); MAI-Cyber-1-Flash matching a much larger "Mythos" model at half cost. MAI on Maia 200: 40% better performance/watt.
• Efficiency as the Azure acceleration engine: Hood repeatedly attributed the beat to CPU/GPU fleet efficiency and process improvements that are "quickly monetized in quarter" given the supply-demand imbalance — a deliberate reframing of acceleration as execution, not just capacity adds.
• The Copilot "super app" consolidation: chat + Cowork (now GA with usage-based billing) + Autopilots (long-running autonomous agents, "powered by OpenClaw") + Code into one flagship app spanning consumer and commercial — "a major step forward," more detail promised soon.
• E7 as the monetization vehicle: Copilot + E5 + Entra + Agent 365 in one SKU; Hood emphasized Agent 365's token-spend observability/manageability (SecOps/FinOps) as the E7 value hook.
• Microsoft Frontier Co.: new outcome-driven engineering organization — 6,000 embedded experts, 330+ projects across 164 customers tested over the past year (Novo Nordisk, LSEG cited). A services-flavored motion to convert AI into measured customer outcomes.
• Xbox reset with a deadline: "necessary decisions... to reset the business for long-term growth," with an explicit commitment to return to growth in FY27 — despite another impairment this quarter.
• Windows as "unmetered intelligence" offload: on-device AI compute as the strategic answer to metered cloud inference costs.

Key numbers and quarter mechanics

• Total: revenue $90.0B (+18%/+17% cc); company GM 67% (down YoY on Azure mix, AI infrastructure, and product usage, partially offset by efficiency); opex +10% (R&D compute, talent, data; low prior-year G&A comp); op margin 45%; headcount -2% YoY. FX roughly in line.
• Cash flow: OCF $55.4B (+30%, cloud billings/collections, partially offset by higher operating lease payments); capex $41B (~2/3 short-lived); finance leases $5.6B (large DC sites); cash PP&E $35.8B; FCF $19.6B; $10.2B returned; FY26 total cash return >$43B.
• Bookings/RPO: bookings +10% (+11% cc) incl. OpenAI, +18% ex-OpenAI; RPO $678B (+84% incl., +25% excl.); duration 2.3 years (from 2.5); ~30% recognized within 12 months (+37%); beyond-12-months +112%; all sequential RPO growth from non-frontier-model customers; FY26 cloud revenue ~90% from non-frontier-model customers.
• P&BP $37.8B (+14%): M365 commercial cloud +16% adjusted (normalizing 2 pts prior-year in-period recognition), +14% reported; commercial products +19% (ahead — large long-duration contracts, Windows Commercial on-prem in-period recognition); consumer cloud +24% (+22% cc), subs +7%; LinkedIn +12% (+10% cc); D365 +13% (+12% cc) — ERP bookings healthy, CRM moderating with longer sales cycles. Segment op margin 58% (down from 60%).
• IC $39.3B (+32%/+31% cc): Azure +43%; on-prem roughly flat (-1% cc), ahead on renewal in-period recognition. Segment op margin 41% (up from 40%), "relatively unchanged YoY."
• MPC $12.9B (-4%/-5% cc): Windows OEM & Devices -7% (OEM -5%; lower PC demand, EOS comp; ahead of the high-teens-decline guide because OEMs/channel again built inventory ahead of component price increases); search ex-TAC +10% (+9% cc, third-party partnership drag); Xbox -10% (-11% cc); content/services -10%. Xbox impairment charges in opex again. Segment op margin 21% (down from 28%).
• Q1 FY27 guide (USD): revenue $89.85–90.95B (+16–17%); COGS $29.6–29.8B (+23–24%); opex $16.8–16.9B (+7–8%); op margins roughly flat YoY; other income ~-$100M ex-OpenAI; tax ~20%; FX <1 pt headwind. Cloud GM relatively stable QoQ.
• Segment guides: P&BP $36.7–37.0B (+11–12%); M365 commercial cloud ~16% cc adjusted / 15% reported, with acceleration expected through FY27 on E7/premium SKUs and July usage-based billing additions; commercial products mid-single digits; consumer cloud mid-teens (lapping price increase); LinkedIn high single digits; D365 low teens (stable). IC $40.95–41.25B (+33–34%); Azure ~45% cc, H1 acceleration expected; on-prem down low-to-mid single digits. MPC $12.2–12.7B; Windows OEM & Devices down low 20s ("wider than normal" range again); search ex-TAC mid-single digits (partnership impact); Xbox content/services down mid-single digits; hardware down.
• FY27: double-digit revenue and operating income growth; opex mid-to-high single digits; operating margins down less than 1 point (a posture shift from FY26's "up ~1 point"); capex up YoY; FCF positive; tax ~20%; M365 commercial products and Server products both down mid-single digits for the year; Windows OEM & Devices down high teens for the year.

Product and launch scorecard

• M365 Copilot — step-change quarter: >30M paid seats (from 20M); net adds more than doubled QoQ (vs +250% YoY adds last quarter); customers with 50,000+ seats up >7x YoY (from "quadrupled"); enterprises deploying to majority of information workers +~75% QoQ. Named wins: NHS England 505,000 (largest healthcare deployment; trial showed 43 min/day saved), KPMG 276,000+, HSBC 200,000, plus 11 named customers at 60,000+ each. Usage: conversations/user ~doubled YoY; satisfaction scores doubled over 3 quarters; latency -25% in-quarter; time-to-high-usage from months to days. Still no Copilot revenue dollar figure.
• E7 — fast early traction: launched ~2 months ago; hundreds of enterprise customers, millions of seats; EY 400,000 employees = largest win to date. First real evidence the integrated AI suite lands.
• Cowork: GA last month; usage-based billing added in July with thousands of customers already paying — the seats+consumption thesis now has a second live billing surface.
• GitHub Copilot — the usage-pricing bet paid immediately: 50M users; revenue accelerated >60% QoQ after the June usage-based change, with continued business/enterprise seat growth plus "significant consumption revenue"; named as a positive contributor to the Azure/IC beat. GitHub overall: 225M users; 1 in 3 pull requests now involves an agent.
• Foundry: 100,000 customers; revenue more than doubled YoY (first Foundry revenue growth disclosure); 1T-token-run-rate customers up 4x YoY (from +30% QoQ last quarter — metric basis changed); Levi Strauss running 1,000+ domain agents on OpenAI+Anthropic; Telefónica corporate agentic platform.
• Fabric: 40,000+ paid customers (+60%+, from 35,000); 17,000+ Foundry+Fabric customers (+60%); new Rayfin agent-first SDK — 2,500+ customers already, powering Replit app backends.
• Databases: PostgreSQL revenue +55%, accelerating for a third straight quarter; PostgreSQL customers also using Foundry +80%; HorizonDB launched (managed PostgreSQL, 3x throughput vs self-managed).
• Agent 365: ~40M agents registered across tens of thousands of companies, two months in (from "tens of millions" last quarter). Still no revenue metrics.
• Security: Purview audited 50B+ Copilot interactions, +~360% YoY (from 35B); Project Perception launched (multimodal agentic red/blue/green-team system) — will be a consumption-based offering at GA.
• Healthcare: on pace for 100M+ automated patient encounters this calendar year; 28M this quarter, +2x YoY; Mass General Brigham 4,000+ providers on Dragon Copilot (21% burnout reduction cited).
• LinkedIn: AI recruiting solutions at 20,000+ companies; "seats increased to 140% quarter-over-quarter" (as stated — ambiguous phrasing, likely a 140% QoQ increase; worth verifying). No update to the $450M agent ARR figure.
• Consumer/legacy: Bing and Edge share gains for 5 straight years; Xbox reset underway with FY27 return-to-growth commitment; no Game Pass metrics this quarter.

Sell-side read-through

• Only six questions, and the tone was notably non-adversarial — no one challenged the FY27 margin-down guide, the Xbox impairment, the useful-life change, or OpenAI concentration.
• Keirstead (UBS) on open/custom models and the frontier-lab tension: Nadella gave the fullest articulation yet of the substitutable-model architecture (harness separate from models; "every firm having a frontier"; even invoked the Hugging Face incident as an argument against single-model dependence). Hood's add was the economic point: Azure is fungible infrastructure regardless of model chosen. The question of how Microsoft benefits while holding large frontier-lab exposure was answered architecturally, not financially.
• Thill (Jefferies) on Azure drivers — execution vs environment: Hood confirmed constraints persist ("demand continues to exceed available supply," even citing spot-market pricing for assets) and attributed the beat to monetizable efficiency gains and faster plug-in times. This is the second straight quarter the beat mechanism is "capacity/efficiency delivered earlier" — a repeatable but hard-to-audit claim.
• Moerdler (Bernstein) on overcapacity risk and component-price pass-through: Hood's most substantive risk answer of the call — capex has pivoted to short-lived assets that can be slowed if demand changes; land/builds are a smaller, staggerable share; diverse book of business; first-party apps as a capacity sink. On pricing: component inflation "impacting everybody equivalently," cloud ROI improves relative to on-prem, and newer contracts are being priced to reflect costs — an explicit statement that inflation is being passed through in new commitments.
• Wood (Morgan Stanley) on Copilot pilots-to-deployment and monetization: Nadella cited time-to-usage collapse (months → days), Outlook/Teams-level engagement, and enterprise wiring (E7/Agent 365/D365 as Cowork plug-ins); Hood pointed to E7's token-spend governance as the ARPU driver. No churn, seat-contraction, or pilot-failure data offered — same evasion pattern as last quarter's Borges question.
• Zelnick (Deutsche Bank) on Project Perception and trust: product answer (agentic red/blue/green teams, multi-model resilience, 50%-cost CyberGym result); no security-revenue sizing.
• Borges (Goldman) on capex ROI: Hood conceded "my math has changed" over the past year — framing ROI now as confidence in TAM expansion plus enumerated margin levers (first-party silicon, model diversification, portfolio mix, hyperscale efficiency) rather than a stated return hurdle. Candid, but it means no quantified ROI framework was offered for a >$50B/quarter capex run rate.
• Nobody asked about: the third straight Xbox impairment (size still undisclosed); the FY27 margin guide flipping from up ~1 pt to down <1 pt; the useful-life extension's optics (lease reclassification cutting reported capex by ~$15B); OpenAI's share of RPO (still no updated percentage); D365 CRM moderation; the Anthropic gain's implications for the OpenAI relationship.

Management credibility

• For: Azure beat the guide by 3–4 pts (43% vs 39–40% cc) with the same stated mechanism as last quarter (efficiency + earlier capacity) — and then guided to ~45% cc with H1 acceleration, putting the promise on the record. The GitHub Copilot usage-pricing change, flagged last quarter as a margin drag, produced >60% QoQ revenue acceleration and improved margins through the quarter — the pricing fix worked as designed, on schedule. Cloud GM came in better than the ~64% guide at 65%, breaking the four-quarter step-down sequence. Copilot seat disclosures again escalated with specifics (30M, NHS 505K, EY 400K). Bookings ex-OpenAI re-accelerated to +18% as the consumption-transition air pocket narrative predicted it should if demand was real. The useful-life change was disclosed proactively with quantified capex impact (~$175B vs ~$190B) and a stated minimal FY27 operating-income benefit.
• Against / soft spots: FY27 operating margins guided down <1 point after FY26 was raised twice to "up ~1 point" — the margin trajectory inflects negative just as capex exceeds $50B/quarter; the explanation (investment) is generic. Xbox impairment for a third consecutive quarter, still with no size or composition disclosed, now paired with a "return to growth in FY27" promise that requires verification. Windows OEM beat the high-teens-decline guide via the same channel-inventory-build dynamic for at least the fifth time, and the Q1 guide (down low 20s) again assumes normalization — this recurring beat-then-guide-down pattern has not resolved. D365 CRM "moderating with longer sales cycles" is new negative language following last quarter's renewal weakness. The Anthropic gain ($3.2B) flattered other income; the $0.27 discrete-item EPS benefit means the headline beat overstates operating outperformance. Still no Copilot revenue dollarization at 30M seats; no OpenAI share of RPO for a second straight quarter.
• Under pressure: Hood's capex-ROI answer (Borges) was honest but framework-only — "my math has changed" plus a lever list, no hurdle rates or payback math. The overcapacity answer (Moerdler) was the strongest risk response: short-lived-asset pivot, staggerable builds, contract-level price pass-through. Nadella's Copilot deployment answer (Wood) again supplied usage anecdotes (including a personal ROIC-dashboard demo) instead of cohort retention data.

What changed versus the prior quarter

• Azure inflected up hard: 43% print vs 39–40% cc guide, and Q1 guided to ~45% cc with H1 acceleration — last quarter's "modest acceleration in H2 CY2026" promise is now being guided into the numbers, earlier and larger.
• Bookings ex-OpenAI re-accelerated: +18% vs +7% — the consumption-transition bookings air pocket closed, at least for one quarter. RPO quality framing improved: all sequential growth from non-frontier-model customers; ~90% of FY26 cloud revenue from non-frontier-model customers.
• Cloud GM beat (65% vs ~64% guide) and guided stable QoQ — the four-quarter guided step-down sequence stopped; GitHub Copilot flipped from named margin drag to named revenue/margin contributor after usage pricing.
• Capex crossed $41B and steps to >$50B in Q1; CY2026 restated to ~$175B purely via lease reclassification (useful life 15→25 years) — reported capex now understates the buildout relative to prior quarters; operating lease payments already rising in OCF.
• FY27 margin posture flipped: down <1 point (from FY26's twice-raised "up ~1 point") — investment now explicitly outruns leverage.
• Copilot scaled 20M → 30M seats with adds doubling QoQ, and E7 went from launch to millions of seats in two months — the suite-attach monetization path has its first evidence.
• GitHub Copilot usage pricing validated: >60% QoQ revenue acceleration; the seats+consumption thesis produced its first clean financial proof point.
• Foundry got its first revenue disclosure (100K customers, revenue >2x YoY); the 1T-token metric jumped to 4x YoY (basis changed from QoQ — comparability caution).
• MPC worsened: third straight Xbox impairment; Xbox content/services -10% (vs -5%); Windows OEM & Devices guided to low-20s decline (from high teens); MPC op margin fell to 21% (from 28%).
• D365 language deteriorated further: from renewal weakness to "CRM continued to moderate with longer sales cycles"; growth slowed to +13% (from +22%).
• New named entities entered the story: Anthropic ($3.2B investment gain), Mistral (Sovereign Cloud partnership), OpenClaw (powering Autopilots) — the multi-model world is now showing up in Microsoft's P&L and product dependencies, not just its catalog.

Bull case

• Azure acceleration is now printed and guided, not promised: 43% with a ~45% cc guide and H1 acceleration expected, on a >$100B base, with efficiency gains that monetize immediately under constraint (management statements; synthesis, not a recommendation).
• Demand quality improved visibly: bookings ex-OpenAI +18%, all sequential RPO growth from non-frontier customers, ~90% of cloud revenue from non-frontier customers, 30% of RPO recognized within 12 months (+37%) — the OpenAI-concentration critique lost some force this quarter.
• The consumption model produced its first financial wins: GitHub Copilot +60% QoQ revenue post-pricing-change; Cowork usage billing live with thousands paying; D365 customer-service credit consumption +4x QoQ; E7 at millions of seats in two months.
• Copilot adoption metrics are compounding at scale: 30M seats, adds doubling QoQ, 7x growth in 50K+ seat customers, months-to-days time-to-usage, NHS/KPMG/HSBC/EY as reference deployments with quantified outcomes (43 min/day).
• The COGS flywheel is now quantified per workload: 89% GPU cost cut in D365, 84% in PowerPoint, 50% in security, 4x Copilot throughput since the start of the year, Maia 200 at 30% better perf/$ — direct counters to the AI-margin skepticism.
• FCF $19.6B with OCF +30% while capex hit $41B — cash generation is keeping pace with the buildout; FY27 guided FCF-positive.
• Model-substitutability architecture de-risks the OpenAI dependency while monetizing everyone else's models (Anthropic, Mistral, xAI) through Foundry — and the Anthropic stake is already producing multi-billion-dollar gains.

Bear case

• FY27 margins guided down while capex steps to >$50B/quarter — the operating-leverage story pauses precisely as spending peaks; COGS guided +23–24% against revenue +16–17% in Q1, so gross-margin compression continues (guidance math; inference on severity).
• The useful-life change reclassifies ~$15B of CY2026 spend out of reported capex into operating leases — economically neutral per management, but it flatters the capex trendline and will complicate capex-to-revenue analysis; operating lease payments are already pressuring OCF.
• The beat quality was flattered by discretes: $0.27 EPS benefit including a $3.2B Anthropic mark — an investment gain, not operating performance — plus retirement-program cost timing.
• Xbox is a deteriorating asset with an undisclosed hole: three straight impairment quarters, content/services -10%, op margin down to 21%, and a "return to growth in FY27" commitment with no bridge provided.
• Windows OEM's inventory-build dynamic repeated again — the Q4 "beat" came from the same channel behavior that has unwound negatively every time; Q1 guided down low 20s with a self-described "wider than normal" range, and FY27 guided down high teens.
• D365 is decelerating with worsening language: +13% (from +22%), CRM moderating on longer sales cycles — the seat-to-consumption transition is still costing bookings in CRM even as it helps elsewhere.
• ROI accountability remains qualitative: Hood's "my math has changed" answer, with no hurdle or payback framework, on a >$200B annualized capex run rate; the demand-durability question from prior quarters was not re-asked or answered with new macro evidence.
• OpenAI disclosure still regressed: no RPO percentage update for a second quarter; quarterly bookings/RPO volatility from OpenAI contract timing explicitly flagged for FY27.

Next-quarter watchlist

1. Azure vs ~45% cc Q1 guide and the promised H1 acceleration; whether efficiency-driven monetization is repeatable or was a one-quarter pull-forward; any update to constraint language (no explicit "through 2026" restatement this quarter — note the softer phrasing).

2. Capex vs >$50B Q1 guide under the new lease accounting; operating-lease payment growth in OCF; whether the ~$175B CY2026 restated figure holds; FCF against the step-up.

3. Cloud GM "relatively stable QoQ" — whether the 65% print holds as AI usage grows; component-price pass-through evidence in new contract pricing (Hood's stated mechanism).

4. FY27 margin trajectory: first quarter under the "down <1 point" guide — Q1 op margins guided roughly flat YoY; watch whether opex stays at 7–8% and whether headcount declines continue (-2% this quarter).

5. M365 Copilot: seat adds off 30M (guided sequential growth in commercial cloud through the year); any first revenue dollarization; E7 attach beyond the launch cohort; Cowork usage-billing revenue evidence; the "super app" launch details promised "soon."

6. GitHub Copilot: whether the >60% QoQ revenue acceleration annualizes into a durable run rate; consumption vs seat mix; margin trajectory post-pricing change.

7. Xbox: impairment size/composition (third quarter undisclosed); content/services vs mid-single-digit decline guide; concrete steps behind the FY27 return-to-growth commitment.

8. Windows OEM & Devices vs low-20s decline guide: whether the channel inventory build finally unwinds; memory-price impact on PC demand; FY27 high-teens decline assumption.

9. D365: CRM moderation and longer sales cycles — stabilization or further deceleration vs the low-teens guide; ERP bookings health; usage-credit expansion beyond customer service (+4x QoQ this quarter).

10. OpenAI/Anthropic: bookings and RPO volatility from prior-year OpenAI contract laps (explicitly flagged); any restored RPO percentage disclosure; how Anthropic economics (gain recognized this quarter) coexist with the OpenAI revenue-share terms through 2030.

11. Foundry and Agent 365: whether Foundry's >2x revenue growth gets a dollar figure; conversion of 40M registered agents into E7/Agent 365 revenue; Rayfin/Replit traction.

12. Frontier Co.: whether the 6,000-person outcome-engineering org shows up as services revenue, cost, or Copilot/E7 attach — and any early ROI evidence from the 330+ completed projects.

13. Useful-life change follow-through: quantified FY27 depreciation benefit ("minimal" per management) and the finance-to-operating lease shift's effect on reported capex comparability.

