$37B AI run rate. 40% Azure. 20M Copilot seats. And a $190B capex bombshell that overshadowed one of the strongest enterprise AI prints on record.
Microsoft's Q3 FY2026 was a legitimately strong quarter by every conventional measure: revenue beat, EPS beat, Azure beat guidance, Copilot delivered proof of engagement, and commercial RPO nearly doubled to $627 billion. By any pre-earnings checklist, this was a bull thesis confirmed.
But the stock slipped 1–2% immediately after the print. The reason: $190 billion in capex guidance for calendar year 2026 — $35 billion above the $155 billion consensus — forced the market to recalculate the FCF and returns timeline. When free cash flow is down 22% year-over-year despite operating cash flow growing 26%, the market doesn't cheer the earnings beat. It prices the capex cycle.
Every data point in this recap should be read through that tension. The bull case is unambiguously stronger after this print than before it. The bear case is also stronger, because the capex required to sustain this growth is larger than anyone modeled.
Microsoft disclosed a $37 billion annualized AI revenue run rate, up 123% year-over-year. This figure aggregates Azure AI services, OpenAI revenue share, Copilot licensing, and enterprise AI SaaS. It is roughly the annual revenue of a Fortune 300 company — created in under three years. Enterprise AI is not a promise; it is a present-day revenue stream.
| Azure Growth Metric | Q2 FY2026 | Q3 FY2026 | Q4 Guide | Consensus |
|---|---|---|---|---|
| Azure YoY Growth | 39% | 40% | 39–40% | 36.7% |
| Microsoft Cloud Revenue | ~$48B | $54.5B | — | — |
| Commercial RPO | — | $627B | — | — |
| RPO YoY Growth | — | +99% | — | — |
The sequential acceleration from 39% to 40% Azure growth is significant because the consensus expected capacity constraints to limit growth, not expand it. CFO Amy Hood's guidance — "Azure growth to show modest acceleration in the second half of the calendar year" — signals the capacity pipeline is converting to revenue, not just burning capex.
The $627 billion commercial RPO (+99% YoY) is the most underappreciated metric in the print. This is contracted future revenue — customers have already committed. The near-doubling of RPO means enterprise AI infrastructure commitments are locking in for 3–5 year horizons. This is not quarterly TAM speculation; it is legally contracted backlog that de-risks the $190B capex thesis.
Microsoft reported 20 million paid M365 Copilot seats as of March 31, 2026 — up from 15 million in January, a 33% increase in a single quarter. The year-over-year growth rate is 250%. At $30/seat/month list price (with enterprise discounting), this implies a Copilot revenue run rate approaching $7.2 billion annually.
The metric that matters most isn't the seat count — it's the engagement data. CEO Satya Nadella stated: "Weekly engagement is now at the same level as Outlook," — Microsoft's most-used application, embedded in the daily workflow of 450 million+ users. Seat count is a licensing metric. Outlook-level engagement means Copilot has achieved habit formation.
| Enterprise Deployment | Seats / Scale | Signal |
|---|---|---|
| Accenture | 740,000 seats | Largest disclosed single-customer deployment |
| Bayer, J&J, Mercedes, Roche | 90,000+ seats each | Cross-sector regulated enterprise validation |
| 50,000+ seat customers | 4x increase QoQ | Large enterprise deployment acceleration |
| GitHub Copilot orgs | 140,000 | Enterprise subscribers up 3x YoY |
| M365 penetration | ~4.4% | 20M of 450M+ M365 subscribers — significant headroom |
The capex story dominated the post-print reaction and deserves precise analysis. Microsoft guided calendar year 2026 capital expenditures at approximately $190 billion — versus $118 billion in 2025 (+61%) and versus consensus expectations of $155 billion (+22%). The Q4 FY2026 quarterly run rate guidance of greater than $40 billion would be the highest single-quarter capex in company history.
| Capex Item | Amount | Context |
|---|---|---|
| CY2026 Total Guidance | ~$190B | vs. $155B consensus — $35B miss to upside |
| CY2025 Actual | ~$118B | +61% YoY growth to 2026 guidance |
| Component cost inflation | ~$25B | HBM, GPU, CPU cost increase baked into $190B |
| Short-lived assets (GPUs/CPUs) | ~2/3 of total | ~$127B in compute hardware |
| Long-lived assets (buildings) | ~1/3 of total | ~$63B in data center infrastructure |
| Q3 FY2026 FCF | $15.8B | Down 22% YoY despite +26% operating cash flow |
| Q3 FY2026 Gross Margin | 67.6% | Narrowest since 2022 — depreciation + AI infra drag |
The math: $190B capex spending produces approximately $127B in GPU/CPU hardware that depreciates over 3–5 years. At $40B+ quarterly capex rates, the annual depreciation charge will grow materially from current levels, creating sustained gross margin pressure through 2026 and into 2027. The FCF recovery is therefore a 2027–2028 story, not a near-term story.
The counter-argument — and management's core thesis — is the $627B commercial RPO backlog. If that backlog converts at expected rates, the capex is supply procurement against contracted demand, not speculative infrastructure build. The risk is the conversion rate, not the demand signal.
