MSFT (Trans): Azure guidance at 50%; demand still outstrips supply

Below is Dolphin Research's compiled$Microsoft(MSFT.US) FY27Q1 earnings call Trans

I. Core Takeaways

1. Shareholder returns and cash flow: Returned $10.2bn via dividends and buybacks this quarter. Operating cash flow was $58.0bn (+29% YoY), driven by stronger cloud billings and collections. FCF came in at $18.6bn, weighed by higher capex. Mgmt expects FY27 FCF to remain positive and will tap additional financing when attractive to preserve flexibility.

2. FY27Q2 guidance

a. Consolidated: Revenue of $96.0–97.1bn (+18%–19%). COGS of $32.7–32.9bn (+26%–27%). OpEx of $18.5–18.6bn (+9%), incorporating the recently announced voluntary separation plan; operating margin to edge down YoY.

b. Other items: Other income/expense ex-OpenAI investments roughly -$0.3bn, with interest income below interest expense including DC financing leases. Effective tax rate ~20%; minimal FX impact.

c. Agents & Infra revenue of $80.1–80.7bn (+24%–25%): Azure CC +49%–50%. Efficiency gains on CPU/GPU partly offset the shift of certain 3P GPU-as-a-Service contracts into 2H; demand still exceeds supply.

d. M365 Commercial cloud CC ~+19% (FX drag <1ppt), with growth accelerating each quarter through the year. Microsoft Cloud GPM ~65%, down YoY.

e. Other sub-segments: M365 Consumer cloud to grow low double digits. Productivity and server licensing down mid-to-high single digits. Industry Solutions cloud up mid-to-high single digits. Frontier and Support Services up low-to-mid single digits.

f. Devices & Consumer revenue of $15.9–16.4bn: Search ads ex-TAC up low double digits. Xbox content and services up mid-to-high single digits, strongest in Q2, slowing in Q3, and further in Q4. Hardware down YoY.

g. Windows OEM and Devices down mid-30% as channel inventory runs high and PC demand remains soft, on top of a tough comp from Win10 end-of-support. This trend likely persists through the fiscal year.

3. Capex

a. $45.2bn in the quarter, below plan due to some delayed 3P capacity deliveries. Roughly three-quarters went to short-cycle assets like CPUs/GPUs (including higher memory and storage prices); $7.4bn via finance leases and $39.4bn cash PP&E.

b. Q2 capex expected to exceed $50bn, with calendar-year spend approx. $170bn due to 3P delays. 2H to accelerate, led by CPUs, GPUs and deferred 3P capacity. Mgmt said revenue and returns supported per $1 of capex improve as efficiency rises.

4. This quarter's metric definitions and drivers

a. Bookings: Commercial bookings -12% YoY (CC -11%), lapping a large Azure commitment last year; this quarter also included one large Azure commitment.

b. RPO: Commercial RPO +73% YoY to ~$677.0bn, with a weighted avg. duration of 2.3 years. About 30% will be recognized over the next 12 months (+29%), and the remainder beyond 12 months +98%.

c. Margin: Total operating margin ticked up YoY to 50%. Microsoft Cloud GPM fell to 66%, as mix shift to Azure outweighed product-level margin gains.

d. Azure +48% beat, driven by Foundry and better CPU monetization. Engineering efficiency, 3P deliveries, self-build pace, and model mix impacting ARPU will cause quarterly volatility.

e. M365 Commercial cloud +18% ex the 1ppt in-period recognition last year: GitHub Copilot beat expectations, while E3 and standalone SKU execution was softer. Paid seats +5%, led by SMBs and frontline workers.

II. Details from the call

2.1 Key management commentary

1. Copilot and biz model

a. Rolled out the biggest Copilot update to date, unifying Chat, Cowork, Autopilot and Claude into a single experience for commercial and consumer users. It is integrated with the full Office suite and positioned as the 'new operating system for work'.

b. Usage: Automated tasks nearly doubled since Jul; DAUs rose nearly 50% QoQ and over 5x YoY; satisfaction up nearly 20% QoQ. 90% of queries use Auto mode, routing across models by quality, speed and cost.

c. Large deployments: Customers with 50k+ seats grew over 8x YoY. Volkswagen plans to deploy to 330k+ employees, one of the largest in Europe. E7, which integrates Copilot, E5, Entra and Agent 365, already has thousands of customers.

d. The model extends from per-seat subscriptions to usage-based pricing, with subscriptions funded by IT and usage funded by biz units based on Agent-delivered value. Copilot customers consuming credits via usage-based pricing grew over 75% QoQ.