Quarter in one view

• Revenue $82.9B (+18%, +15% cc); EPS $4.27 (+21%, +18% cc adjusted for OpenAI); operating income +20% (+16% cc); operating margin 46%, up slightly YoY — "exceeded expectations across revenue, operating income, and earnings per share." Microsoft Cloud revenue $54.5B (+29%, +25% cc); cloud GM 66%, slightly better than expected, down YoY on AI investment.
• AI business ARR surpassed $37B, +123% YoY — first explicit AI revenue run-rate disclosure on the call.
• Azure +40% (+39% cc), ahead of expectations vs 37–38% cc guide — beat attributed to capacity delivered earlier in the quarter enabling increased consumption across AI and non-AI services. Demand still exceeds supply; "constrained at least through 2026" formally restated.
• RPO $627B, +99% YoY including OpenAI; +26% ex-OpenAI ("in line with historic seasonality"); duration ~2.5 years; 25% recognized in next 12 months (+39%); beyond-12-months portion +138% — the ex-OpenAI decomposition is now the standard framing. Commercial bookings -4% (-6% cc) including OpenAI; +7% ex-OpenAI.
• Capex $31.9B, down sequentially (build-out variability, finance-lease timing); ~2/3 short-lived; finance leases $4.7B; cash PP&E $30.9B. FCF recovered to $15.8B (from $5.9B) on OCF of $46.7B (+26%). $10.2B returned to shareholders.
• Q4 capex guided to over $40B (including ~$5B from higher component pricing); CY2026 capex ~$190B, including ~$25B from higher component pricing — memory/component inflation is now a quantified, material capex driver.
• FY26 op-margin guidance raised again: now "up about one point" YoY (from "up slightly"), inclusive of ~$900M one-time voluntary-retirement costs in Q4. Headcount declined YoY and will decline again in FY27.
• Another ~1GW of capacity added; Fairwater Wisconsin online six weeks ahead of schedule; on track to double footprint in two years.

What management is focused on

• The business-model transition: seats → seats + consumption. The dominant theme of both prepared remarks and Q&A. Hood: per-seat businesses become "a license business plus a consumption business... it'll also have a meter, just like you see in Azure," and bookings "may not all flow through bookings in the same way." Nadella: "any per-user business... will become a per-user and usage business." GitHub Copilot moved to usage-based pricing effective June 1; ~60% of D365 customer-service customers already buy usage-based credits; Copilot credit consumptive offer up ~2x QoQ.
• "Who pays for it" answered with evals/outcomes: Nadella's response to Weiss — dollars come from compressed workflows, cost reduction, or revenue lift measured by customer evals; IT budgets get "reshaped by business outcomes... and reallocation from other line items."
• Capacity execution as the binding constraint: dock-to-live times for new GPUs down ~20% since start of year; Fairwater Wisconsin six weeks early ("allowing us to recognize revenue earlier"); 40% inference throughput improvement on most-used Copilot models; constrained at least through 2026 with Azure acceleration expected in H2 CY2026.
• First-party silicon and models as COGS levers: Maia 200 live in Iowa and Arizona (30%+ tokens-per-dollar improvement); Cobalt in nearly half of DC regions (Databricks, Siemens, Snowflake cited), supply "expanding significantly"; MAI-Transcribe-1 (67% GPU efficiency gain) and MAI-Image-2 (up to 260%) powering Bing/PowerPoint and sold via Foundry (Shutterstock, WPP).
• The "IQ layer" / Work IQ as the moat: Work IQ spans 17+ EB of data (+35% YoY); Fabric + Foundry + M365 + security graph framed as a unified context engine; 15,000+ customers use both Foundry and Fabric (+60% YoY).
• Consumer reset: explicit "win back fans" language — Windows fundamentals (performance on lower-memory devices, update experience), Xbox "recommitting to core fans," Game Pass price changes.

Key numbers and quarter mechanics

• Total: revenue $82.9B (+18%/+15% cc); company GM 68% (down YoY); opex +9% (+8% cc, low prior-year comp in S&M and G&A); op margin 46%; headcount down YoY. FX roughly in line.
• OpenAI mechanics: adjusted other income +$961M ex-OpenAI (investment gains partially offset by FX remeasurement losses). EPS given on adjusted ex-OpenAI basis. New agreement terms disclosed: IP royalty-free through 2032; revenue share to Microsoft through 2030; Microsoft's rev share to OpenAI eliminated.
• Bookings/RPO: bookings -4% (-6% cc) incl. OpenAI, +7% ex-OpenAI; RPO $627B (+99% incl. OpenAI, +26% ex-OpenAI); ~25% recognized within 12 months (+39%); beyond-12-months +138%; duration ~2.5 years.
• P&BP $35.0B (+17%/+13% cc): M365 commercial cloud +19% (+15% cc), ahead; paid commercial seats +6%; commercial products +1% (-3% cc, Office 2024 normalization); consumer cloud +33% (+29% cc, subs +7%, ~95M subscribers); LinkedIn +12% (+9% cc); D365 +22% (+17% cc) but bookings hit by weaker renewals as customers balance seat vs consumption models. Segment op margin 60%.
• IC $34.7B (+30%/+28% cc): Azure +40% (+39% cc); on-prem +slightly (-3% cc). Segment op margin 40% (down from 42%).
• MPC $13.2B (-1%/-3% cc): Windows OEM +slightly, ahead of expectations — OEMs/channel built inventory ahead of memory price increases again; search ex-TAC +12% (+9% cc); gaming -7% (-9% cc); Xbox content/services -5% (-7% cc); further gaming impairment and related expenses in opex. Segment op margin 28%.
• Cash flow: OCF $46.7B (+26%); capex $31.9B; FCF $15.8B (recovered from $5.9B); finance leases $4.7B; cash PP&E $30.9B.
• Q4 FY26 guide (USD): revenue $86.7–87.8B (+13–15%); COGS $29.4–29.6B (+22–23%, incl. ~$350M retirement-program cost); opex $19.3–19.4B (+~7%, incl. ~$550M retirement cost); other income ~-$100M ex-OpenAI (interest expense incl. finance-lease interest exceeds interest income); tax ~19%; FX <1 pt revenue tailwind. Cloud GM ~64% (down from 66% print — fourth consecutive step-down in the guide/print sequence).
• Segment guides: P&BP $37.0–37.3B (+12–13%); M365 commercial cloud 15–16% cc adjusted (13–14% reported; prior year had 2 pts in-period recognition); net paid Copilot seat adds to increase sequentially; commercial products mid-single digits; consumer cloud low-20s (lapping price increase); LinkedIn ~10%; D365 low double digits (down sequentially on comps and bookings trends). IC $37.95–38.25B (+27–28%); Azure 39–40% cc; on-prem down mid-single digits. MPC $11.75–12.25B; Windows OEM down high teens (~6 pts EOS comp, ~6 pts inventory unwind, ~6 pts lower PC market from memory pricing; "wider than normal" range); search ex-TAC high single digits; Xbox content/services down low teens (comp + Game Pass price changes); hardware down.
• FY26/FY27: FY26 op margins up ~1 pt YoY (raised). FY27: double-digit revenue and operating income growth; opex growth mid-to-high single digits; headcount down YoY; lapping EOS/OEM-inventory/transactional comps. Azure: modest acceleration in H2 CY2026 vs H1.

Product and launch scorecard

• M365 Copilot — the quarter's headline: >20M paid seats (from 15M); record seat adds +250% YoY (from +160%), fastest since launch, with QoQ acceleration continuing; customers with 50,000+ seats quadrupled YoY; Accenture at 740,000+ seats (largest win to date); Bayer, J&J, Mercedes, Roche each 90,000+. Usage: queries/user +~20% QoQ; first-party agent MAU +6x YTD; weekly engagement "at the same level as Outlook." Agent mode now default in Word/Excel/PowerPoint; Cowork (delegated async tasks) launched; 625+ updates in the past year (+50%). Still no Copilot revenue dollar figure.
• GitHub Copilot: ~140,000 organizations; enterprise subscribers nearly tripled YoY; CLI usage ~doubling MoM; usage-based pricing announced, effective June 1 — a margin/COGS alignment move Hood tied to cloud-GM pressure from Copilot usage. (No paid-subscriber count update vs 4.7M last quarter.)
• Foundry: 10,000+ multi-model customers; 5,000+ using open-source models; Anthropic+OpenAI dual users up 2x QoQ (from 1,500+); 300+ customers on track for 1T+ tokens (from 250+), +30% QoQ; Bayer built an in-house agent platform (20,000+ MAU). Foundry Agent Service adds durable stateful agents.
• Fabric: 35,000 paid customers (+60% YoY, from 31,000); OneLake data ~4x YoY; 15,000+ Foundry+Fabric customers (+60%). No revenue run-rate update ($2B last quarter). Cosmos DB revenue +50% YoY on AI app workloads.
• Agent 365: tens of thousands of companies managing tens of millions of agents — first adoption quantification, still no revenue metrics.
• Security: Security Copilot customers +2x YoY; data-security triage agents handled 2M+ unique alerts; Purview audited 35B Copilot interactions, +7x YoY (from 24B); sim-ship Defender protections for AI-discovered vulnerabilities.
• LinkedIn agentic products: Talent Solutions agents surpassed $450M annualized revenue run rate — a rare dollarized agent revenue disclosure.
• Copilot Studio / low-code: ~90% of Fortune 500 have active agents; Copilot credit consumption ~2x QoQ.
• MAI first-party models: MAI-Transcribe-1 and MAI-Image-2 launched, in Bing/PowerPoint, heading to Copilot/Teams; first commercial licensing via Foundry (Shutterstock, WPP).
• Consumer: Windows MAU devices >1.6B; Bing MAU 1B for the first time; Edge share gains 20 consecutive quarters; M365 Consumer ~95M subscribers; record Xbox MAU and streaming hours despite revenue decline; Game Pass price changes announced.

Sell-side read-through

• Six questions; the demand-sustainability question led. Weiss (Morgan Stanley, his final MSFT call) asked the macro question directly: CIO surveys show excitement but flat IT budgets and GDP — "who's paying for all of this?" Hood's answer reframed bookings itself (seat license + meter; usage may bypass bookings), and Nadella pointed to eval-measured outcomes and budget reallocation. This is a framework answer, not evidence — no quantified spend-displacement data offered.
• Keirstead (UBS) pressed the CY2026 ~$190B capex / ~$120B H2 ramp: Hood expressed confidence in the supply chain, attributed ~$25B to component pricing (implying the rest is volume, leaning short-lived assets), and reiterated the first-party-vs-Azure allocation tension "will persist." No partner-financing or third-party capacity details given.
• Thill (Jefferies) teed up the AI-margin question (why are hyperscaler margins rising?): Hood argued AI margins are tracking better than the cloud transition at the same stage, citing consumption pricing, royalty-free OpenAI IP, first-party silicon, and efficiency work. Notably candid that OpenAI IP "is obviously free to us for a long time."
• Moerdler (Bernstein) on the capex-vs-revenue disconnect: Hood pointed to short-lived assets as the revenue-correlated spend, the $600B+ book of business "we still need to deliver," and M365 commercial cloud acceleration (and guided Q4 improvement) as the first place usage models show up in revenue; GitHub next. Nadella added the structural-TAM argument (knowledge work, coding, security) and "model capabilities are exponential... you have to be ready."
• Borges (Goldman) on Copilot learnings/E7/Cowork: Nadella gave a product-form-factor answer (chat → agents → edit mode → Cowork delegation; multi-model harness decoupled from models; Work IQ context) — no churn, seat-contraction, or failure-mode discussion despite the "what's not working" framing.
• Materne (Evercore) got the OpenAI agreement news: royalty-free IP through 2032, revenue share to Microsoft through 2030, elimination of Microsoft's rev share to OpenAI. Nadella framed it as win-win evolution acknowledging customer model-diversity demands. Financial modeling impact beyond those terms was not detailed.
• Jaluria (RBC) on seat-vs-consumption mix and E7: Nadella — seats are "entitlement to some consumption," overages go to metered consumption with commitment-based discounts; mix will be defined by customer evals. No 3–5-year mix quantification given.
• Nobody asked about: the gaming impairments (second consecutive quarter); the D365 renewal weakness; cloud GM guided to ~64%; the $25B component-price capex inflation's margin implications; Windows OEM's fifth inventory unwind; OpenAI's share of the $627B RPO (no updated percentage given this quarter).

Management credibility

• For: Azure beat again (39% cc vs 37–38% guide) with a concrete mechanism (earlier capacity delivery; Fairwater Wisconsin six weeks ahead of schedule). FCF recovered to $15.8B as the lease-mix explanation from last quarter predicted — the $5.9B print now looks like timing, as Hood framed it. Copilot seat disclosures continue to escalate and are now specific (20M+, +250% adds, named 90K–740K-seat customers). FY26 margin guidance raised for a second straight quarter ("up about one point") while absorbing $900M of one-time costs. The OpenAI agreement terms were disclosed with specifics (2032 IP, 2030 rev share) rather than characterized vaguely. Component-price inflation was quantified ($5B in Q4, $25B in CY26) rather than left as a qualitative risk.
• Against / soft spots: Cloud GM guided down a fourth consecutive time (~64%), now with GitHub Copilot usage explicitly named as a margin drag — the usage-based transition carries a COGS cost the seat narrative doesn't. Gaming took impairment charges for a second straight quarter with no size or detail volunteered. D365 bookings weakened on renewals — disclosed but buried in segment commentary, and the Q4 D365 guide decelerates. Windows OEM beat again on the same dynamic that has repeatedly failed to normalize (channel inventory building ahead of memory prices) — and the Q4 guide now assumes the unwind plus a memory-driven PC-market decline, a fifth normalization attempt. The "constrained at least through 2026" restatement extends the constraint horizon beyond prior "through FY26" language. No updated OpenAI share of RPO was given after last quarter's 45% disclosure — the decomposition shifted to growth rates (+26% ex-OpenAI) rather than the dollar split.
• Under pressure: Hood handled the capex-magnitude (Keirstead) and capex-vs-revenue (Moerdler) questions with more specificity than prior quarters (price vs volume split, short-lived correlation, $600B+ book of business). The weakest answers were on demand durability (Weiss) — framework without data — and Copilot "what's not working" (Borges), which went unanswered.

What changed versus the prior quarter

• FCF recovered: $5.9B → $15.8B as finance leases fell further ($6.7B → $4.7B) but OCF jumped to $46.7B (+26%); cash PP&E ($30.9B) now roughly equals capex — the cash-cost of the buildout is no longer lease-shielded, but operating cash generation covered it this quarter.
• Capex trajectory re-accelerated sharply: Q3 printed $31.9B (down sequentially as guided), but Q4 guided to >$40B and CY2026 to ~$190B with ~$25B from component pricing — memory inflation moved from a flagged risk to a quantified capex line.
• AI revenue disclosed for the first time: $37B+ ARR, +123% — a new standing metric that partially answers the "no AI revenue number" gap.
• OpenAI agreement restructured: royalty-free IP through 2032, revenue share to Microsoft through 2030, Microsoft's rev share to OpenAI eliminated. RPO decomposition shifted from the 45% dollar split to growth-rate framing (+99% incl., +26% excl.).
• Copilot crossed 20M paid seats with accelerating adds (+250% vs +160%) and mega-deals stepped up an order of magnitude (Accenture 740K vs Publicis 95K last quarter). GitHub Copilot moved to usage-based pricing (June 1) — the seats+consumption thesis became an actual pricing action.
• Azure guide stepped up: 39–40% cc for Q4 (vs 37–38% for Q3) with "modest acceleration" promised in H2 CY2026 — management is now guiding acceleration, not deceleration, on a larger base.
• Constraint language hardened and extended: "constrained at least through 2026" (calendar) vs prior softer "demand exceeds supply."
• FY26 margin guidance raised again ("up about one point" vs "up slightly"), now absorbing a $900M voluntary-retirement charge and a YoY headcount decline — a new cost-action signal.
• MPC deterioration continued: second straight gaming impairment; Xbox content/services guided to low-teens decline (from mid-single digits); Windows OEM guided down high teens with a three-part bridge (EOS comp, inventory, memory-driven PC market).
• New pressure points surfaced: D365 renewal weakness; GitHub Copilot usage named as a cloud-GM drag; search recovered to +12% (+9% cc) from last quarter's miss.

Bull case

• Demand evidence broadened and dollarized: $37B AI ARR (+123%), RPO $627B with the beyond-12-months portion +138%, ex-OpenAI RPO +26% "in line with historic seasonality," and Copilot seat adds accelerating at 20M+ scale with 740K-seat single-customer wins (management statements; synthesis, not a recommendation).
• Azure acceleration is now guided, not just hoped for: 39–40% cc Q4 guide on a tougher comp, plus explicit "modest acceleration in H2 CY2026" — with capacity delivery (dock-to-live -20%, Fairwater early) as the stated mechanism.
• The consumption transition is showing up in the P&L where management said it would: M365 commercial cloud accelerated to +19% and is guided better in Q4; LinkedIn agents at $450M ARR; ~60% of D365 service customers on usage credits; Copilot credits ~2x QoQ.
• Margin delivery keeps improving mid-buildout: FY26 op margins now guided up ~1 pt despite $900M one-time costs; P&BP at 60%; headcount declining; AI margins claimed better than cloud at the same stage, with royalty-free OpenAI IP through 2032 as a structural COGS advantage.
• FCF concern from last quarter resolved as timing: $15.8B print with OCF +26% demonstrates the buildout can be cash-funded at current demand levels.
• Full-stack economics compounding: Maia 200 (30%+ tokens/$), Cobalt in ~half of DC regions with expanding supply, MAI models cutting first-party COGS (67–260% efficiency gains), 40% inference throughput gains — the tokens-per-watt-per-dollar flywheel is producing measurable outputs each quarter.

Bear case

• Capex is escalating faster than revenue: ~$190B CY2026 (with $25B pure price inflation) against FY27 guided to "double-digit" growth; Q4 capex >$40B vs $31.9B print. The capex-to-revenue conversion question (Moerdler) was answered with framework, not a timeline or ratio (inference from the guidance math).
• Cloud GM guided down a fourth straight time (~64%), and the usage-based model itself is now a stated drag (GitHub Copilot consumption) — the more successful the agent strategy, the more COGS it generates until pricing catches up (management's own linkage).
• The demand-durability question remains unanswered with data: Weiss's "who pays for this" got an evals-and-reallocation framework; flat IT budgets and GDP were not rebutted with evidence. Bookings ex-OpenAI grew only +7%, and D365 renewals weakened — early signs that the seats+consumption transition creates bookings air pockets (management disclosed the dynamic; the severity is inference).
• OpenAI concentration disclosure regressed: no updated percentage of RPO this quarter; the +99% headline RPO growth is heavily OpenAI-driven while the ex-OpenAI +26% is the cleaner number — and the new agreement's economics (eliminated rev share to OpenAI) were not quantified.
• Consumer/legacy drag is structural, not transitional: second consecutive gaming impairment, Xbox content/services guided to low-teens decline, Windows OEM's fifth inventory unwind now compounded by a memory-driven PC-market decline, MPC revenue declining again.
• Constraint horizon extended ("at least through 2026") — Azure remains supply-capped and the residual claimant after first-party and R&D allocations, with the first-party-vs-Azure tension acknowledged to "persist."
• Q4 opex and COGS both embed one-time costs and 22–23% COGS growth against 13–15% revenue growth — gross-margin compression continues even as operating margin is protected by headcount actions.

Next-quarter watchlist

1. Azure vs 39–40% cc Q4 guide and evidence for the promised H2 CY2026 acceleration; capacity-added cadence vs the ~1GW/quarter rate; any update to "constrained at least through 2026."

2. Capex vs >$40B Q4 guide and the ~$190B CY2026 plan — price vs volume split, finance-lease mix, and whether cash PP&E continues tracking total capex; FCF trajectory against it.

3. Cloud GM vs ~64% guide — AI COGS, memory/component costs, and GitHub Copilot usage drag vs efficiency gains; whether the step-down sequence finally stops.

4. GitHub Copilot usage-based pricing (June 1) — early evidence on revenue acceleration (Hood's stated expectation), churn, or usage throttling; updated subscriber metrics.

5. M365 Copilot — sequential seat-add increase (guided) off 20M+; whether +250% add growth holds; any first revenue dollarization; E7 launch details and pricing structure.