Days before earnings, Microsoft announced a restructured partnership with OpenAI. The key terms: Microsoft's license to OpenAI technology extended to 2032 on a non-exclusive basis — OpenAI can now partner with Amazon, Google, and others. In exchange, Microsoft no longer pays revenue share to OpenAI (prior model had bidirectional payments). OpenAI continues paying Microsoft 20% of its revenue through 2030, and commits to Azure as its primary cloud provider.
| Change | Old Structure | New Structure | Impact |
|---|---|---|---|
| Microsoft → OpenAI payments | Revenue share paid | Eliminated | Margin accretive for MSFT |
| OpenAI → Microsoft payments | 20% of OpenAI revenue | 20% through 2030 (capped) | Continued revenue stream |
| OpenAI model license | Exclusive | Non-exclusive (2032) | OpenAI now competes via AWS/GCP |
| OpenAI compute anchor | Azure primary | Azure primary (confirmed) | RPO credibility maintained |
The Stargate Initiative — the $500 billion OpenAI/SoftBank/Oracle joint venture to build 10 GW of AI data centers by 2029 — is adjacent but structurally separate from Microsoft's own capex. Oracle will provide ~4.5 GW of compute. Microsoft's role is leasing capacity at adjacent Stargate sites, creating a hybrid infrastructure model that captures overflow demand without being primary funder.
Microsoft's Q3 FY2026 results are a validation-and-complication for the enterprise AI thesis. The core demand signals are stronger than expected. The cost structure and timeline to returns are more challenging than modeled.
| Thesis Item | Status | Print Evidence |
|---|---|---|
| Training cost inflation persists | ✅ Validated | $25B of $190B capex is component cost inflation (HBM, GPU) |
| Inference scaling demand | ✅ Validated | Azure AI +123% YoY; capacity-constrained, not demand-constrained |
| Enterprise pricing power | ✅ Validated | $30/seat Copilot, no discounting reported; $627B RPO locked |
| Copilot habit formation | ✅ Validated | Weekly engagement at Outlook-level; Accenture 740K seats |
| Margin compression timeline | ⚠️ Pulled Forward | 67.6% gross margin now, not in 18 months; FCF -22% YoY |
| Copilot gross margin >60% | 🔍 Unproven | Run rate implied ~$7.2B; margin structure not disclosed |
| Long-term capex ROI | 🔍 Unproven | $190B capex must convert $627B RPO; timeline 3–4 years |
Sell-side sentiment is constructive but qualified. Buy/Overweight ratings dominate across coverage; zero Sell ratings tracked. Price targets cluster in the $450–$600 range with a $565 consensus average, implying ~36% upside from the $414 area post-print. The caveat: target convergence suggests the market agrees on the direction but disagrees on the timeline to returns.
| Scenario | Price Target | Coverage % | Assumption |
|---|---|---|---|
| Bull Case | $580–$600 | ~40% | Azure stays >40%, Copilot ARPU expands via agents, $627B RPO converts |
| Base Case | $550 | ~45% | Azure moderates to 35–37%, FY2028+ margin recovery, Copilot 30–40M seats |
| Bear Case | $450–$480 | ~15% | AI capex doesn't produce proportional revenue; margin recovery stalls; FCF constrained through 2027 |
| Consensus Average | ~$565 | — | ~36% upside from ~$414 post-print level |
The post-print investment thesis in one sentence: "Microsoft is executing on enterprise AI, but at the cost of near-term returns." The $37B AI run rate, 40% Azure growth, and $627B RPO are proof that the strategy is working. The $190B capex, 67.6% gross margin, and FCF decline are proof that it's expensive.
| Time Horizon | Thesis | Confidence |
|---|---|---|
| Next 6 Months Q2–Q3 CY2026 |
Capex remains elevated; FCF stays under pressure; Azure growth modestly accelerates as new capacity comes online; Copilot seat growth continues but ARPU expansion unproven | HIGH |
| FY2027 H2 2026 → H1 2027 |
Margin recovery begins as depreciation normalizes; Copilot seats reach 30–40M; AI revenue run rate exceeds $50B; agent-based consumption pricing (Copilot Agent Suite) starts contributing | MEDIUM-HIGH |
| FY2028+ 2027 onward |
ROIC on capex cycle validates; FCF recovery to pre-AI levels; potential for stock re-rating if margins recover to 72%+. Bull case: MSFT at $600+. Bear case: capex cycle extends, $450–480 range. | MEDIUM |
Microsoft FY26 Q3 Earnings Press Release (April 29, 2026) · Microsoft FY26 Q3 Earnings Call Transcript (April 29, 2026) · Microsoft Investor Relations (microsoft.com/en-us/investor/earnings/fy-2026-q3) · CNBC: "Microsoft (MSFT) Q3 earnings report 2026" (April 29, 2026) · TechCrunch: "Microsoft says it has over 20M paid Copilot users, and they really are using it" · Wedbush, RBC Capital, Bernstein, JPMorgan analyst notes (April 30 – May 3, 2026) · Visible Alpha consensus data.