6. OpenAI — Q4 accounting impact size/direction; any updated share of RPO; revenue-share and IP economics flowing into reported lines; ex-OpenAI bookings growth vs +7%.

7. D365 — whether renewal weakness and the seat-to-consumption shift worsen (guided deceleration to low double digits); bookings trend disclosure.

8. Windows OEM vs high-teens decline guide — the three-part bridge (EOS comp, inventory, memory-driven PC market); whether the inventory unwind finally completes; memory-price pass-through to PC demand.

9. Gaming — impairment size and composition (second straight quarter, still undisclosed); Xbox content/services vs low-teens decline guide; Game Pass price-change impact on subscribers and revenue.

10. FY27 setup — opex growth mid-to-high single digits and headcount decline vs double-digit revenue/income growth commitment; retirement-program cost takeout durability; FY27 margin posture.

11. AI ARR metric — whether $37B/+123% becomes a standing disclosure and how it reconciles with Azure AI contribution.

12. Consumption-mix evidence — Copilot credit growth, D365 usage-credit penetration beyond customer service, and whether bookings ex-OpenAI re-accelerates as the model transition matures.

13. Agent 365 and Foundry — conversion of "tens of millions of agents managed" and 300+ 1T-token customers into revenue disclosures; multi-model customer growth as the Anthropic relationship deepens.

Quarter in one view

• Revenue $81.3B (+17% cc); EPS $4.14 (+24% cc, adjusted for OpenAI impact); operating income +21% cc; operating margin 47%, up YoY and ahead of expectations — "again exceeded expectations across revenue, operating income, and earnings per share." Microsoft Cloud revenue $51.5B (+26% cc), first time above $50B; cloud GM 67% (slightly better than expected, down YoY on AI investment).
• Azure +39% cc, "slightly ahead of expectations" vs ~37% cc guide — beat attributed to efficiency gains across the fungible fleet enabling reallocation of capacity to Azure that was monetized in the quarter. Demand still exceeds supply. Hood disclosed: had all GPUs that came online in Q1+Q2 been allocated to Azure, "the KPI would have been over 40."
• RPO $625B, +10% YoY, weighted average duration ~2.5 years (up from ~2 years); ~45% of commercial RPO is OpenAI; the remaining ~$350B grew 28% — the first explicit quantification of OpenAI's share of backlog. 25% of RPO recognized in next 12 months (+39%); beyond-12-months portion +56%.
• Commercial bookings +23% cc — driven by "the previously large Azure commitment from OpenAI" (the $250B deal now in bookings), the previously announced Anthropic commitment from November, and healthy core annuity growth. Deceleration from +112% reflects the prior-quarter OpenAI mega-booking comp.
• OpenAI accounting change: post-recapitalization, Microsoft now records its share of the change in OpenAI's net assets rather than share of operating losses — producing a gain that drove GAAP other income to $10B this quarter. Ex-OpenAI, other income was slightly negative and below expectations (net investment losses).
• Capex $37.5B (vs $34.9B last quarter); ~two-thirds short-lived assets (GPUs/CPUs); finance leases $6.7B; cash PP&E $29.9B — and FCF collapsed to $5.9B, down sequentially from $25.7B, explicitly because of higher cash capex from a lower finance-lease mix. OCF $35.8B (+60%). $12.7B returned to shareholders (+32%).
• Nearly 1GW of total capacity added this quarter alone; Fairwater Atlanta + Wisconsin connected via AI WAN into a "first-of-kind AI super factory."
• FY26 operating margin guidance raised: now "up slightly" (from "relatively unchanged"), citing H1 prioritization and favorable mix from Windows OEM and commercial on-prem.

What management is focused on

• Reframing the capex debate away from Azure: Hood's central argument — Azure guidance is "an allocated capacity guide," not a demand signal. GPUs are allocated first to first-party apps (M365 Copilot, GitHub Copilot), then R&D, then Azure. Nadella: "we don't wanna maximize just one business... build the best LTV portfolio." This is the direct response to the stock's after-hours decline.
• Tokens per watt per dollar as the governing metric: 50% throughput increase on OpenAI inferencing powering Copilots; Maya 200 accelerator brought online (10+ FLOPS FP4, 30%+ better TCO vs latest fleet hardware); Cobalt 200 CPU (+50% performance vs first-gen).
• Sovereignty as a demand theme: DC investments announced in seven countries this quarter; region-specific models (Cohere cited); fine-tuning framed as customers capturing "tacit knowledge... as their core IP."
• Agent platform as the new app platform: Agent 365 launched — cross-cloud agent control plane (governance, identity, security) with Adobe, Databricks, SAP, ServiceNow, Workday, NVIDIA and others integrating; "first provider to offer this type of agent control plane across clouds."
• M365 Copilot monetization finally quantified: 15M paid seats, record seat adds (+160% YoY), DAU +10x YoY — the headline product disclosure of the call.
• Memory pricing as an emerging cost risk: flagged three times — transactional purchasing ahead of memory price increases (on-prem, OEM), potential PC market impact, and future capex/cloud-GM impact building gradually over the 6-year depreciation schedule.

Key numbers and quarter mechanics

• Total: revenue $81.3B (+17% cc); company GM 68% (down slightly YoY); opex +5% cc (R&D compute/AI talent + gaming impairment charges); op margin 47%; FX increased results slightly less than expected, particularly in Intelligent Cloud.
• OpenAI mechanics: equity-method accounting shifted to share of change in net assets post-recapitalization → $10B GAAP other income (a gain, vs $4.1B loss last quarter). Ex-OpenAI, other income slightly negative, below expectations on net investment losses. EPS given on adjusted ex-OpenAI basis.
• P&BP $34.1B (+16% cc): M365 commercial cloud +17% cc (Copilot contribution increasing; ARPU led by E5 + Copilot); paid commercial seats +6% to >450M; commercial products +13% cc (Office 2024 transactional beat again); consumer cloud +29% cc (subs +6%); LinkedIn +11% cc (Marketing Solutions); D365 +19% cc. Segment op margin 60% (up YoY).
• IC $32.9B (+29% cc): Azure +39% cc; on-prem +21% cc, well ahead — SQL Server 2025 launch plus transactional purchasing ahead of memory price increases. Segment op margin 42% (down slightly).
• MPC $14.3B (-3%): Windows OEM +5% (Windows 10 EOS + elevated inventory + pre-memory-price purchasing; devices dragged the combined line to +1%); search ex-TAC +9% cc, below expectations — "execution challenges" plus third-party normalization; gaming -9% cc; Xbox content/services -6% cc, below expectations on first-party content; gaming impairment charges hit opex. Segment op margin 27% (roughly flat).
• Cash flow: OCF $35.8B (+60%, strong cloud billings/collections); capex $37.5B; FCF $5.9B, down sharply sequentially — finance-lease mix fell to $6.7B from $11.1B, so cash PP&E ($29.9B) caught up with total capex.
• Q3 FY26 guide (USD): revenue $80.65–81.75B (+15–17%); COGS $26.65–26.85B (+22–23%); opex $17.8–17.9B (+10–11%, low prior-year comp); op margins down slightly YoY; other income ~+$700M ex-OpenAI (equity portfolio FMV gain + interest income); tax ~19%; FX +3 pts revenue. Cloud GM ~65% (down from ~66% guide and 67% print). Capex to decrease sequentially (build-out variability, finance-lease timing); short-lived mix similar to Q2.
• Segment guides: P&BP $34.25–34.55B (+14–15%); M365 commercial cloud 13–14% cc; commercial products down low single digits; consumer cloud mid-to-high 20s; LinkedIn low double digits; D365 high teens. IC $34.1–34.4B (+27–29%); Azure 37–38% cc; on-prem down low single digits. MPC $12.3–12.8B; Windows OEM down ~10% (EOS normalization + inventory unwind, wide range on memory pricing); search ex-TAC high single digits; Xbox content/services down mid-single digits (Game Pass growth partial offset); hardware down.
• FY26: op margins now expected up slightly (raised); Q4 FX <1 pt tailwind.

Product and launch scorecard

• M365 Copilot — first hard numbers: 15M paid seats; record seat adds +160% YoY with accelerating QoQ seat growth; DAU +10x YoY; conversations per user doubled YoY; customers with 35,000+ seats tripled YoY. Named deals: Publicis 95,000+ seats; Fiserv, ING, US Dept. of Interior, Westpac, two universities each 35,000+. "Multiples more" enterprise chat users than paid seats. WorkIQ positioned as the differentiation layer ("biggest quarter-over-quarter improvement in response quality to date"). Still no Copilot revenue figure.
• GitHub Copilot: 4.7M paid subscribers, +75% YoY (first paid-subscriber disclosure); Copilot Pro+ individual subs +77% QoQ; Siemens full-platform adoption after 30,000+ developer rollout; Agent HQ + Copilot CLI + new Copilot SDK (embed Copilot runtime in third-party apps).
• Fabric: >$2B annual revenue run rate (first dollar disclosure), 31,000+ customers (from 28,000 paid), revenue +60% YoY (third straight quarter at ~+60%).
• Foundry: customers spending $1M+/quarter grew ~80%; 250+ customers on track to process 1T+ tokens this year; 1,500+ customers using both Anthropic and OpenAI models; GPT-5.2 and Claude 4.5 added. No total customer count given this quarter (80,000 last quarter).
• Security: 1.6M security customers (first count), 1M+ using 4+ workloads; Security Copilot rolling out to all E5 customers (major distribution expansion); Purview audited 24B Copilot interactions, +9x YoY (from 16B); iCertus SOC triage time -75%.
• Dragon Copilot: 21M patient encounters, +3x YoY (from 17M); 100,000+ medical providers; Mount Sinai going system-wide.
• Maya 200 / Cobalt 200: Maya 200 online with 30%+ TCO improvement, scaling first for superintelligence-team inference/synthetic data, then Copilot/Foundry inference — vertical integration with optionality (NVIDIA/AMD partnerships retained).
• Agent 365: launched this quarter; cross-cloud agent governance; eight named partner integrations. New category claim, no revenue metrics.
• Consumer Copilot: app daily users ~3x YoY; Copilot Checkout with PayPal, Shopify, Stripe.
• Windows: 1B Windows 11 users, +45% YoY; share gains across Windows, Edge, Bing.
• Gaming: weak quarter — content/services -6% cc below plan, impairment charges taken, record PC players/paid streaming hours as the offset. Game Pass cited only as a Q3 guide offset.

Sell-side read-through

• Seven questions; the tone was set by Weiss (Morgan Stanley) naming the elephant: stock down after-hours despite a beat because "CapEx is growing faster than we expected, and maybe Azure is growing a little bit slower than we expected" — the ROI question posed directly. Hood's answer (Azure guide = allocated capacity; GPUs go first to first-party apps and R&D; "over 40" KPI if all Q1/Q2 GPUs went to Azure) is the most explicit allocation hierarchy disclosed to date — and confirms Azure is the residual claimant on capacity.
• Moerdler (Bernstein) pressed the duration mismatch: 6-year server depreciation vs 2.5-year RPO duration. Hood's answer: GPUs for the largest customers "are sold for the entire useful life of the GPU"; the short blended duration reflects short-dated M365 contracts; Azure-only RPO is longer; margins improve over asset life via efficiency. The strongest version of the duration-matching defense yet — but "largest customers" includes OpenAI, tying the defense back to counterparty risk.
• Thill (Jefferies) asked the 45%-OpenAI-backlog question directly. Hood pivoted to the 55% (~$350B, +28%, "larger than most peers, more diversified") and offered no new durability evidence on OpenAI beyond "great partnership... provider of scale." No guardrails, no credit mechanics — same gap as last quarter.
• Keirstead (UBS) on capacity magnitude: Hood declined to size Fairwater Atlanta/Wisconsin ramps ("multiyear deliveries... abstract away from that") — no capacity trajectory given beyond "working as quickly as we can."
• Murphy (JPMorgan) on Maya 200 vs TPU/Trainium/Blackwell: Nadella framed silicon as a real competency but emphasized fleet TCO flexibility over vertical integration — "you have to be ahead for all time to come."
• Zelnick (Deutsche Bank) on frontier-firm spend expansion: Nadella's answer was qualitative (WorkIQ as the compounding asset across M365/security/GitHub) — no quantified spend-expansion frame.
• Lenschow (Barclays) on CPU/cloud migration: Nadella confirmed AI workloads drive compute/storage demand alongside GPUs (agents spawn containers on CPUs) and SQL Server 2025 IaaS adoption >2x prior version.
• Nobody asked about: the FCF collapse to $5.9B; the OpenAI accounting change producing a $10B GAAP gain; the gaming impairment; search's execution miss; the Anthropic commitment's size; why cloud GM is guided down again to ~65%.

Management credibility

• For: Azure beat again (39% cc vs ~37% guide) with a specific, checkable mechanism (fleet efficiency → reallocated capacity monetized in-quarter). The 45% OpenAI RPO disclosure is a major transparency step — the concentration question answered with a number after two quarters of refusal. M365 Copilot finally got paid-seat disclosure (15M) after repeated quarters of usage-only metrics. Fabric got a dollar run rate ($2B). The capacity-allocation hierarchy (first-party → R&D → Azure) is candid about Azure being residual. FY26 margin guidance was *raised*, not sandbagged. Search and gaming misses were acknowledged plainly ("execution challenges," "below expectations").
• Against / soft spots: FCF fell to $5.9B from $25.7B and the prepared-remarks framing ("decreased sequentially, reflecting the higher cash capital expenditures") understates a ~77% sequential drop — no Q&A challenge, no bridge beyond lease mix. The OpenAI accounting change converts a $4.1B quarterly loss line into a $10B gain via net-asset revaluation — mechanically disclosed, but the swing's drivers (recap valuation) are opaque and unauditable from the call. Cloud GM guided down a third consecutive time (68% → ~66% → 67% print → ~65%). Windows OEM inventory normalization missed a fourth time (still elevated; unwind again assumed for Q3, now with a ~-10% guide). On-prem's +21% beat leans on pre-memory-price pull-forward — a demand-borrowing dynamic management disclosed but didn't quantify.
• Under pressure: Hood's "sold for the entire useful life" answer on GPU contracts is reassuring on duration but silent on counterparty concentration within those contracts. The OpenAI durability answer (Thill) was the weakest of the call — relationship language, no economics. Keirstead's capacity-magnitude question was deflected entirely.

What changed versus the prior quarter

• OpenAI swung from $4.1B loss to $10B gain on an accounting-method change (share of net-asset changes post-recapitalization, not operating losses) — GAAP optics transformed; ex-OpenAI other income was actually slightly negative and below plan.
• OpenAI concentration quantified for the first time: ~45% of $625B RPO; non-OpenAI RPO ~$350B, +28%. Duration extended to ~2.5 years (from ~2). The $250B commitment is now inside bookings/RPO (bookings +23% cc vs +112%).
• Anthropic entered the bookings narrative — a November commitment cited as a Q2 bookings driver; first-party model exclusivity framing continues to loosen (Claude 4.5 on Foundry; Researcher Agent supports Claude).
• FCF broke: $25.7B → $5.9B as finance leases fell to $6.7B from $11.1B and cash PP&E jumped to $29.9B. The lease-mix shield that protected FCF last quarter reversed.
• Capex mix shifted further short-lived: ~two-thirds (from ~half); Q3 guided to a sequential capex *decrease* — the first sequential decline guided since the buildout began.
• FY26 op-margin guidance raised to "up slightly" (from "relatively unchanged") — despite cloud GM guided down to ~65%.
• M365 Copilot crossed the disclosure threshold: 15M paid seats, +160% seat adds, DAU +10x — from zero seat disclosure two quarters ago. GitHub Copilot paid subs disclosed (4.7M) for the first time.
• Capacity adds accelerated: ~1GW in one quarter; Fairwater became a two-site "AI super factory" (Atlanta + Wisconsin via AI WAN); Maya 200 silicon online.
• New cost risk introduced: memory pricing — affecting OEM, on-prem transactional demand (pull-forward), and future capex/cloud GM.
• MPC deteriorated: gaming -9% cc with impairment charges; search ex-TAC missed (+9% cc) on execution; Windows OEM still inventory-inflated with a ~-10% Q3 guide.
• Capacity-constraint language softened: no explicit "constrained through end of FY26" restatement; instead "demand continues to exceed supply" plus a claim of "significantly accelerating growth rates in both Q3 and Q4" capacity additions (transcript phrasing is ambiguous — appears to reference supply adds).

Bull case

• The backlog is real, huge, and now decomposed: $625B RPO (+10% on a massively expanded base), ~$350B ex-OpenAI growing 28% with breadth "by segment, by industry, and by geo"; 25% recognizable within 12 months (+39%) — near-term revenue visibility is high (management statements; synthesis, not a recommendation).
• Copilot monetization is inflecting with numbers: 15M paid seats, seat adds +160% and accelerating, 35K+ seat customers tripling, Publicis at 95K seats — ARPU expansion (E5 + Copilot) is now a stated driver of 17% cc M365 commercial cloud growth on a 450M-seat base.
• GPU risk is contracted away: largest-customer GPU contracts cover the entire useful life; margins improve over asset life via software efficiency (50% throughput gain on OpenAI inference this quarter) — the depreciation-vs-duration bear argument got its strongest rebuttal yet.
• Capacity is the only Azure constraint, and it's compounding: ~1GW added in a single quarter; Hood's "over 40" counterfactual shows underlying Azure demand above the print; Q3/Q4 supply additions "significantly accelerating."
• Margin delivery upgraded mid-buildout: FY26 op margins now guided up slightly; P&BP at 60%, opex at +5% cc against +17% cc revenue.
• Full-stack optionality: Maya 200 (30%+ TCO gain), Agent 365 cross-cloud control plane with eight major ISVs, Foundry hosting Anthropic + OpenAI with 1,500+ dual-model customers, GitHub at 4.7M paid subs — Microsoft monetizes every model outcome.
• Security distribution unlock: Security Copilot to all E5 customers converts an installed base into an agent channel; 1.6M security customers, 1M+ on 4+ workloads.

Bear case

• FCF collapsed to $5.9B (from $25.7B) while capex rose to $37.5B — the finance-lease shield is variable, cash PP&E is catching up, and memory-price inflation will raise future capex. The "FCF holds through the buildout" thesis is broken for now (inference from the sequential print and Hood's lease-mix explanation).
• 45% of the backlog is one counterparty whose durability answer remains relationship rhetoric; GPU-useful-life contracts are concentrated in "largest customers" — the duration defense and the concentration risk are the same contracts (inference from Hood's two answers combined).
• The $10B GAAP gain is an accounting artifact of the OpenAI recapitalization — GAAP earnings quality is now dominated by OpenAI net-asset revaluations in both directions, and ex-OpenAI other income was actually negative and below plan.
• Azure's beat mechanism is efficiency reallocation, not new supply — a finite lever; the guide (37–38% cc) still implies deceleration, and Azure remains the residual claimant after first-party and R&D allocations.
• Cloud GM guided down a third straight time (~65%) with memory costs now building into the depreciation stack for six years.
• Beats again contained pull-forwards: Office 2024 transactional, SQL Server 2025 launch + pre-memory-price server buying (+21% on-prem), elevated OEM inventory (fourth missed normalization) — Q3 guides assume normalization across all three, with OEM guided down ~10%.
• Consumer/legacy softness is broadening: gaming -9% cc with impairments and a first-party content miss; search ex-TAC missed on execution; MPC revenue declining and guided to $12.3–12.8B (down again).
• Q3 opex growth steps up to 10–11% (from +5%) against 15–17% revenue — op margins guided down slightly YoY in H2's first quarter.

Next-quarter watchlist

1. Azure vs 37–38% cc Q3 guide — whether "significantly accelerating" Q3/Q4 capacity additions materialize; any repeat of the >40% all-GPU counterfactual; AI vs core split disclosure.

2. FCF trajectory — does $5.9B recover as finance leases normalize, or is cash PP&E now structurally tracking capex; Q3 capex vs guided sequential decrease; memory-price impact on unit costs.

3. OpenAI accounting — size and direction of the Q3 net-asset-based gain/loss; any disclosure of the recap valuation mechanics; whether the 45% RPO share grows.

4. RPO quality — ex-OpenAI growth vs 28%; duration vs 2.5 years; next-12-months recognition vs +39%; Anthropic commitment size and terms.

5. Cloud GM vs ~65% guide — AI COGS + memory costs vs efficiency gains; pace of decline.

6. M365 Copilot — seat count progression from 15M; whether seat-add acceleration (+160%) holds; any first revenue/ARPU dollarization; E5-attached Security Copilot rollout economics.

7. Windows OEM vs ~-10% guide — fifth attempt at inventory normalization; memory-price impact on the PC market; post-EOS demand cliff.

8. On-prem vs guided low-single-digit decline — how much of the +21% was pull-forward; SQL Server 2025 durability.

9. Gaming — impairment details (size, what was written down); content/services vs mid-single-digit decline guide; Game Pass disclosure.

10. Search ex-TAC vs high-single-digit guide — whether "execution challenges" are fixed; third-party normalization endpoint.

11. FY26 margin math — "up slightly" commitment vs Q3's guided slight YoY decline and opex acceleration to 10–11%; what has to go right in Q4.

12. Capacity language — formal restatement or abandonment of "constrained through end of FY26"; GW-added cadence; Fairwater ramp specifics after Hood's deflection.

13. Foundry metrics — whether $1M+/quarter customers (+80%) and 1T-token customers (250+) become the standing disclosures; total customer count vs 80,000 last quarter.

14. Agent 365 — early adoption evidence for the cross-cloud control-plane claim; partner-integration conversion to revenue.

Quarter in one view

• Revenue $77.7B (+18%/+17% cc); EPS $4.13 (+23%/+21% cc, adjusted for OpenAI investment impact); operating income +24%/+22% cc; operating margin 49%, up YoY and ahead of expectations — "exceeding expectations across revenue, operating income and EPS." Microsoft Cloud revenue $49.1B (+26%/+25% cc), cloud GM 68% (slightly better than expected, down YoY on AI investment).
• Azure +40%/+39% cc vs ~37% cc guide — beat driven again by core infrastructure, "primarily from our largest customers"; Azure AI services only "generally in line" — the third consecutive quarter the upside is non-AI. Demand "again exceeded supply across workloads."
• Commercial bookings +112%/+111% cc, "significantly ahead," driven by Azure commitments from OpenAI plus $100M+ contracts across Azure and M365 — explicitly excludes the incremental $250B OpenAI Azure commitment announced the day before the call. RPO $392B (+51%), nearly doubled in 2 years, weighted average duration ~2 years.
• Capex $34.9B (vs >$30B guide), incl. $11.1B finance leases; cash PP&E $19.4B; ~half short-lived assets (GPUs/CPUs) — and the FY26 capex framework was revised: FY26 capex growth now expected *higher* than FY25, reversing the prior "moderation" guidance. OCF $45.1B (+32%); FCF $25.7B (+33%).
• OpenAI definitive agreement closed (announced prior day): Microsoft has "roughly 10x-ed" its investment; OpenAI contracted incremental $250B of Azure services; rev share, exclusive IP rights and Azure API exclusivity continue until AGI or through 2030; model/product IP rights extended through 2032. Q1 results not impacted by the new deal.
• Other income/expense: $4.1B OpenAI equity-method loss in Q1 (vs -$1.7B total other income last quarter) — Hood confirmed it is "all due to our percentage of losses in OpenAI," nothing else. Going forward, guidance will exclude OpenAI impacts entirely due to expected volatility from the PBC conversion.
• Capacity constraint extended a fourth time: now "capacity constrained through at least the end of our fiscal year" (prior: better shape by December / constrained through H1).

What management is focused on

• The OpenAI deal as the headline: Nadella led with it — 10x return on investment, $250B incremental Azure contract, IP/exclusivity runway to 2030–2032. Framed as "more certainty" on the AGI clause; Nadella stated AGI "as defined at least by us in our contract is ever going to be achieved anytime soon" [sic — meaning not anytime soon].
• Capacity buildout at unprecedented scale: total AI capacity +80% this year; total data center footprint to roughly double over 2 years; Fairwater (Wisconsin) — "world's most powerful AI data center," online next year, scaling to 2GW alone; first large-scale NVIDIA GB300 cluster deployed.
• Fungibility as the risk-management answer: repeated across three answers — fleet spans pretraining to inference, continuously modernized ("ride Moore's Law"), software efficiency (30%+ token throughput gain per GPU on GPT-4.1/GPT-5 this quarter). Used to rebut concentration, bubble, and workload-loss questions.
• Short-lived asset matching as the bubble defense: Hood's core argument — GPU/CPU spend is matched to contract durations; "the lifetime of these and the lifetime of the contracts are very similar"; long-lived assets (15–20 yr leases) are space/power, where confidence of need "is very high."
• Copilot system breadth: 900M MAU touching AI features (from 800M); first-party Copilots surpassed 150M MAU (from 100M+); M365 Copilot chat adoption +50% QoQ; agent users doubled QoQ; Agent Mode, Teams mode, App Builder launched.
• Saying "no" as strategy: Nadella explicitly confirmed Microsoft declines demand that is too concentrated "by customer, by location, by type of skewing" — the first direct acknowledgment of walking away from large AI-native contracts (in response to the lost-deal question).

Key numbers and quarter mechanics

• Total: revenue $77.7B (+18%/+17% cc); company GM 69% (down slightly YoY on AI scaling, offset by Azure/M365 efficiency); opex +5%/+4% cc; op margin 49%; $10.7B returned to shareholders; FX roughly in line.
• OpenAI loss mechanics: $4.1B equity-method loss in other income; ex-OpenAI, other income was +$401M (interest income exceeding expense incl. finance-lease interest). EPS figures given on an adjusted ex-OpenAI basis.
• P&BP $33.0B (+17%/+14% cc): M365 commercial cloud +17%/+15% cc (1 pt from in-period recognition); ARPU led by E5 + Copilot; paid seats +6% (SMB/frontline-led); commercial products +17%/+14% cc (Office 2024 transactional beat again); consumer cloud +26%/+25% cc, subs +7% to >90M; LinkedIn +10%/+9% cc (Talent Solutions still weak); D365 +18%/+16% cc. Segment op margin 62% (+3 pts).
• IC $30.9B (+28%/+27% cc): Azure +40%/+39% cc; on-prem +1%/flat cc (Windows Server 2025 transactional beat); segment GM down YoY on AI investment; op margin 43% (down only slightly).
• MPC $13.8B (+4%): Windows OEM + devices +6% — significantly ahead, driven by Windows 10 end-of-support demand plus still-elevated inventory (normalization missed again); search ex-TAC +16%/+15% cc (third-party partnership benefit better than expected); gaming -2%/-3% cc; Xbox content/services +1%/flat cc (third-party content beat). Segment op margin 30% (+3 pts).
• Cash flow: OCF $45.1B (+32%); capex $34.9B; FCF $25.7B (+33%) — flat sequentially despite +$10.7B capex increase, because finance leases ($11.1B) don't hit cash PP&E.
• Q2 FY26 guide (USD): revenue $79.5–80.6B (+14–16%); COGS $26.35–26.55B (+21–22%); opex $17.3–17.4B (+7–8%); op margins roughly flat YoY, down sequentially (seasonality); other income ~+$100M ex-OpenAI; tax ~19%; FX +2 pts revenue. Cloud GM ~66% (down YoY on AI investment + Azure mix). P&BP $33.3–33.6B (+13–14%); M365 commercial cloud 13–14% cc; consumer cloud mid-20s; LinkedIn ~10%; D365 mid-to-high teens. IC $32.25–32.55B (+26–27%); Azure ~37% cc; on-prem down low-to-mid single digits. MPC $13.95–14.45B; Windows OEM down low-to-mid single digits (inventory unwind assumed again, wider-than-normal range); devices down; search ex-TAC low double digits; Xbox content/services down low-to-mid single digits; hardware down.
• Capex guide: sequential increase in Q2; FY26 growth rate now higher than FY25 — a formal reversal of the prior moderation framework.

Product and launch scorecard

• Azure AI / Foundry: 80,000 customers (from 14,000 Agent Service customers last quarter — metric broadened to all Foundry; not comparable); 80% of Fortune 500; 11,000+ models incl. GPT-5 and Grok 4; new Microsoft Agent Framework (KPMG audit deployment cited); MAI first-party models debuted "among the top" on leaderboards; Phi SLMs 60M+ downloads, 3x YoY. Azure AI services revenue still only in line.
• M365 Copilot: >90% of Fortune 500 now use it; fastest adoption of any M365 suite; chat users +50% QoQ ("tens of millions"); named deals: Accenture, Bristol-Myers Squibb, EY, UK HMRC each 15,000+ seats; Lloyds 30,000 seats (46 min/day saved per employee); PwC added 155,000 seats this quarter to 200,000+ deployed, 30M+ interactions in 6 months — the largest single named deployment yet. "Large majority" of enterprise customers buying more seats. Still no total seats or dollars.
• GitHub Copilot: 26M+ users (from 20M, +30% QoQ); GitHub at 180M+ developers, "fastest growth in its history"; 80% of new developers start with Copilot in week one; 500M+ PRs merged in past year; Agent HQ launched — mission control for third-party coding agents (OpenAI, Anthropic, Google, Cognition, xAI).
• Fabric: revenue +60% (from +55%); 28,000 paid customers (from 25,000 "customers" — the "paid" qualifier restored, +12% QoQ).
• Dragon Copilot: 17M+ patient encounters this quarter, up nearly 5x YoY (from 13M, +31% QoQ); 650+ healthcare orgs purchased ambient listening.
• Security: 100T daily signals; 1B Entra MAU; Purview auditing 16B Copilot interactions (+72% QoQ); 40,000 Sentinel customers; Phishing Triage Agent 6.5x analyst efficiency. Still no security revenue.
• Gaming: record content/services revenue for the quarter; 155M MAU; Minecraft all-time high; Xbox Ally launched; PC player records. No Game Pass figure this quarter.
• Consumer: Copilot app daily users +50% QoQ; Edge share gains 18 consecutive quarters; Bing share gain; M365 Premium consumer subscription launched; Copilot wake word/vision/actions on Windows 11.

Sell-side read-through

• Seven questions; tone congratulatory but the substance was sharper than last quarter — concentration, bubble risk, and the lost-hyperscaler-deal question all surfaced directly.
• Weiss (Morgan Stanley) asked the AGI/zeitgeist question: what could weaken Microsoft's position? Nadella's answer: "jagged intelligence" persists for a long time, so value accrues to organizing-layer systems (M365 Copilot, Agent HQ, security) rather than raw models — a strategic moat argument, and a notable downplaying of near-term AGI despite it being a contractual trigger.
• Thill (Jefferies) on bookings concentration: Hood pointed to RPO breadth across products/customer sizes and the ~2-year weighted duration as evidence of near-term consumption — "people are not consuming unless there's value." No quantified split of OpenAI vs non-OpenAI within the +112%.
• Moerdler (Bernstein) asked the bubble question directly: Hood's defense — spend is against *booked* business, short-lived assets match contract durations, long-lived assets are power/space with very high confidence of use; "we've been short now for many quarters. I thought we were going to catch up, we are not." Nadella added fleet fungibility + ARPU expansion (Copilot ARPU vs M365 ARPU "expansive," like cloud vs server).
• Keirstead (UBS) got the cleanest answer of the call: the $4.1B other-income loss is entirely Microsoft's share of OpenAI losses under the equity method — no accounting change, no deal impact. This is the first explicit quantification of the OpenAI loss line.
• Murphy (JPMorgan) on AI-native counterparty risk (contracts "20x their current revenue scale"): Nadella — fungible fleet, first/third-party balance, enterprise adoption just starting; Hood — gear isn't deployed until contracts begin delivery, giving lead time to assess counterparties. No explicit guardrails or credit-mitigation mechanics disclosed.
• Zelnick (Deutsche Bank) on revenue lost to capacity constraints: Hood — Azure "bears most of the revenue impact" because M365 Copilot, security, GitHub, and internal R&D/product teams are prioritized first; "it is safe to say that the number could be higher" — no quantification.
• Rangan (Goldman) asked about the hyperscaler that "took away the business that was rightfully Microsoft's": Nadella confirmed Microsoft declines demand that is too concentrated by customer/location/sku or single-meter (accelerator-only) — "each time we say no to, the day after, I feel better." This is a de facto confirmation that Microsoft walked from at least one large AI-native deal.
• Nobody asked about: the FY26 capex guidance reversal (moderation → higher growth); the fourth capacity-timeline slip; why guidance now excludes OpenAI (volatility framing accepted without challenge); Copilot dollars.

Management credibility

• For: Azure beat again (39% cc vs ~37% guide) with honest attribution — AI services "in line," upside in core infrastructure, third straight quarter. The $4.1B OpenAI loss was disclosed and directly confirmed as entirely equity-method losses when pressed — a real transparency improvement after two quarters of opacity. The 1-pt in-period recognition benefit in M365 commercial cloud was proactively disclosed. FCF held at $25.7B despite the capex step-up. Fabric's "paid customers" metric was restored after last quarter's wording drift. Windows OEM beat was candidly attributed to elevated inventory rather than normalized demand.
• Against / soft spots: Capacity-balance guidance slipped a fourth time — January: better by June → Q4: hoped by December → now: "constrained through at least the end of our fiscal year." The FY26 capex framework reversed in one quarter — from "growth moderates below FY25" to "growth higher than FY25," with no Q&A challenge. Guidance now excludes OpenAI impacts — understandable given PBC-conversion volatility, but it removes the loss line from the guided P&L just as losses hit $4.1B/quarter. Windows OEM inventory normalization has now been guided and missed three times (still elevated; unwind again assumed for Q2). Foundry's customer metric shifted from Agent Service (14,000) to all-Foundry (80,000) without reconciliation.
• Under pressure: Hood handled the bubble and concentration questions with the most detailed capital-allocation logic given to date (duration matching, booked-business backing, deployment lead times). Nadella's "we say no" answer was candid but confirms large deals were declined — the revenue cost of that discipline is undisclosed. The AGI answer ("not anytime soon") is notable given AGI is a contractual trigger in the new OpenAI agreement.

What changed versus the prior quarter

• OpenAI moved from unquantified drag to center stage: definitive agreement signed ($250B incremental Azure commitment, IP to 2032, exclusivity to AGI/2030, ~10x investment gain); the quarterly equity-method loss was quantified for the first time at $4.1B (vs -$1.7B total other income last quarter); and guidance will henceforth exclude OpenAI entirely.
• Bookings exploded: +37% → +112%/+111% cc, explicitly OpenAI-driven this quarter (last quarter OpenAI was not mentioned in bookings context); RPO $368B → $392B (+51%); new disclosure of ~2-year weighted average duration to counter long-dated-backlog concerns.
• Capex framework reversed: $24.2B → $34.9B print, and FY26 guidance flipped from "growth moderates" to "growth higher than FY25." Finance leases $6.5B → $11.1B; ~half of spend now short-lived assets.
• Capacity constraint extended again: "constrained through H1, better by December" → "constrained through at least the end of FY26" — the fourth slip.
• Azure accelerated to +40%/+39% cc (from +39%), same beat composition (core infrastructure, largest customers; AI services in line). Q2 guided ~37% cc — same as Q1's guide.
• Cloud GM guided down to ~66% (from ~67%) on AI investment plus Azure mix shift.
• FCF flat sequentially ($25.6B → $25.7B) despite +$10.7B capex — finance-lease mix shields cash flow; OCF accelerated to +32%.
• Copilot disclosure stepped up again: 150M+ Copilot MAU (from 100M+), 900M AI-feature MAU (from 800M), PwC 200K seats — but still no revenue.
• Windows OEM beat (+6% vs guided decline) on Windows 10 end-of-support demand — but inventory normalization missed a third time.
• Microsoft acknowledged declining large AI-native contracts for the first time (concentration/sku/location criteria) — a new strategic disclosure prompted by the lost-deal question.

Bull case

• Demand is outrunning every prior frame: bookings +112%, RPO $392B at ~2-year duration (short-dated, consumption-backed per Hood), Azure 39% cc on a ~$80B+ run-rate base with demand "significantly ahead" of capacity — and the $250B OpenAI increment isn't even in these numbers (management statements; synthesis, not a recommendation).
• The OpenAI overhang converted to contracted economics: $250B incremental Azure commitment, IP rights to 2032, exclusivity to 2030/AGI, ~10x investment gain — the biggest uncertainty on the sheet is now a contracted revenue pipeline.
• Capital risk is duration-matched: half of capex is short-lived GPUs/CPUs matched to contract lengths; long-lived spend is power/space with multi-decade utility; finance leases keep FCF intact ($25.7B, +33%) even at $34.9B capex.
• Copilot adoption is compounding with hard proof points: PwC 200K seats, Lloyds 30K with measured time savings, >90% Fortune 500 penetration, chat usage +50% QoQ, agent users doubling QoQ — seat expansion precedes ARPU expansion.
• Margin delivery through the buildout: 49% op margin, P&BP at 62% (+3 pts), MPC at 30% (+3 pts), cloud GM 68% beat — efficiency gains (30%+ tokens/GPU) visibly offsetting AI COGS.
• Platform optionality: Agent HQ positions GitHub as the control layer for *all* coding agents including competitors'; Foundry at 80,000 customers with 11,000+ models — Microsoft monetizes rivals' models either way.

Bear case

• The capacity promise has slipped four times and now stretches to end of FY26 — Azure's ceiling remains supply-set for at least three more quarters, and Hood confirmed Azure "bears most of the revenue impact" of prioritization (inference: unquantified but admitted revenue left on the table).
• Capex discipline framework abandoned in one quarter: FY26 growth now *higher* than FY25 with no crossover framing, cloud GM guided down to ~66%, and total spend rising sequentially — the cost curve is steepening while the revenue guide (Azure ~37% cc) implies deceleration (inference from guidance comparison).
• OpenAI exposure is now enormous on both sides: $4.1B/quarter equity-method losses (growing), bookings growth explicitly OpenAI-driven, a $250B forward commitment from a counterparty whose revenue is a fraction of that — and guidance will no longer include OpenAI impacts, reducing visibility precisely as exposure peaks (inference from Murphy's "20x revenue scale" framing and Hood's exclusion decision).
• Concentration is acknowledged but unquantified: no split of OpenAI vs core within +112% bookings or the RPO balance; Nadella confirmed walking away from rival deals, meaning the OpenAI relationship is being deepened while alternatives are declined.
• Beats again contained non-recurring elements: 1 pt in-period recognition in M365 commercial cloud, Office 2024 transactional, Windows Server 2025 transactional, elevated OEM inventory, third-party search benefit — and Q2 guides assume normalization that has repeatedly failed for OEM.
• Guided decelerations are broad: total revenue 14–16% (from 18%), Azure ~37% cc (from 39% cc), search ex-TAC low double digits (from 16%), gaming content/services negative, OEM negative — Q2 leans on bookings conversion, not acceleration.
• Copilot remains undollarized despite the strongest seat quarter yet — usage metrics keep improving while revenue attribution stays absent.

Next-quarter watchlist

1. Azure vs ~37% cc Q2 guide — core-infrastructure durability; whether capacity additions (Fairwater ramp, GB300s) allow upside; any AI-contribution disclosure.

2. Capacity language — does "constrained through at least end of FY26" hold or slip a fifth time; power/site delivery and finance-lease timing commentary.

3. Capex — Q2 print vs "sequential increase"; FY26 "higher than FY25" trajectory; short-lived vs long-lived mix; finance-lease volume after $11.1B.

4. OpenAI accounting — size of the Q2 equity-method loss (excluded from guidance but still in GAAP results); any PBC-conversion accounting effects; whether the $250B commitment enters bookings/RPO and how it's disclosed.

5. Bookings/RPO quality — core bookings growth "on a low expiry base adjusted for OpenAI"; any quantification of OpenAI's share of RPO; duration stability at ~2 years.

6. Cloud GM vs ~66% guide — AI COGS scaling vs efficiency gains; Azure mix-shift impact.

7. M365 commercial cloud vs 13–14% cc — underlying trend ex recognition; whether PwC-scale seat adds appear in ARPU; any Copilot dollar disclosure.

8. Windows OEM — fourth attempt at inventory normalization (guided down low-to-mid single digits, wide range); post-Windows-10-EOS demand cliff risk.

9. Search ex-TAC vs low-double-digit guide — third-party partnership normalization.

10. FCF durability — whether ~$25.7B holds as cash PP&E catches up with finance-lease-heavy capex; OCF growth vs supplier payments.

11. Concentration disclosures — follow-through on declined-deal strategy; any counterparty guardrails; enterprise vs AI-native mix in Azure growth.

12. FY26 margin mechanics — op margin guided flat YoY in Q2 with seasonal sequential decline; whether the full-year "relatively unchanged" commitment survives the capex revision.

13. Foundry metric — whether the 80,000-customer figure is the new baseline; Agent Service trajectory vs 14,000.

14. Gaming — content/services vs guided low-to-mid-single-digit decline after the record quarter; Game Pass disclosure cadence.

Quarter in one view

• Revenue $76.4B (+18%/+17% cc); EPS $3.65 (+24%/+22% cc); operating income +23%/+22% cc; operating margin 45%, up 2 pts YoY — "significantly exceeded expectations" in the largest quarter of the year. FY25 totals: revenue >$281B (+15%), operating income >$128B (+17%), and Hood confirmed the FY25 operating-margin commitment was exceeded.
• Azure +39% (no cc given), "significantly ahead," vs the 34–35% cc guide — driven by accelerated core (non-AI) infrastructure growth, "primarily from our largest customers." Notably, Azure AI services revenue was only "generally in line with expectations" — the second consecutive quarter where the beat was non-AI. Azure surpassed $75B in annual revenue, +34%.
• Commercial bookings >$100B for the first time, +37%/+30% cc on a strong prior-year comparable — driven by core annuity execution and more $10M+ and $100M+ contracts across Azure and M365. RPO $368B (+37%/+35% cc), +$53B sequentially; ~35% within 12 months (+21%); beyond-12-months +49%; annuity mix 98%.
• Capex $24.2B incl. $6.5B finance leases ($17.1B cash PP&E); Q1 FY26 guided over $30B — the "plateau" is over; spend is re-accelerating. FY26 capex growth will moderate vs FY25 with more short-lived assets, but H1 growth will exceed H2 on finance-lease delivery timing. OCF $42.6B (+15%); FCF $25.6B.
• Capacity constraint extended again: Hood — "I talked about, my gosh, in January and said I thought we'd be in better supply demand shape by June. And now I'm saying I hope I'm in better shape by December." Guided capacity-constrained through H1 FY26.
• Other income -$1.7B, worse than the ~-$1.2B guide, "primarily due to losses on investments accounted for under the equity method" — OpenAI losses still unquantified but visibly growing.
• FY26 guide: double-digit revenue and operating income growth; operating margins "relatively unchanged" YoY; tax 19–20%; FX +2 pts revenue tailwind. Q1: Azure ~37% cc; total implied revenue ~$74.7–75.8B; cloud GM ~67%.

What management is focused on

• Scale as the headline: Microsoft Cloud >$168B annual revenue (+23%); Azure >$75B (+34%) with share gains "every quarter this year"; 2+ gigawatts of new capacity stood up in 12 months; 400+ data centers across 70 regions; every region now AI-first with liquid cooling.
• Software efficiency as the margin answer: GPT-4o-family software optimizations alone deliver 90% more tokens per GPU vs a year ago; Nadella repeated "the difference between a hoster and a hyperscaler is software"; Hood tied this to compounding S-curves from the prior cloud transition.
• Backlog as the capex justification: Hood anchored the >$30B Q1 capex guide to $368B of contracted backlog — "the spend that we're making is correlated to basically contracted on the books business." She explicitly refused to pick a capex/revenue crossover date: "picking a data point usually means you're going to pick to be too conservative in terms of market share gain."
• Migrations as a durable Azure driver: Nestle's 200+ SAP instances / 10,000+ servers / 1.2PB migration cited as one of the largest ever; Nadella said migrations are "at best, maybe in the middle innings," plus cloud-native scaling and AI workloads as the three demand pockets.
• Copilot breadth over Copilot dollars: 100M+ MAU across Copilot apps; 800M+ MAU touching AI features; largest seat-add quarter since launch with named expansions (Barclays 15K→100K, UBS 55K→all employees, Adobe/KPMG/Pfizer/Wells Fargo 25K+ seats each) — still no revenue or total seat count.
• Sovereignty and quantum as new arcs: Microsoft Sovereign Cloud launched (public + private); "world's first operational deployment of a Level 2 Quantum computer" with Atom Computing.

Key numbers and quarter mechanics

• Total: revenue $76.4B (+18%/+17% cc); company GM 69% (-1 pt); opex +6%/+5% cc; op margin 45% (+2 pts); tax ~17%; other income -$1.7B vs ~-$1.2B guided; $9.4B returned in Q4, >$37B for FY25; headcount roughly flat YoY (vs +2% last quarter).
• FX: roughly in line in Q4; FY26 guided +2 pts revenue/COGS, +1 pt opex; Q1 +2 pts total revenue.
• P&BP $33.1B (+16%/+14% cc), ahead: M365 commercial cloud +18%/+16% cc — with 2 points from in-period revenue recognition; underlying trends "relatively stable" ex-recognition; ARPU from E5 + Copilot; paid seats +6% (SMB/frontline-led). Commercial products +9%/+7% cc (Office 2024 transactional beat again). Consumer cloud +20% (January price increase; subs +8%). LinkedIn +9%/+8% cc (Talent Solutions still weak). D365 +23%/+21% cc. Segment op income +21%/+19% cc.
• IC $29.9B (+26%/+25% cc), ahead: Azure +39%; on-prem server -2%/-3% cc (ahead on transactional purchasing with higher in-period recognition); EPS +7%/+6% cc. Segment GM -4 pts YoY on AI scaling; op income +23%.
• MPC $13.5B (+9%), ahead: Windows OEM + devices +3% (inventory still elevated — normalization again deferred); search ex-TAC +21%/+20% cc with ~8 points from third-party partnerships including a low prior-year comp; gaming +10%; Xbox content/services +13%/+12% cc (first-party content, Game Pass). Segment GM +3 pts; op income +34%/+33% cc.
• Cash flow: OCF $42.6B (+15%, cloud billings/collections offset by higher supplier payments); capex $24.2B; FCF $25.6B (vs $20.3B last quarter — the inflection held and grew).
• Q1 FY26 guide (USD): P&BP $32.2–32.5B (+14–15%, ~3 pts FX); IC $30.1–30.4B (+25–26%, ~1 pt FX); MPC $12.4–12.9B. Azure ~37% cc. M365 commercial cloud 13–14% cc; commercial products mid-to-high single digits; consumer cloud low 20s; LinkedIn high single digits; D365 high teens. On-prem down low-to-mid single digits. Windows OEM + devices down mid-to-high single digits (inventory unwind assumed, wider-than-normal range); search ex-TAC low-to-mid teens (partnership benefit normalizing); gaming down mid-to-high single digits, content/services down mid-single digits. COGS $24.3–24.5B (+21–22%); opex $15.7–15.8B (+5–6%); other income ~-$1.3B; tax 19–20%. Cloud GM ~67%. Q1 capex >$30B.

Product and launch scorecard

• Azure AI / Foundry: 500T+ tokens served by Foundry APIs this year, up 7x+ (new metric — narrower than last quarter's 100T+/quarter all-up figure); Foundry Agent Service at 14,000 customers (from 10,000+); 80% of the Fortune 500 use Foundry; 15 OpenAI models sim-shipped same-day; Grok (xAI) added, Black Forest Labs and Mistral "very soon." Azure AI services revenue only in line — the growth is in core infrastructure.
• M365 Copilot: largest seat-add quarter since launch; record customers returning for more seats; named proof points (Barclays 100K, UBS full deployment, four 25K+ seat buyers); "tens of thousands" of orgs used Researcher/Analyst in first weeks. Still no total seats, attach rate, or dollars — but the named-customer specificity is the strongest adoption evidence yet.
• Agents / Copilot Studio: 3M agents created this year via SharePoint + Copilot Studio (vs 1M+ in Q3 alone — annual framing, not comparable QoQ); Copilot Tuning launched; Teams group agents (Facilitator, Interpreter); hundreds of third-party agents (Adobe, SAP, ServiceNow, Workday).
• GitHub Copilot: 20M users (from 15M, +33% in one quarter); enterprise customers +75% QoQ; 90% of Fortune 100; Coding Agent (async) launched; Code Review Agent doing "millions of code reviews each month." Nadella noted rival tools (Claude Code, Codex, Cursor) still drive GitHub repos/PRs — platform benefit regardless of agent used.
• Fabric: revenue +55% YoY, 25,000+ customers (from 21,000 paid, +19% QoQ — note metric wording shifted from "paid customers" to "customers"); "fastest-growing database product in our history."
• Dragon Copilot: 13M+ physician-patient encounters this quarter, up nearly 7x YoY (from 9.5M last quarter — the new quarterly metric persisted and grew ~37% QoQ); Mercyhealth 1,000+ physicians, 100K+ hours saved, expanding to 5,000 providers.
• Security: nearly 1.5M customers (from 1.4M); Sentinel added a modern data lake (350+ connectors); Defender secures ~2M general AI apps; Purview used by 75% of M365 Copilot customers. Still no security revenue figure.
• Gaming: Game Pass annual revenue nearly $5B for the first time (new disclosure); 500M MAU; top publisher on Xbox and PlayStation; Black Ops 6 at 50M players; Minecraft record MAU and revenue; 500M+ cloud-gaming hours this year; ~40 games in development.
• New launches: Microsoft Sovereign Cloud; Copilot Mode in Edge; Level 2 Quantum deployment with Atom Computing.

Sell-side read-through

• Seven questions; tone was euphoric — Weiss: "I don't think I've ever seen a quarter where everything came together this well"; Moerdler: "I didn't know how you were going to beat last quarter, and you did it." Zero challenge on the -$1.7B other-income miss, the OpenAI loss trajectory, or the capex re-acceleration.
• Weiss (Morgan Stanley) asked the most strategic question — AI labs as customers, competitors, and in-sourcing risk: Nadella's answer was historical analogy (Netflix/AWS) plus "head apps shape the platform, then broad diffusion follows," citing Cosmos DB/Postgres learning from ChatGPT workloads. He did not address the in-sourcing or contract-concentration risk directly.
• Rangan (Goldman) pressed the capex-vs-Azure crossover: Hood explicitly refused to frame a pivot point — "I am not as focused on trying to pick a date at which revenue growth and CapEx growth will meet and cross" — and anchored everything to the $368B backlog. This is a deliberate de-commitment from the "capex growth decelerates toward revenue" framing, even as FY26 capex growth is guided below FY25.
• Keirstead (UBS) on migration durability: Nadella gave three drivers (migrations "middle innings," cloud-native scaling including AI-come-stay-for-more customers, new AI workloads) — no quantification of migration's share of the 39%.
• Moerdler (Bernstein) on SaaS AI monetization and margins: Nadella — usage expansion, per-user + consumption blending; Hood — "a blending of these models will continue." No margin trajectory given for SaaS AI specifically.
• Turrin (Wells Fargo) on flat FY26 margins: Hood credited revenue growth, S-curve efficiency, and talent focus — no quantified AI-internal-productivity offset despite the question inviting one.
• Nobody asked about: the other-income miss vs guide and the growing equity-method (OpenAI) losses; the absence of the AI-points-of-Azure-growth disclosure (dropped this quarter after Hood flagged it was getting hard to separate); the $13B AI run rate (gone for a second quarter); Copilot dollars; or the Windows OEM inventory normalization that has now been guided and missed twice.

Management credibility

• For: Azure beat guide again (39% vs 34–35% cc) and the beat composition was again honestly attributed — Hood stated plainly that Azure AI services were only "in line" and the upside was core infrastructure, resisting the easy AI narrative. The FY25 operating-margin commitment was exceeded. FCF delivered $25.6B, confirming the Q3 inflection was not a one-off. M365 Copilot disclosure improved in specificity (named customers, seat counts per deal) even without dollar figures. The 2-point in-period recognition benefit in M365 commercial cloud was proactively disclosed.
• Against / soft spots: The capacity-balance date slipped a third time — January: "better shape by June"; now: "I hope I'm in better shape by December," with constraints guided through H1 FY26. Hood owned the slip explicitly, but the pattern stands. Other income missed guide by ~$500M (-$1.7B vs ~-$1.2B) on equity-method losses with no quantification or explanation of the OpenAI loss trajectory. The AI-points disclosure disappeared one quarter after Hood said AI/non-AI separation was getting harder — the forecast became the fact, reducing Azure transparency at the peak. Windows OEM inventory normalization was guided for Q4 and did not happen (still +3%, "elevated"); it is now guided for Q1 with a wider range. The Fabric metric shifted from "paid customers" to "customers" without explanation.
• Under pressure: Rangan's crossover question got the most revealing answer of the call — Hood's refusal to pick a date is candid but also removes the anchor investors used to model capex deceleration. Weiss's AI-lab-risk question got analogy, not risk management.

What changed versus the prior quarter

• Azure accelerated again: +33%/+35% cc → +39%, with the driver rotating from "non-AI execution repair" to "core infrastructure, primarily largest customers." The AI-points disclosure (13 → 16) was dropped entirely — no AI contribution figure given, and AI services revenue described as merely in line.
• Capex re-accelerated: $21.4B → $24.2B, with Q1 guided >$30B — the "plateau" era lasted two quarters. FY26 framing: growth below FY25 but H1 > H2, more short-lived assets, justified by $368B backlog rather than a revenue-crossover date.
• Capacity endgame slipped again: "a little short as we exit the year / constraints beyond June" → "constrained through H1 FY26," balance hoped for by December.
• Bookings stepped up massively: +18% → +37%/+30% cc, first >$100B quarter; RPO $315B → $368B (+$53B sequential, vs +$17B last quarter); beyond-12-months +47% → +49%. Driver shifted from a named OpenAI commitment to core annuity + large $10M/$100M+ contracts — OpenAI was not mentioned in the bookings context at all.
• Other income deteriorated: -$623M → -$1.7B, missing the -$1.2B guide — equity-method losses (OpenAI) are growing and still unquantified.
• FCF kept climbing: $20.3B → $25.6B even as capex rose — OCF +15% funded it.
• Headcount went from +2% YoY to roughly flat — efficiency posture tightening.
• M365 commercial cloud printed 18% but with a disclosed 2-pt in-period recognition benefit — underlying ~16% cc, stable; Q1 guided 13–14% cc, a deceleration.
• Search's third-party partnership was quantified for the first time (~8 pts of the +21%) and guided to normalize — ex-TAC decelerates to low-to-mid teens.
• Gaming guided negative (mid-to-high single-digit decline) after the strongest content quarter in years — tough comps.
• New annual disclosures: Azure >$75B (+34%), Microsoft Cloud >$168B (+23%), Game Pass ~$5B, 2GW capacity added, 500T Foundry tokens.

Bull case

• Azure at 39% on a $75B+ base with demand still above supply — and the upside is core infrastructure from the largest customers, the highest-quality kind, while AI services merely being "in line" means AI upside is not even in the print (management statements; synthesis, not a recommendation).
• The backlog is the balance sheet of the AI buildout: $368B RPO (+37%), long end +49%, first $100B+ bookings quarter — Hood's explicit argument that capex is "correlated to contracted on-the-books business" is now numerically supported.
• FCF inflection confirmed and compounding: $6.5B → $20.3B → $25.6B across three quarters, with OCF +15–16% funding a rising capex line — the operating leverage thesis survived the capex re-acceleration.
• Copilot adoption evidence got concrete: largest seat-add quarter, Barclays 15K→100K, UBS to full deployment, four 25K+ seat deals, GitHub Copilot enterprise customers +75% QoQ — the seat-expansion motion is scaling into the largest enterprises.
• Efficiency is quantified and accelerating: 90% more tokens per GPU via software alone; cloud GM of 68% beat despite AI scaling; FY26 op margins guided flat while investing — margin discipline holding through the buildout.
• Breadth: D365 +23%, consumer cloud +20%, MPC op income +34%, Game Pass ~$5B, security at ~1.5M customers — the non-Azure portfolio is contributing growth and margin simultaneously.

Bear case

• The capacity promise has now slipped three times (end-FY25 → beyond June → December/H1 FY26). Each slip is demand-driven, but Azure's ceiling remains supply-set, and the >$30B Q1 capex guide shows the cost of chasing it is rising, not falling (inference from three quarters of evolving guidance).
• Capex re-acceleration without a crossover framework: Hood explicitly declined to identify when capex and revenue growth converge, and H1 FY26 capex growth will exceed H2 — the "FY26 moderation" could still mean a very large absolute number against cloud GM guided down to ~67% (inference).
• OpenAI exposure is growing on both sides of the ledger: equity-method losses drove a ~$500M other-income miss (-$1.7B) with no quantification, while Azure's acceleration is attributed to "largest customers" — concentration risk is rising precisely as disclosure of it shrinks (inference; OpenAI not named in the Azure growth context).
• Azure transparency regressed at the peak: the AI-points disclosure vanished one quarter after management flagged it was hard to separate; the $13B AI run rate is gone for a second quarter; Fabric's "paid customers" became "customers." Each change reduces falsifiability as numbers get large.
• Q4 beats again contained non-recurring elements: 2 pts of M365 commercial cloud from in-period recognition, ~8 pts of search ex-TAC from third-party partnerships, Office transactional purchasing, elevated Windows OEM inventory — and Q1 guides assume all of these normalize, the same assumption that failed twice for OEM.
• Guided decelerations are broad: M365 commercial cloud 13–14% cc (from 16% underlying), search ex-TAC low-to-mid teens (from 21%), gaming negative, Windows OEM negative — Q1's growth rests heavily on Azure holding ~37% cc.
• Copilot monetization remains undollarized four-plus quarters post-launch despite the strongest adoption quarter — the gap between usage evidence and revenue evidence persists.

Next-quarter watchlist

1. Azure vs the ~37% cc Q1 guide — whether core-infrastructure momentum from largest customers sustains; whether any AI-contribution disclosure returns or is permanently retired; capacity-delivery timing as a swing factor (Hood flagged quarterly variability explicitly).

2. Capacity language — does "constrained through H1" and "better shape by December" hold, or slip a fourth time? Watch power/DC-space and finance-lease delivery commentary.

3. Capex — Q1 print vs the >$30B guide; finance-lease mix; whether FY26 "moderation" survives contact with H1 demand; any update to the short-lived/long-lived split.

4. Other income vs ~-$1.3B guide — any quantification of OpenAI equity-method losses after the -$1.7B miss; trajectory of the loss line as OpenAI scales.

5. Bookings and RPO — growth on a growing expiry base after the $100B quarter; whether large-contract mix (and any OpenAI tranches) re-emerges; beyond-12-months after +49%.

6. Cloud GM vs ~67% guide — pace of AI COGS scaling against the flat-FY26-op-margin commitment; whether efficiency gains (90% tokens/GPU) visibly offset.

7. M365 commercial cloud vs 13–14% cc guide — underlying trend ex the 2-pt recognition benefit; whether record Copilot seat adds show up in ARPU; any dollar/seat disclosure after the named-deal quarter.

8. Windows OEM inventory unwind — third attempt at normalization (guided down mid-to-high single digits, wide range); tariff commentary.

9. Search ex-TAC vs low-to-mid-teens guide — whether the third-party partnership benefit actually normalizes as guided.

10. Gaming vs guided decline — content/services down mid-single digits against the strongest comp; Game Pass trajectory after the ~$5B disclosure.

11. FCF durability — whether $25.6B holds as capex steps to $30B+; OCF growth vs supplier-payment pressure.

12. AI-lab concentration — any disclosure on customer concentration within Azure's acceleration; follow-through on Weiss's in-sourcing/competition question.

13. Fabric metric — whether "customers" vs "paid customers" is clarified; paid-count trajectory vs 25,000.

14. FY26 margin mechanics — any quantification of internal AI productivity gains or opex leverage behind the flat-margin guide as the year begins.

Quarter in one view

• Revenue $70.1B (+13%/+15% cc); EPS $3.46 (+18%/+19% cc); operating income +16%/+19% cc; operating margin 46%, up 1 pt YoY — beat across the board, with Hood attributing it to "focused execution from our sales and partner teams." Opex +2%/+3% cc, below expectations (cost efficiency plus investments shifted to Q4); headcount +2% YoY, down slightly QoQ.
• Microsoft Cloud $42.4B (+20%/+22% cc), ahead of expectations; cloud GM 69%, down 3 pts YoY on AI scaling, in line.
• Azure +33%/+35% cc — a re-acceleration and beat vs the 31–32% cc guide — with 16 points from AI (up from 13). Critically, Hood clarified in Q&A that the outperformance was in non-AI (enterprise acceleration, scale-motion improvement), while AI upside came only from capacity delivered early. The Q2 non-AI execution miss was repaired faster than the "through H2" framing implied.
• Capacity constraint extended: Hood had "hoped to be in balance by the end of Q4" but now expects to be "a little short, still, a little tight as we exit the year," and guided AI capacity constraints beyond June — the end-of-FY25 balance commitment from last quarter slipped on demand growing "a bit faster" than supply.
• Bookings +18%/+17% cc, "significantly ahead," again driven by an Azure commitment from OpenAI plus core annuity execution; RPO $315B (+34%/+33% cc), +$17B sequentially; ~40% within 12 months (+17%); beyond-12-months +47%; annuity mix 98%.
• Capex $21.4B incl. leases ($16.7B cash PP&E), slightly below expectations on lease timing; Q4 guided up sequentially, with H2 total unchanged from January guidance. OCF $37B (+16%); FCF $20.3B — a dramatic reversal from -29% YoY last quarter.
• Other income -$623M, better than the ~-$1B guide (net gains on derivatives/investments; equity-method losses "slightly higher than expected" — OpenAI losses still not quantified).
• Q4 guide: Azure 34–35% cc; total revenue implied by segments ~$73.2–74.3B; FX now a 1-pt tailwind (April dollar weakness); cloud GM ~67%; FY25 op margins still guided up slightly; FY26 capex framework unchanged (growth below FY25, more short-lived assets).

What management is focused on

• Efficiency as the demand engine, with new hard metrics: dock-to-lead times for new GPUs down nearly 20%; AI performance up nearly 30% ISO power across the blended fleet; cost per token more than halved; model capability doubling every six months. Nadella's framing: each cloud generation's advantage over the last "is efficiency."
• Non-AI Azure execution as the repaired story: Hood repeatedly credited "focused execution" — enterprise-segment acceleration and scale-motion improvement — and was careful (correcting Keirstead's premise) to say the beat was not an AI demand surprise.
• Capacity honesty with a slipped date: opened DCs in 10 countries across four continents this quarter; acknowledged being "short power" (i.e., DC space) in specific locations; pulled space readiness earlier; but conceded constraints now run beyond June.
• Token volume as the new AI scale proof: 100+ trillion tokens processed this quarter, up 5x YoY, with a record 50 trillion in the last month alone — a new disclosure framing AI consumption.
• Copilot as an agent platform: "hundreds of thousands" of customers (up 3x YoY), record seat-expansion quarter, Researcher/Analyst deep-reasoning agents, role-based agents (Sales, Customer Service), 1M+ custom agents created this quarter (+130% QoQ), Copilot Studio at 230,000+ orgs including 90% of the Fortune 500.
• Gaming margin expansion continues: "focus on margin expansion," 500M+ MAU, PC Game Pass revenue +45%, record 150M+ cloud-gaming hours, Minecraft movie halo (+75% weekly active users).
• Recession positioning (Moerdler question): Nadella's answer — software as "the most malleable resource" to fight inflation/growth pressure; help customers do more with less and take share. No demand-signal deterioration acknowledged; Hood said April demand signals "remained consistent."

Key numbers and quarter mechanics

• Total: revenue $70.1B (+13%/+15% cc); company GM 69% (-1 pt YoY, AI scaling); opex +2%/+3% cc (below expectations; some investments shifted to Q4); op margin 46% (+1 pt); tax ~18%; other income -$623M vs ~-$1B guided (derivatives/investment gains; equity-method losses slightly worse than expected); $9.7B returned (+15%); headcount +2% YoY, down slightly QoQ ("reducing layers with fewer managers").
• FX: roughly in line in Q3 (COGS 1 pt unfavorable vs expectations); Q4 FX now a +1 pt revenue tailwind on April dollar weakness — a swing from the 2-pt headwind guided last quarter.
• P&BP $29.9B (+10%/+13% cc), ahead on LinkedIn, M365 commercial products, and M365 consumer: M365 commercial cloud +12%/+15% cc (in line; E5 + Copilot ARPU); paid seats +7% to over 430M (SMB/frontline-led); commercial products +5%/+8% cc (ahead on Office transactional purchasing); consumer cloud +10%/+12% cc (ahead on January price increase; subs 87.7M, +9%); LinkedIn +7%/+8% cc (ahead across all businesses; Talent Solutions still weak); D365 +16%/+18% cc (in line). Segment op income +15%/+18% cc.
• IC $26.8B (+21%/+22% cc), ahead on Azure: Azure +33%/+35% cc with 16 pts from AI; non-AI accelerated in enterprise, scale motions improved; AI capacity landed early. On-prem server -6%/-4% cc (slightly below; lower in-period recognition on renewal mix); EPS +5%/+6% cc (slightly ahead). Segment GM -4 pts YoY; op income +17%/+18% cc.
• MPC $13.4B (+6%/+7% cc), ahead across all businesses: Windows OEM + devices +3% (tariff uncertainty kept inventories elevated — the expected normalization did not happen); search ex-TAC +21%/+23% cc (significantly ahead — third-party partnership usage again, plus rate expansion and volume); gaming +5%/+6% cc; Xbox content/services +8%/+9% cc (ahead on first- and third-party content). Segment GM +2 pts; op income +21%/+23% cc.
• Cash flow: OCF $37B (+16%, strong cloud billings/collections, offset by higher tax payments); capex $21.4B; FCF $20.3B (vs $6.5B and -29% YoY last quarter — the trough thesis confirmed for now).
• Q4 guide (USD): P&BP $32.05–32.35B (+11–12% cc); IC $28.75–29.05B (+20–22% cc); MPC $12.35–12.85B (widened range for OEM inventory variability). Azure 34–35% cc. M365 commercial cloud ~14% cc; commercial products mid-single digits; consumer cloud mid-teens; LinkedIn high single digits; D365 mid-to-high teens. On-prem server down mid-single digits; EPS mid-to-high single digits. Windows OEM down low-to-mid single digits (inventory normalization assumed, wider outcome range), devices down high teens; search ex-TAC high teens; gaming mid-single digits, content/services high single digits. COGS $23.6–23.8B (+19–20% cc); opex $18.0–18.1B (~+5% cc); other income ~-$1.2B (equity-method investments); tax ~19%. Cloud GM ~67% (down YoY). Q4 capex up sequentially; H2 total unchanged. FY25 op margins up slightly. FY26 capex framework unchanged.

Product and launch scorecard

• Azure AI: 16 points of Azure growth (from 13); 100T+ tokens processed in the quarter (+5x YoY), 50T in the final month alone — new consumption disclosure; capacity delivered early to customers; Foundry now at 70,000+ enterprises/digital natives (new enterprise-count framing vs 200K MAU last quarter); Agent Service at 10,000+ orgs four months in; Phi downloaded 38M times (from 20M); BitNet b1.58 (CPU-only 1B-param model) coming to Foundry; models from OpenAI, Cohere, DeepSeek, Meta, Mistral, Stability.
• M365 Copilot: "hundreds of thousands of customers," up 3x YoY; record number of customers returning to buy more seats; deal sizes growing; major update announced (agents, notebooks, search, create; Researcher and Analyst deep-reasoning agents). Still no total seats, attach rate, or dollar figure — and notably fewer hard cohort metrics than last quarter's >10x expansion disclosure.
• Copilot Studio / agents: 230,000+ orgs (from 160,000+), 90% of the Fortune 500; 1M+ custom agents created this quarter across SharePoint and Copilot Studio, +130% QoQ (vs 400K cumulative in three months last quarter); "computer use" agents; SharePoint-site-to-agent conversion.
• GitHub Copilot: 15M+ users, up 4x+ YoY (first user-count disclosure); agent mode in VS Code; Code Review Agent has reviewed 8M+ PRs; SWE agent in preview; VS Code/Visual Studio 50M+ MAU. Still no revenue figure.
• Fabric: 21,000+ paid customers (+80% YoY; from 19,000, +11% QoQ); real-time intelligence is the fastest-growing workload (40% of customers using it five months post-GA); 50%+ of customers use 3+ workloads; OneLake data up 6x YoY.
• Dragon Copilot (DAX successor): ~9.5M physician-patient encounters documented last quarter, +50%+ QoQ — a step-change from the 2M monthly-encounters metric (scope/rebranding change not explained); City of Hope, Ottawa Hospital, Tufts, WellStar.
• Data: PostgreSQL usage accelerated a third straight quarter (~60% of Fortune 500); Cosmos DB revenue growth accelerated again (OpenAI named as a customer).
• Security: 1.4M security customers; 900,000+ with 4+ workloads (+21% YoY); Entra 900M+ MAU; Security Copilot agents launched; 84T daily threat signals. Still no security revenue figure.
• LinkedIn: 1B+ members, double-digit growth; video time +36%, comments +32%; AI coaching learners +2x QoQ; Premium Pages SMB subscribers +75% QoQ; Marketing Solutions accelerated for a second consecutive quarter.
• Advertising: total advertising revenue surpassed $20B over the last 12 months — new disclosure; Bing/Edge share gains again; Copilot Search/Vision/Discover.
• Gaming: 500M+ MAU; top publisher by pre-orders/pre-installs on Xbox and PlayStation; PC Game Pass revenue +45% YoY; cloud gaming record 150M+ hours; Copilot for Gaming and Muse generative-gameplay model; Minecraft movie top-grossing film of the year with +75% game WAU. Content/services +8%/+9% cc — best in several quarters.
• Windows: Copilot+ PC exclusive features (Recall, Click to Do) rolled out broadly; Windows 11 commercial deployments +~75% YoY ahead of Windows 10 EOS.

Sell-side read-through

• Eight questions; tone was congratulatory throughout ("fantastic quarter," "inspiring," "amazing") — a sharp contrast to prior quarters' capex/ROI pressure. No one challenged guidance, margins, or the OpenAI relationship.
• Weiss (Morgan Stanley) pressed the only contentious topic — reports of Microsoft walking away from DC commitments: Nadella framed build/lease adjustments as continuous and unremarkable ("you all pay a lot more attention nowadays"), emphasizing workload shape, location, and compounding S-curves; conceded being "short power" in specific places. Hood added 5–7 year land-to-build lead times and admitted the hoped-for Q4 balance slipped: "a little short, still, a little tight as we exit the year."
• Thill (Jefferies) on migration acceleration: Nadella decomposed Azure demand into four pockets — classic migration, data growth (Postgres, Cosmos, Fabric, even Databricks/Snowflake on Azure), cloud-native compute, and the AI-to-non-AI ratio (ChatGPT itself consumes Cosmos, Postgres, core compute/storage).
• Moerdler (Bernstein) asked the recession question directly: Nadella gave a share-gain/help-customers answer with no acknowledgment of demand softening; Hood's guidance preamble ("demand signals... remained consistent" through April) is the only macro hedge.
• Keirstead (UBS) misread the Azure beat as AI-driven and was corrected: Hood — "the real outperformance in Azure this quarter was in our non-AI business"; AI upside was only early supply delivery. She did not answer whether 16 AI points could rise in June — a notable dodge given the Q4 guide implies it.
• Rangan (Goldman) on capital efficiency: Hood claimed AI margins are "better than they were at this point, by far" versus the equivalent stage of the server-to-cloud transition — a new, unquantified claim. Nadella: hyperscale vs hosting "is software."
• Murphy (JPMorgan) on DeepSeek and GPU useful life: Nadella reiterated ~10x software-driven improvement per Moore's Law cycle; Hood said depreciable-life changes require "a long history" — no change coming, but framed as a software question, leaving the door open.
• Materne (Evercore) on non-AI durability: Hood — enterprise-segment acceleration, scale motions "a little better" with "still some work to do"; consistent across geos; improved execution, nothing beyond that.
• Zukin (Wolfe) on AI/non-AI blur: Hood — "it's getting harder and harder to separate what an AI workload is from a non-AI workload," signaling the AI-points disclosure may not persist in current form.
• Nobody asked about: the OpenAI commitment's size or tranche structure (second consecutive quarter driving bookings), the unquantified OpenAI equity-method losses, the missing $13B AI run-rate update, Copilot dollars, or the Dragon Copilot metric change.

Management credibility

• For: The Q2 non-AI execution problem was fixed ahead of the "through H2" timetable — Hood owned the miss last quarter and delivered enterprise acceleration and scale-motion improvement this quarter, and proactively corrected Keirstead to keep the credit accurate rather than let the AI narrative inflate. Azure beat the 31–32% cc guide at 35% cc. FCF recovered to $20.3B as the capex plateau took hold. Capex came in slightly under and H2 guidance was held unchanged despite the Q4 sequential increase — the plateau commitment is holding. Opex discipline continued (+2%, investments shifted to Q4 disclosed). FX guidance swung from headwind to tailwind and was updated transparently.
• Against / soft spots: The capacity-balance commitment slipped — "roughly in line with near-term demand by end of FY25" became "a little short... as we exit the year" and "AI capacity constraints beyond June." This is the second dated capacity promise to move, though this time the cause is demand upside plus early deliveries, not slippage. The $13B AI run rate was not updated or defined — the milestone metric from last quarter simply disappeared, replaced by token volumes. Copilot disclosure regressed from cohort economics (>10x expansion, price per seat "quite good") to softer counts ("hundreds of thousands of customers, up 3x"). The Dragon Copilot 9.5M-encounter figure is not comparable to the prior 2M-monthly metric, with no bridge given. OpenAI losses remain unquantified inside a better-than-guided other-income line.
• Under pressure: Weiss got a candid capacity answer including the slipped balance date — accountable but the slip itself is the news. Keirstead's follow-up on whether AI points rise in June went unanswered. Hood's AI-margin-superiority claim (Rangan) is unverifiable as stated.

What changed versus the prior quarter

• Azure re-accelerated and beat: 31% (Q2) → +33%/+35% cc vs 31–32% cc guide; AI contribution 13 → 16 points. The miss driver rotated back out — non-AI went from self-inflicted miss to the source of outperformance, with the H2-long repair finishing early.
• The capacity endgame slipped: "roughly in line by end of FY25" → constraints "beyond June," attributed to demand growing faster than supply plus early deliveries.
• FCF inflected hard: $6.5B (-29% YoY) → $20.3B, on OCF +16% and a slightly-below-expectations capex print ($21.4B vs $22.6B).
• Bookings normalized but stayed ahead: +67%/+75% cc → +18%/+17% cc (vs flat guide), still OpenAI-driven; RPO $298B → $315B (+$17B sequential); beyond-12-months +45% → +47%.
• FX swung from a 2-pt headwind to a +1 pt Q4 tailwind on April dollar weakness.
• Other income improved: -$2.3B (Cruise impairment) → -$623M, better than guide; Q4 guided ~-$1.2B with equity-method losses as the stated driver.
• AI proof metric rotated: the $13B run rate vanished; replaced by 100T+ tokens (+5x YoY) and Foundry enterprise counts. AI-points disclosure itself was flagged as increasingly hard to separate.
• Copilot disclosure softened: from cohort economics and named seat counts to "hundreds of thousands of customers, up 3x" and record expansion — directionally positive, less falsifiable.
• Windows OEM stayed elevated (+3%) as tariff uncertainty kept inventories high; normalization pushed to Q4 (guided down low-to-mid single digits, widened MPC range).
• Search's third-party partnership did not normalize — ex-TAC +21%/+23% cc again, guided high teens for Q4 (raised from mid-teens pattern).
• Gaming content/services improved to +8%/+9% cc (from +2%), best since the Activision close dynamics, with PC Game Pass revenue +45%.
• Cloud GM guided down to ~67% for Q4 (from ~69% actuals) — AI COGS scaling accelerates even as FY25 op-margin-up guidance is maintained.

Bull case

• Azure beat and re-accelerated with the cleanest possible composition: the upside came from non-AI execution (enterprise acceleration, scale-motion repair) — the exact weakness bears cited — while AI's 16 points were delivered with capacity still constrained and Q4 guided 34–35% cc (management statements; synthesis, not a recommendation).
• Demand exceeds supply with a visible backlog: $315B RPO, long end +47%, bookings ahead for a second straight quarter, and constraints extended because demand grew faster — the supply-limited thesis now has management's explicit confirmation.
• The capex plateau held and FCF inflected: $21.4B, H2 unchanged, FY26 growth below FY25 with revenue-correlated mix — and FCF jumped to $20.3B, demonstrating the operating leverage of the plateau in a single quarter.
• Efficiency metrics are now quantified and compounding: token cost more than halved, +30% AI performance per unit power, GPU dock-to-lead -20% — supporting Hood's claim that AI margins already exceed the comparable cloud-transition stage.
• The agent layer is scaling: 1M+ custom agents in a quarter (+130% QoQ), 230K Copilot Studio orgs, 90% Fortune 500 penetration, 15M GitHub Copilot users (4x YoY) — monetization surface broadening beyond seat-based Copilot.
• Breadth of beats: LinkedIn, consumer (price increase working), gaming content, search rate expansion, and MPC op income +21%/+23% cc show the non-Azure portfolio contributing, with a $20B advertising TTM milestone.

Bear case

• The capacity promise slipped again — "in line by end of FY25" became "constraints beyond June." Even if demand-driven, the pattern of moving dates (Q1 vendor slippage, Q2 execution, Q3 demand upside) means Azure's ceiling keeps being set by supply Microsoft doesn't control on schedule (inference from three quarters of evolving explanations).
• Cloud GM guided to ~67% for Q4, down from 69% — AI COGS scaling is accelerating into the margin line even as opex discipline holds the operating line; the FY26 margin math with continued capex growth is unaddressed.
• Bookings remain OpenAI-dependent: two consecutive quarters where an OpenAI Azure commitment is the named driver, with no size, tranche structure, or terms disclosed, and OpenAI's losses still buried in an unquantified equity-method line (Q4 other income guided to -$1.2B primarily for this).
• Key disclosures are rotating away from falsifiability: the $13B AI run rate disappeared one quarter after being the headline; Copilot metrics softened from cohort economics to customer counts; Hood signaled AI/non-AI Azure points may become inseparable — each change reduces verifiability just as numbers get large (inference from disclosure pattern).
• Q3 beats again contained pull-forward elements: Windows OEM stayed elevated on tariff-driven inventory (guided to decline, wider range); search's third-party partnership persisted; Office transactional purchasing beat again. Q4 guides assume normalization — the same assumption that failed this quarter for OEM.
• Copilot monetization remains undollarized three-plus quarters in, and the encounter-metric change for Dragon Copilot (2M monthly → 9.5M quarterly) obscures trend without a bridge.
• Macro hedging is thin: the entire Q4 guide rests on "demand signals remained consistent through April" — one month of data in a tariff-volatile environment, with the MPC range already widened for uncertainty.

Next-quarter watchlist

1. Azure vs the 34–35% cc Q4 guide — AI point contribution vs 16 (and whether Hood's non-AI/AI separation comments foreshadow disclosure changes); non-AI enterprise momentum and scale-motion progress.

2. Capacity constraint language — does "beyond June" get a new end date, and does the slipped FY25 balance commitment get re-committed for FY26? Watch power/DC-space commentary specifically.

3. AI run-rate disclosure — whether the $13B figure returns, is updated, or is permanently replaced by token volumes; any definition of composition.

4. Bookings and OpenAI — size/structure of any further OpenAI tranches vs core annuity motions; RPO beyond-12-months after +47%; Q4 bookings vs "solid growth on a significant prior-year comparable."

5. Capex — Q4 sequential increase vs the unchanged H2 total; FY26 guidance detail (growth rate, short-lived mix) as the year turns; any depreciable-life commentary following Murphy's question.

6. Cloud GM vs ~67% Q4 guide — the pace of AI COGS scaling against the FY25 op-margin-up commitment and into FY26.

7. FCF durability — whether $20.3B was a collections/tax-timing fluke or the new run rate as capex plateaus.

8. Other income vs ~-$1.2B guide — any quantification of OpenAI equity-method losses; further impairments.

9. M365 Copilot — whether record seat expansion converts to any dollar/seat disclosure; M365 commercial cloud vs ~14% cc guide; seat-growth moderation as flagged.

10. Windows OEM normalization — whether elevated tariff-driven inventory finally unwinds (guided down low-to-mid single digits, wide range); devices down high teens; tariff commentary.

11. Search ex-TAC vs high-teens guide — whether the third-party partnership ever normalizes; rate expansion durability.

12. Gaming — content/services vs high-single-digit guide; whether PC Game Pass +45% and Minecraft momentum sustain; any subscriber disclosure.

13. Dragon Copilot metric — whether the 9.5M-encounter figure is bridged to the old DAX metric or restated going forward.

14. Macro — any change from "April demand signals consistent" given tariff escalation risk; recession-scenario follow-through on Nadella's share-gain framing.

Quarter in one view

• Revenue $69.6B (+12%); EPS $3.23 (+10%); operating income +17%/+16% cc; operating margin 45%, up 2 pts YoY — better than expected, driven by efficiency while scaling AI. Opex +5%, below expectations; headcount +2% YoY, flat QoQ.
• Microsoft Cloud $40.9B (+21%), first time above $40B; cloud GM 70%, down 2 pts YoY on AI scaling, in line.
• Azure +31% (cc not separately stated for the print; Q3 guided 31–32% cc), with 13 points from AI (AI services +157% YoY, ahead of expectations). Non-AI was slightly below expectations due to go-to-market execution challenges in the "scale motion" (partner/indirect customers) — a new, self-inflicted miss driver, distinct from last quarter's third-party capacity slippage. Still AI-capacity constrained; Hood now says "by the end of FY25 we should be roughly in line with near-term demand."
• AI business run rate surpassed $13B, up 175% YoY, "above expectations" — the $10B milestone committed last quarter was cleared and raised; scope still undefined.
• Bookings +67%/+75% cc, "significantly ahead," driven by a new large Azure commitment from OpenAI plus broad core-motion strength; RPO $298B (+34%/+36% cc), +$39B sequentially (largest ever per Bracelin); ~40% within 12 months (+21%); beyond-12-months +45%; annuity mix 97%.
• Capex $22.6B incl. leases ($15.8B cash PP&E), in line; guided flat at Q2 levels for Q3/Q4. OCF $22.3B (+18%); FCF $6.5B, down 29% YoY.
• Other income -$2.3B, worse than the ~-$1.5B guide, due to a Cruise impairment charge (OpenAI equity-method losses embedded but not separately quantified this quarter).
• Q3 guide: Azure 31–32% cc; total revenue implied by segments ~$67.9–68.7B; FX now a 2-pt revenue headwind (~$1B worse than October assumptions); cloud GM ~69%.

What management is focused on

• Reframing the capex debate around efficiency and Jevons-style demand: Nadella opened with scaling laws compounding across pre-training and inference, >2x price-performance per hardware generation, >10x per model generation from software — "as AI becomes more efficient and accessible, we will see exponentially more demand." Fleet management is now "a more continuous cycle governed by both revenue growth and capability growth."
• The $13B AI run rate as the monetization proof point — led the prepared remarks, "above expectations."
• Owning the non-AI Azure miss as an execution problem: Hood repeatedly attributed the shortfall to "go-to-market execution challenges" in the scale/indirect motion from summer sales-motion changes balancing AI vs migration work; fixes underway but impact persists "through H2."
• OpenAI as demand, not just drag: the "new large Azure commitment" is the headline bookings driver; Nadella emphasized API exclusivity on Azure, the ROFR, and "a lot more coming soon." Stargate framed as Microsoft keeping a fungible fleet rather than funding OpenAI's training buildout.
• Copilot expansion metrics: first-availability cohort expanded seats >10x over 18 months; daily users doubled QoQ again; usage intensity +60% QoQ; Copilot Chat broadens TAM.
• Gaming profitability over growth: "focused on improving the profitability of the business... higher margin content and platform services" — MPC op income +32% on mix and execution.
• Capacity endgame dated: more than doubled DC capacity in three years; added more capacity last year than any prior year; roughly in line with near-term AI demand by end of FY25.

Key numbers and quarter mechanics

• Total: revenue $69.6B (+12%); GM 69%, up slightly YoY (mix shift to higher-margin businesses, Gaming/Search improvement, offset by AI scaling); opex +5% (below expectations); op margin 45% (+2 pts); tax 18% (slightly below); other income -$2.3B vs ~-$1.5B guide — Cruise impairment; $9.7B returned; headcount +2% YoY, flat QoQ.
• FX: not significant in Q2; Q3 FX now a 2-pt revenue headwind (~$1B worse than October), 2 pts on COGS, 1 pt on opex; Q4 FX expected to cut revenue/COGS growth by >1 pt.
• P&BP $29.4B (+14%/+13% cc), ahead on M365 Commercial: M365 commercial cloud +16%/+15% cc (slightly ahead; E5 and Copilot ARPU); paid seats +7% (SMB/frontline-led); M365 commercial products +13%, significantly ahead on Office 2024 transactional purchasing; consumer cloud +8%, subs 86.3M (+10%, mix shift to Basic); LinkedIn +9% (Talent Solutions slightly below on hiring-market weakness); D365 +19%/+18% cc. Segment op income +16%/+15% cc.
• IC $25.5B (+19%), with more unfavorable FX than expected; ex-FX, Azure non-AI, on-prem, and services all slightly below, offset by Azure AI above: Azure +31% (13 pts AI, +157% YoY); on-prem server -3% (slow Windows Server 2025 transactional purchasing); EPS services -1%. Segment GM -4 pts YoY; op income +14%.
• MPC $14.7B (roughly flat), ahead on Windows OEM pre-builds, a third-party Search partnership, and CoD: Windows OEM + devices +4% (commercial inventory builds ahead of Windows 10 EOS plus tariff uncertainty); search ex-TAC +21%/+20% cc (third-party partnership usage; rate + volume); gaming -7%/-8% cc; Xbox content/services +2% (ahead; Blizzard/Activision content). Segment GM +6 pts YoY; op income +32%/+30% cc.
• Cash flow: OCF $22.3B (+18%); capex $22.6B; FCF $6.5B (-29% YoY) — the decline steepened from -7% last quarter.
• Q3 guide (USD): P&BP $29.4–29.7B (+11–12% cc); IC $25.9–26.2B (+19–20% cc); MPC $12.4–12.8B. Azure 31–32% cc (AI contribution grows as capacity lands; non-AI execution drag persists through H2; capacity constrained in Q3, roughly in line by FY25-end). M365 commercial cloud 14–15% cc; commercial products ~flat; LinkedIn low-to-mid single digits; D365 mid-teens. On-prem server down mid-single digits. Windows OEM ~flat (inventory normalizes; tariff caveat), devices declining; search ex-TAC mid-teens (partnership normalizes); gaming low single digits, content/services low-to-mid single digits. COGS $21.65–21.85B (+19–20% cc); opex $16.4–16.5B (+5–6% cc); other income ~-$1B; tax ~18%. Bookings guided roughly flat YoY (flat expiry base, tough large-Azure-contract comp). Cloud GM ~69%. FY25 op margins now guided up slightly YoY; FY26 capex growth lower than FY25 with mix shifting to short-lived assets.

Product and launch scorecard

• Azure AI: 13 points of Azure growth (from ~12), AI services +157% YoY, ahead of expectations; new large OpenAI Azure commitment (first tranche recognized in bookings; ROFR retained); Azure AI Foundry at 200,000+ MAU two months post-launch; DeepSeek R1 added to Foundry/GitHub catalog day-one (with red teaming/safety scanning); Phi SLMs downloaded 20M+ times; 30+ industry models (Bayer, Paige.AI, Rockwell, Siemens). Azure OpenAI apps on Azure databases/app services more than doubled YoY.
• M365 Copilot: strongest quantitative disclosure yet — first-quarter-availability cohort expanded seats >10x over 18 months; majority of existing enterprise customers buying more seats; Novartis at 40,000 seats (adding thousands per quarter); Barclays, Carrier, Pearson, University of Miami each 10,000+ seats this quarter; daily users >2x QoQ again; usage intensity +60% QoQ. Named as an ARPU driver alongside E5 again. Still no total seats, attach rate, or dollar figure.
• Copilot Chat / Studio: Copilot Chat extends agents to every employee; Copilot Studio at 160,000+ orgs (from 100,000+), 400,000+ custom agents created in three months (>2x QoQ).
• GitHub Copilot: 1M+ sign-ups in the first week for Copilot in VS Code (free tier); customers ASOS, Spotify, HP, HSBC, KPMG; GitHub at 150M developers (+50% in two years). No revenue/subscriber metric.
• Fabric: 19,000+ paid customers (from 16,000, +19% QoQ) — "fastest growing analytics product in our history"; Power BI 30M+ MAU (+40% YoY).
• DAX Copilot: 2M+ monthly physician encounters, +54% QoQ (from 1.3M); Mass General Brigham, Michigan Medicine, Vanderbilt.
• LinkedIn: Premium surpassed $2B annual revenue for the first time; subscriber growth ~+50% over two years; ~40% of subscribers using AI features; hiring took share despite weak market.
• Search: ex-TAC +21%/+20% cc, fourth consecutive strong quarter — but this one flattered by a third-party partnership's usage, guided to normalize to mid-teens in Q3; Edge >30% US share on Windows, 15 straight quarters of share gains.
• Gaming: Black Ops 6 top-selling on Xbox and PlayStation, most launch-quarter players of any paid CoD; Indiana Jones 4M+ players; record 140M cloud-gaming hours; Game Pass record quarterly revenue, PC subs +30% — but content/services only +2% and guided low-to-mid single digits; hardware still declining.
• Copilot+ PCs: 15% of US premium laptops this holiday; DeepSeek R1 distilled models coming to local NPUs; Windows 10 EOS driving commercial pre-builds (with tariff-uncertainty pull-forward).
• Security: 80+ new capabilities in a year; Security Copilot resolving incidents 30% faster; Purview auditing 2B+ Copilot interactions (from 1B). No security revenue figure again.

Sell-side read-through

• Nine questions; the capex/ROI debate softened into architecture questions — DeepSeek, Stargate/OpenAI, model mix, Copilot packaging — while Azure's miss mechanics got only one direct airing (Weiss).
• Weiss (Morgan Stanley) got the key mechanical answer: the shortfall is non-AI ACR in the scale motion — summer sales-motion changes misbalanced AI vs migration work; adjustments underway but impact persists through H2. AI results were better than expected (delivery dates pulled in by weeks). Q3's 31–32% cc guide embeds the same capacity constraint; Hood "feels good" on AI, watching non-AI. Nadella's add: "you would rather win the new than just protect the past."
• Moerdler (Bernstein) on the AI beat: Hood confirmed Copilot outperformance — seats (new + expansion), usage, and price per seat "quite good" — the first pricing signal disclosed.
• Thill (Jefferies) on DeepSeek: Nadella's Jevons answer — 10x inference improvement per cycle from software, DeepSeek innovations will be commoditized, falling token prices expand consumption; "all good news" for a hyperscaler and PC platform provider.
• Keirstead (UBS) on Stargate/OpenAI: Nadella — OpenAI "committed in a big way to Azure," bookings are "just the first tranche," ROFR preserves future benefit; Microsoft builds a fungible fleet balancing training/inference, continuously modernized (Moore's Law 2x, optimization 10x — don't overbuy at one point in time). Hood tied capex to ~$300B RPO and the pivot from long-lived infrastructure to CPU/GPU spend "more directly correlated to revenue." No one challenged the Cruise impairment or asked for the OpenAI loss figure this quarter.
• Zelnick (Deutsche Bank) on Copilot packaging: Copilot Chat + Studio as the ubiquitous tier, full Copilot above it; o1-powered Think Deeper free globally — inference cost declines flowing into broader distribution.
• Reback (Stifel) on open vs proprietary models: Nadella — multi-model is the steady state; Foundry as the "new app server" managing model churn; no percentage answer given.
• Sills (BofA) on Copilot use cases: departmental land (sales, finance, supply chain) then collaboration network effects drive enterprise-wide spread; "think with AI, work with people."
• Bracelin (Piper Sandler) on the bookings sea change (+$39B sequential RPO, 75% cc bookings): Hood — OpenAI commitment is a big component but ongoing, not one-time; plus strong core motions (renewals, upsell, E5 momentum) and large Azure commitments from both existing and new customers. Q3 bookings guided flat against this comp.
• Nobody asked about: the Cruise impairment size, the OpenAI equity-method loss run rate (last quarter's ~-$1.5B guide went unaddressed), FCF -29%, gaming's modest content/services growth despite record engagement, or the missing security revenue metric.

Management credibility

• For: The $10B AI run-rate commitment was met and exceeded on schedule ($13B, +175%) — the dated, falsifiable promise from last quarter was kept. Q3 Azure guide of 31–32% cc matches the Q2 print trajectory with the decel mechanism now explicitly owned as internal execution, not external vendors. Opex below expectations again; op margin beat and FY25 margin guidance raised to "up slightly." Capex in line at $22.6B and guided flat — the "up sequentially" pattern broke as signaled. Hood proactively decomposed the bookings spike (OpenAI vs core motions) and flagged its Q3 flat comp. The capacity-constraint end date (roughly in line by end of FY25) is specific and testable.
• Against / soft spots: Azure non-AI missed on Microsoft's own go-to-market execution — the second consecutive quarter with a new explanation for Azure softness (Q1: third-party capacity slippage; Q2: scale-motion misbalancing), and the fix now stretches "through H2," pushing the clean-acceleration story further out. The -$2.3B other income missed the guide on a Cruise impairment disclosed only in passing, with no updated OpenAI loss figure. The $13B AI run rate remains undefined in composition. Copilot metrics improved (cohort expansion, named seat counts) but still no total seats or dollars. Search's beat relied on a third-party partnership that normalizes next quarter. Windows OEM's beat was inventory pre-builds and tariff fear — pull-forward, not demand.
• Under pressure: Weiss got a candid, detailed execution-miss explanation — the most accountable answer of the call. Keirstead's Stargate question got strategic framing (fungible fleet, ROFR) but no direct response on whether Microsoft is stepping back from funding OpenAI training capex. Reback's open-vs-proprietary percentage question went unanswered numerically.

What changed versus the prior quarter

• Azure decelerated as guided (34% cc → 31%) but the miss driver rotated: from third-party capacity slippage (external) to non-AI go-to-market execution in the scale motion (internal), with the drag now expected through H2. AI contribution rose ~12 → 13 points and AI beat expectations.
• The AI run rate printed $13B vs the $10B "next quarter" commitment — milestone delivered early and above expectations; +175% YoY growth disclosed for the first time.
• Bookings exploded: +30%/+23% cc → +67%/+75% cc; RPO $259B → $298B (+$39B sequential, largest ever); beyond-12-months RPO +27% → +45% — driven by the new OpenAI Azure commitment plus broad core-motion strength. Q3 bookings guided flat against this comp.
• Capex trajectory inflected: $20B → $22.6B but guided flat for Q3/Q4 (vs "up sequentially" every prior quarter); FY26 capex growth guided below FY25 with mix shifting to short-lived, revenue-correlated assets. FCF decline worsened: -7% → -29% YoY.
• Margin story improved again: op margin 45% (+2 pts YoY) vs last quarter's guided expansion; FY25 op margins now guided up slightly YoY (from "down ~1 pt" framing two quarters ago).
• Other income worsened beyond the OpenAI guide: -$2.3B vs ~-$1.5B guided, on a Cruise impairment — a new, unflagged charge; OpenAI loss run rate not updated.
• M365 Copilot disclosure stepped up: from adoption anecdotes to cohort economics (>10x seat expansion over 18 months, 10K+ seat deals named, usage intensity +60% QoQ, price per seat "quite good") — still no totals.
• Windows OEM flipped to a beat (+4%) on Windows 10 EOS pre-builds and tariff pull-forward — but guided flat for Q3 as inventory normalizes; last quarter's devices "execution challenges" not revisited.
• Search's streak continued (+21% cc) but with a new caveat — third-party partnership usage inflating the print, normalizing to mid-teens in Q3.
• Gaming pivoted explicitly to profitability: MPC op income +32%, segment GM +6 pts; content/services +2% (vs ~8% organic implied last quarter) with record Game Pass revenue and PC subs +30%.
• FX became a headwind: from neutral in Q2 to a 2-pt Q3 revenue drag (~$1B worse than October assumptions).

Bull case

• The $13B AI run rate, +175% YoY and above expectations, is the third consecutive escalation of the AI monetization proof — and Azure AI specifically beat while capacity-constrained, meaning the number is supply-limited, not demand-limited (management framing; synthesis, not a recommendation).
• Bookings/RPO step-change: +75% cc bookings, $298B RPO with the long end +45%, and Hood's insistence the OpenAI commitment is ongoing rather than one-time — multi-year revenue visibility materially improved even ex-OpenAI (strong E5, renewals, new and expanding Azure commitments).
• The capex peak is now visible: flat Q3/Q4 spend, FY26 growth below FY25, mix shifting to revenue-correlated short-lived assets, and capacity roughly matching demand by FY25-end — the convergence Hood promised "like the last cloud cycle" now has dates and mechanics.
• Margin delivery beat the narrative: +2 pts op margin, opex +5%, FY25 margins guided up slightly — efficiency is funding the AI build, and Gaming/Search margin execution (MPC op income +32%) shows the playbook working.
• Copilot evidence is compounding toward monetization: >10x cohort seat expansion, 10K+ seat enterprise deals, usage intensity +60% QoQ, and the first positive price-per-seat signal — with Copilot Chat as a top-of-funnel for the installed base.
• Fabric (+19% QoQ customers), DAX (+54% QoQ encounters), LinkedIn Premium ($2B), and Foundry (200K MAU in two months) show the AI stack monetizing beyond Azure infrastructure.

Bear case

• Azure missed on self-inflicted execution, and the fix takes two more quarters: non-AI scale-motion problems persist "through H2," meaning the H2 acceleration thesis — already deferred once — now rests on a sales-motion repair, not just capacity landing. Q3's 31–32% cc guide embeds no re-acceleration.
• FCF fell 29% YoY and other income is leaking from multiple directions: -$2.3B this quarter (Cruise impairment on top of OpenAI losses), with the OpenAI run rate no longer quantified and Q3 other income guided to another ~-$1B.
• The bookings blowout is partly a related-party event: the largest sequential RPO increase ever is substantially an OpenAI commitment — Microsoft's backlog growth is increasingly tied to a loss-making partner whose training capex Microsoft just declined to underwrite (inference from Stargate framing).
• Quality of beats elsewhere is questionable: Windows OEM beat on tariff-fear pre-builds (guided flat next quarter); search beat on a third-party partnership (normalizing next quarter); M365 commercial products +13% on transactional Office 2024 purchasing (guided flat). Several Q2 strengths are pull-forwards or one-offs (inference from guidance).
• Copilot remains undollarized despite the best adoption disclosure yet — >10x expansion off an undisclosed base is unverifiable, and the ARPU contribution remains inseparable from E5.
• Gaming's record engagement produced +2% content/services growth — the ratable-recognition and hardware headwinds keep muting the segment, and Q3 guidance (low single digits) implies no near-term change.
• FX turned into a 2-pt headwind (~$1B) with Q4 also negative — a new external drag absent from the prior quarter.

Next-quarter watchlist

1. Azure vs the 31–32% cc Q3 guide — AI point contribution vs 13; whether non-AI scale-motion fixes show progress; any update on the "roughly in line with demand by end of FY25" capacity commitment — now the central, dated credibility test.

2. AI run-rate disclosure — does the $13B get updated, and does management finally define its composition (Azure AI vs Copilots vs other)?

3. Non-AI Azure execution — specific evidence the summer sales-motion rebalancing is working; Hood flagged H2-long impact, so Q3 commentary on trajectory matters more than the print.

4. Bookings vs the flat YoY guide — how much is OpenAI tranche two vs core motions; RPO beyond-12-months trend after +45%.

5. Capex vs "flat at $22.6B" — first test of the plateau; FY26 mix-shift detail (short-lived vs long-lived) and any convergence timing update.

6. FCF trajectory — whether -29% YoY was the trough; OCF coverage as capex plateaus.

7. Other income vs ~-$1B guide — the OpenAI equity-method loss run rate (unquantified this quarter) and any further impairments.

8. M365 Copilot quantification — cohort expansion and price-per-seat language set up ARPU disclosure; watch M365 commercial cloud vs 14–15% cc guide and seat-growth moderation.

9. Cloud GM vs ~69% guide — AI COGS scaling pace against the margin-up FY25 commitment.

10. Windows OEM normalization — whether the tariff/EOS pre-build reverses as guided (flat YoY); devices decline depth; tariff commentary.

11. Search ex-TAC vs mid-teens guide — the clean read on rate expansion once the third-party partnership normalizes.

12. Gaming — content/services vs low-to-mid single-digit guide; whether record Game Pass revenue/PC subs ever convert to disclosed subscriber numbers; hardware decline.

13. Copilot+ PC and Windows 10 EOS evidence — premium-laptop share progression toward the "majority of PCs" claim.

14. Missing metrics — Azure AI customer count, GitHub Copilot subscribers/revenue, security revenue, Game Pass total subs: whether any return, or the rotation is now permanent.

Quarter in one view

• Revenue $65.6B (+16%); EPS $3.30 (+10%); operating income +14%; operating margin 47%, down 1 pt YoY (up 1 pt ex-Activision). Microsoft Cloud $38.9B (+22%), roughly in line; cloud GM 71%, down 2 pts YoY (slightly better than expected on Azure improvement, still AI-scaling-driven).
• Azure +33%/+34% cc, above the 28–29% cc guide, with ~12 points from AI (up from 8; Hood said "similar to last quarter" — the prior-quarter 8-pt figure was pre-recast, so comparability is approximate). Beat driven by in-period revenue recognition; Hood framed "pure consumption" as ~33%. Demand still exceeds capacity; non-AI contribution fell ~1 pt sequentially but was "in line across regions" — last quarter's European softness did not recur as a disclosed issue.
• Headline disclosure: AI business on track to surpass a $10B annual revenue run rate next quarter — "fastest business in our history to reach this milestone," all inference revenue per Nadella.
• Capex $20B incl. leases ($14.9B cash PP&E), in line; guided up sequentially again. OCF $34.2B (+12%); FCF $19.3B (-7%) on capex.
• Bookings +30%/+23% cc, ahead; RPO $259B (+22%/+21% cc) (recast basis; prior $269B not comparable); ~40% within 12 months (+17%); beyond-12-months +27%; annuity mix 98%.
• Q2 Azure guided 31–32% cc — a deceleration Hood attributed to (a) Q1's recognition benefit and (b) third-party leased capacity pushed out of Q2 into H2/Q3; H2 acceleration reiterated. Q2 other income guided to ~-$1.5B, primarily the equity-method share of OpenAI losses — the first explicit quantification.

What management is focused on

• The $10B AI run-rate milestone as the ROI answer: both Nadella and Hood used it to reframe the capex debate — "it's all inference," "we're not selling raw GPUs for other people to train... we turn that business away," and Hood's "the revenue is what builds the next cycle of training." This is the coordinated response to three quarters of capex-payback questions (inference from repetition).
• Training spend is now explicitly rate-limited by inference monetization: Nadella's answer to Weiss — training capital "will be rate limited by your monetization of inference," fleet refreshed annually on Moore's Law, depreciated over lifecycle. A new doctrinal guardrail statement.
• Capacity constraint narrative shifted to third-party delivery: the Q2 decel is blamed on leased third-party capacity (end-to-end, "inclusive of kits") delivering late, pushed "mainly into H2 and in general Q3" — not GPUs, power, or cooling per Hood's direct answer to Murphy.
• Copilot → agents → autonomous agents as the product architecture: Copilot as UI for AI, Copilot Studio to build agents, autonomous agents raising exceptions to Copilot; 10 out-of-the-box Dynamics agents launched; the full stack also exposed as Azure building blocks.
• OpenAI relationship defense: Nadella called it "super beneficial" for both sides, emphasized the investment stake value and product portfolio built on it; Hood disclosed the equity-method losses are capped at the $13B total investment (per the 10-Q).
• Security as operational priority: Secure Future Initiative at 34,000 engineer-equivalents; Defender securing 750,000+ gen-AI app instances; Purview auditing 1B+ Copilot interactions. No security revenue figure this quarter.

Key numbers and quarter mechanics

• Total: revenue $65.6B (+16%); GM 69%, down 2 pts YoY (cloud GM + Activision purchase accounting); opex +12% (9 pts Activision), lower than expected on cost efficiency; headcount +8% YoY, +2% ex-Activision; tax ~19%; other income -$283M, much better than the ~-$650M guide on FX remeasurement and investment gains (equity-method losses "as expected"); $9B returned.
• Activision net impact: ~3 pts revenue growth, 2-pt drag on op income growth, -$0.05 EPS (from -$0.06), $911M purchase accounting/integration/transaction costs.
• P&BP $28.3B (+12%/+13% cc), ahead across all businesses (recast segment): M365 commercial cloud +15%/+16% cc (small in-period recognition benefit; ARPU from E5 and M365 Copilot); paid seats +8%, driven by SMB/frontline; M365 commercial products +2%/+3% cc; M365 consumer cloud +6%/+7% cc, subs 84.4M (+10%); LinkedIn +10%/+9% cc; Dynamics +14%, D365 +18%/+19% cc (~90% of Dynamics). Segment op income +16%.
• IC $24.1B (+20%/+21% cc), in line: Azure +33%/+34% cc (~12 pts AI; non-AI contribution down ~1 pt sequentially); on-prem server -1% (soft transactional purchasing ahead of Windows Server 2025, offset by recognition benefit); services -1%/flat cc. Segment GM -3 pts YoY on AI scaling; op income +18%.
• MPC $13.2B (+17%; 15 pts Activision), above on gaming and search: Windows OEM + devices +2% (OEM better on market mix; devices lower on "execution challenges in the commercial segment"); search ex-TAC +18%/+19% cc, ahead, with rate expansion plus volume; gaming +43%/+44% cc (43 pts Activision); Xbox content/services +61% (53 pts Activision — implying ~8% organic, better than last quarter's ~3%). Segment op income -4%.
• Q2 guide (USD): P&BP $28.7–29.0B (+10–11% cc); IC $25.55–25.85B (+18–20% cc); MPC $13.85–14.25B. Azure 31–32% cc with stable consumption QoQ, AI contribution "similar," record sequential Azure dollar adds; H2 acceleration reiterated. M365 commercial cloud ~14% cc with moderating seat growth; H2 P&BP growth stable vs Q2; Copilot revenue to "grow gradually." Cloud GM ~70%, down YoY. Capex up sequentially. COGS $21.9–22.1B (+11–13% cc); opex $16.4–16.5B (+~7% cc); op margin expansion guided; other income ~-$1.5B (OpenAI equity-method loss); tax ~19%. Gaming down high single digits on hardware; Xbox content/services roughly flat (Black Ops 6 revenue recognized ratably — Game Pass and always-online both defer recognition); Windows OEM/devices down low-to-mid single digits; search ex-TAC high teens.

Product and launch scorecard

• Azure AI: ~12 points of Azure growth (from 8, recast caveat); Azure OpenAI usage more than doubled in six months; GE Aerospace assistant to 52,000 employees (500K queries, 200K documents in three months); o1 model support added; first cloud with NVIDIA Blackwell GB200; Maia 100 and Cobalt 100 (50% better price-performance) in production. Azure AI customer count not disclosed this quarter (was 60,000+).
• M365 Copilot: daily users more than doubled QoQ (third straight quarter of ~2x); ~70% of Fortune 500 now use it; Vodafone 68,000-seat rollout (3 hrs/person/week saved in trial); UBS 50,000 seats — largest financial-services deal; "fastest-adopted new M365 suite," faster than E3/E5 transitions; responses 2x faster, quality ~3x better; Pages launched. Hood: Copilot ARPU impact is masked by lower-ARPU SMB/frontline seat mix — the most direct explanation yet for why Copilot isn't visible in the numbers. Still no seats, attach rate, or dollars.
• GitHub Copilot: enterprise customers +55% QoQ; Copilot Workspace (agentic spec-to-code), Auto Fix (3x faster vulnerability fixes), GitHub Spark, multi-model support. No subscriber, org-count, or revenue metric this quarter — the $2B GitHub run-rate framing from last quarter was not updated.
• Copilot Studio / agents: 100,000+ organizations, up >2x QoQ (from 50,000); 10 autonomous agents added to Dynamics 365; D365 Copilot MAU +60% QoQ.
• Power Platform: ~600,000 orgs using AI capabilities, +4x YoY (from 480,000 last quarter).
• Fabric: 16,000+ paid customers (from 14,000, +14% QoQ); >70% of Fortune 500.
• DAX Copilot: 1.3M+ physician encounters/month at 500+ orgs; "faster revenue growth than GitHub Copilot in its first year" — a rare comparative monetization claim.
• Gaming: record MAU; Game Pass Q1 records for revenue and ARPU; Black Ops 6 = biggest CoD launch ever, record day-one players and record Game Pass subscriber adds; PlayStation/Steam unit sales +60% YoY. Subscriber count still undisclosed; Q2 content/services guided flat due to ratable recognition.
• Search: ex-TAC +18%/+19% cc, third consecutive strong quarter, now with rate expansion alongside volume; Nadella framed it as a $10B+ business growing faster than the market.
• Copilot+ PCs: all three silicon vendors (AMD, Intel, Qualcomm) now shipping; Click to Do launched; Windows 10 end-of-support (one year out) positioned as the Windows 11 upgrade driver. Devices still declining on commercial execution issues.

Sell-side read-through

• Seven questions; tone congratulatory, but the constraint/ROI theme persisted — Weiss (training capital guardrails, power), Moerdler (capex growth trajectory), Murphy (supply-limitation specifics), Rangan (inference payback) were all variations on the same debate.
• Weiss (Morgan Stanley): got the new doctrine — training capex rate-limited by inference monetization; external constraints are DCs and power, near-term tightness partly third-party leases; Nadella "feels pretty good" about H2 supply/demand matching.
• Thill (Jefferies) pinned the Q2 Azure decel: Hood decomposed it cleanly — Q1's 34% cc was ~33% "pure consumption" plus recognition benefit; Q2's 1–2 pts of decel is "majority" supply pushouts; underlying consumption stable. The quarter's most mechanically important exchange.
• Moerdler (Bernstein) on capex trajectory: Hood — capex growth will slow and revenue growth will rise until they converge, "like the last [cloud] cycle," pace dependent on adoption; some training spend lands in opex, not COGS. No timing commitment on convergence.
• Keirstead (UBS) asked the OpenAI question directly: Nadella gave relationship framing, no diversification signal addressed head-on; Hood disclosed the $13B investment cap on equity-method losses and called the P&L impact "mechanical." The ~-$1.5B Q2 other-income guide is the first dollar quantification of the OpenAI loss drag.
• Rangan (Goldman) on inference payback: Nadella — "we're not waiting for inference to show up," the $10B is all inference, raw-GPU resale is "a business we turn away," adverse-selection dig at VC-funded GPU buyers; Hood confirmed the invest-now-grow-later reading.
• Murphy (JPMorgan) pressed on supply components: Hood refused component-level attribution — it's third-party end-to-end delivery pushed to H2/Q3; declined to size the Q2 point impact beyond the two-factor decomposition.
• Jaluria (RBC) asked why Copilot isn't visible in results: Hood's answer — it already is, in ARPU, masked by SMB/frontline seat mix; also expect Copilot engagement to show in search ex-TAC. The closest management has come to explaining the Copilot disclosure gap.
• Nobody asked about: devices' commercial "execution challenges," gaming's flat Q2 content/services guide, the recast segment comparability, or security revenue (absent this quarter after returning last quarter).

Management credibility

• For: Azure beat the guide (34% cc vs 28–29% cc) — and Hood proactively decomposed the beat into consumption vs recognition rather than letting the print stand alone. The H2 acceleration commitment was reiterated with a specific mechanism (third-party capacity landing in H2/Q3). Capex "in line" at $20B; opex below expectations; cloud GM slightly better than guided. The $10B AI run-rate claim is dated (next quarter) and falsifiable. OpenAI losses finally quantified and capped ($13B) after three quarters of unexamined equity-method losses. Hood's answer to Jaluria (Copilot ARPU masked by seat mix) is a specific, checkable explanation rather than evasion.
• Against / soft spots: The Q2 Azure guide embeds a deceleration blamed on third-party delivery slippage — a new dependency outside Microsoft's direct control, disclosed only in the Q&A. The "12 points of AI, similar to last quarter" language glosses a recast that makes the 8→12 jump non-comparable. Metric rotation continues: Azure AI customer count dropped, GitHub Copilot metrics absent entirely, security revenue absent after one quarter back, Game Pass subs still hidden, M365 Copilot still undollarized despite being a named ARPU contributor for three quarters. The $10B "AI business" run rate is undefined in scope (which products count) — impressive-sounding but unverifiable as disclosed (inference).
• Under pressure: Thill and Murphy got specific, mechanical answers on Azure. Keirstead's OpenAI question got warmth and an accounting cap, but no response to the "diversifying at the model layer" signal he flagged. Moerdler's capex-convergence timing question got an analogy, not a date.

What changed versus the prior quarter

(Using the supplied prior-quarter summary as baseline; note FY25 segment recast limits some comparability.)

• Azure re-accelerated on the print: 30% → 34% cc, beating a conservative guide; AI contribution 8 → ~12 points (recast caveat); European non-AI softness disappeared from the narrative — non-AI was "in line across regions," though its point contribution fell ~1 pt sequentially.
• The capacity story changed shape: from "demand exceeds capacity, H2 relief" to a specific Q2 miss mechanism — third-party leased capacity pushed into H2/Q3, causing guided decel to 31–32% cc before the reiterated H2 acceleration.
• AI got a dollar figure for the first time: $10B annual run rate "next quarter," framed as all-inference — a direct response to the "no dollar-denominated AI return" criticism, though scope is undefined.
• OpenAI moved from unexamined drag to disclosed drag: ~-$1.5B Q2 other-income guide, losses capped at the $13B investment; prior quarters' "equity-method losses as expected" now have a number.
• Capex $19B → $20B, guided up again; FCF turned negative YoY (-7%) on the build.
• Bookings re-accelerated: +17%/+19% cc → +30%/+23% cc; RPO $259B (recast) with beyond-12-months +27% (from +21%) — long-duration commitments strengthening.
• Organic gaming improved: content/services organic ~3% → ~8% implied (61% less 53 pts); but Q2 guided flat on ratable Black Ops 6 recognition — a new revenue-timing headwind.
• O365/M365 commercial cloud re-accelerated modestly (+13%/+14% cc → +15%/+16% cc) with recognition help; seat growth +7% → +8%; Copilot ARPU now explicitly described as masked by mix.
• Operating margin trajectory improved: FY25 "down ~1 pt" framing gave way to Q2 guided margin expansion; Q1 op margin 47%, down 1 pt (up 1 ex-Activision).
• Headcount +3% → +8% (+2% ex-Activision) — organic hiring still modest.
• Devices developed a new problem: "execution challenges in the commercial segment," with Q2 guided down low-to-mid single digits for OEM+devices combined.

Bull case

• The $10B AI run rate converts the ROI debate into a number: if printed next quarter as committed, AI revenue goes from Azure-points abstraction to a disclosed scale figure growing faster than any prior Microsoft business (management framing; synthesis, not a recommendation).
• Azure beat with consumption intact: ~33% pure consumption, stable Q1→Q2, record sequential dollar adds guided, and the Q2 decel is supply timing, not demand — with H2 acceleration reiterated and now mechanically explained (third-party capacity landing).
• Bookings and RPO strengthened at the long end: +30% bookings, beyond-12-months RPO +27%, $100M+ Azure contract growth — multi-year demand visibility improved while capex critics focused on the P&L.
• Copilot adoption evidence keeps compounding: 70% of Fortune 500, UBS 50K seats, Vodafone 68K with quantified time savings, fastest suite adoption in M365 history, and a plausible disclosed reason (mix masking) for why revenue lags seats.
• Margin story is better than guided: opex below expectations, Q2 op margin expansion guided, cloud GM beating on Azure improvement even while scaling AI — the "down 1 pt" FY25 frame now looks conservative (inference from guidance shift).
• Search is a third consecutive beat with rate expansion — monetization, not just engagement.

Bear case

• Azure decelerates into Q2 on a supply excuse that is now external: third-party lessors slipped, pushing capacity to H2/Q3 — the H2 acceleration thesis now depends partly on vendors Microsoft doesn't control, and the guide embeds the miss.
• The OpenAI drag is now real and sized: ~-$1.5B in a single quarter of other income, recurring mechanically until the $13B cap — a structural EPS headwind that arrived with one quarter's notice.
• FCF declined 7% YoY with capex still rising sequentially — the cash-flow cost of the build is now visible, and capex/revenue convergence has no date.
• Copilot remains undollarized after three quarters as a named contributor: the mix-masking explanation is plausible but also convenient — it cannot be verified without an ARPU or seat-dollar disclosure management declines to give (inference).
• Gaming momentum is accounting-flattered and about to flatten: Q2 content/services guided flat despite the biggest CoD launch ever, because Game Pass and always-online defer recognition — the subscription shift trades near-term revenue for annuity, and hardware still drags the segment to a high-single-digit decline.
• Devices has an execution problem, not just a market problem — commercial-segment missteps disclosed for the first time, with Copilot+ PCs not yet offsetting.
• Disclosure rotation continues: Azure AI customers, GitHub Copilot metrics, and security revenue all disappeared this quarter; the $10B AI figure has no defined composition (inference from disclosure history).

Next-quarter watchlist

1. The $10B AI run-rate print — does management confirm it in Q2 as committed, and do they define its composition (Azure AI vs Copilots vs other)?

2. Azure vs the 31–32% cc Q2 guide — AI point contribution vs "similar to last quarter"; whether consumption stays stable; any further third-party capacity slippage; evidence the H2 acceleration is tracking (the central credibility test, now two quarters running).

3. Other income vs the ~-$1.5B guide — the OpenAI equity-method loss run rate and any commentary on trajectory toward the $13B cap.

4. Capex vs "increase sequentially" off $20B — and any update on when capex growth converges toward cloud revenue growth (Moerdler's unanswered timing question).

5. M365 Copilot quantification — Hood has now pre-positioned ARPU as the disclosure venue; watch M365 commercial cloud vs ~14% cc guide, seat-growth moderation, and whether any Copilot dollar/seat figure finally appears.

6. Cloud gross margin vs ~70% guide — pace of AI COGS scaling against Azure improvement.

7. Gaming recognition mechanics — content/services vs the flat guide; whether Black Ops 6 Game Pass adds convert to disclosed subscriber or ARPU figures; hardware decline depth.

8. Devices commercial execution — whether the disclosed challenges are fixed; Copilot+ PC and Windows 10 end-of-support upgrade evidence in Windows OEM.

9. Search ex-TAC vs high-teens guide — a fourth consecutive strong quarter would establish the rate-expansion story.

10. Bookings on the growing expiry base — Hood's volatility caveat; RPO growth and the beyond-12-months +27% trend.

11. FCF trajectory — whether -7% YoY worsens as capex rises; OCF coverage of the build.

12. Missing metrics — whether Azure AI customer count, GitHub Copilot subscribers/orgs, and security revenue return, or whether their absence is now permanent.

13. Headcount and opex — +2% organic headcount vs the ~7% cc Q2 opex guide; whether efficiency discipline holds as AI hiring continues.