2. GitHub and developers

a. Agent-generated PRs rose 6x in six months, now over one-third of all PRs on the platform. The company is expanding capacity and refactoring GitHub's core for higher reliability.

b. Copilot Fusion is fully live, with multi-model orchestration delivering up to 40% faster performance at up to 77% lower cost. After usage-based pricing kicked in, Agent credit consumption more than doubled. GitHub Advanced Security revenue grew over 50% YoY.

3. Industry use cases

a. Dynamics 365 added 30 Copilot skills, bringing CRM/ERP capabilities directly into workflows.

b. LinkedIn: Over 10mn members use AI job search daily, and 25k customers use Agent recruiting solutions. The recruiting assistant helps find qualified candidates 33% faster with 83% fewer resume views.

c. Healthcare: Automated 300mn clinical documentation events this quarter (+70% YoY). Ten providers have each recorded over 1mn events since the start of the year.

4. Foundry and in-house models

a. Foundry serves 120k+ customers with revenue up over 3x YoY. Customers building multi-model Agent solutions rose 5x YoY, and open-weight model revenue grew 10x. Substantial advances in OpenAI models last quarter directly boosted Azure OpenAI.

b. MAI targets parity with frontier models within 3–6 months post-release at far lower training cost, passing that cost advantage to customers. MAI Base 2 has completed pretraining, delivering 4–12x training compute efficiency vs. comparable open-weight models, and will power reasoning, code and security models.

c. Adoption: MAI-Cyber handles over 40% of scans for the MDASH code scanner at 18% lower cost per scan. MAI Corp Flash had 3mn users in GitHub Copilot last month. Microsoft Decision One topped 4k customers in its first month.

d. The Agent harness behind Copilot is being opened to Foundry customers, allowing them to retain model choice.

5. Data and Agent governance

a. Fabric has 44k paying customers (+50%+ YoY), with 19k using both Foundry and Fabric. PostgreSQL growth accelerated for the fourth straight quarter, and PostgreSQL customers also using Foundry rose 90% YoY.

b. Microsoft IQ unifies Work IQ, Fabric IQ and Foundry IQ, now used by nearly 60k customers (+25% QoQ).

c. Over 50mn Agents are registered in Agent 365 (+30%+ QoQ). Customers have audited 7.6bn Agent activities with Purview (+5x QoQ). The security approach separates intelligence from permissions, enforcing identity, authorization and isolation outside the model.

6. Infrastructure

a. Demand is accelerating; 24 new DCs were added across five continents this quarter. Time from rack to run for GB300 in key regions shortened ~20% QoQ, and GB300 accounts for the majority of GPU deployments.

b. The fleet is interchangeable, allowing dynamic allocation to the highest-value workloads. Copilot throughput rose over 50% QoQ.

c. First NVIDIA Vera Rubin racks have landed. Maia 200 delivers up to 1.7x tokens output, and Maia 300 targets 40% more tokens per watt than leading GPU systems. Cobalt CPU adoption is strong. Azure growth spans large enterprises and SMBs, across core cloud and AI infra.

7. Devices and consumer

a. Windows is positioned as the 'secure Agent endpoint', partnering with AMD, Intel, Qualcomm and NVIDIA on next-gen AI PCs. Introduced Microsoft Execution Containers (MXC), already adopted by Agents like OpenAI Codex and Claude Code.

b. Bing and Edge continue to gain share. LinkedIn membership surpassed 1.4bn, with feed content viewing time up 15% YoY.

c. Gaming: Minecraft hit record engagement and revenue. The new Call of Duty earned the franchise's best reviews in over a decade, and GTA VI lands on Xbox next month.

2.2 Q&A

Q: As new AI capacity comes online, over the next few years will returns be driven more by utilization and unit costs, or by stacked monetization across Fabric, Foundry and Copilot?

A: Both, with leverage at each layer and additive benefits from cross-layer synergy. We treat all compute as a single pool, maximizing efficiency through DC design, ramp cadence and workload optimization, then allocate based on demand signals.

The new segment structure reflects this: Work IQ underpins Copilot and is called by Foundry. Customers build multi-Agent supply chains spanning structured and unstructured data on Foundry, with databases like PostgreSQL on Azure.

Agents are registered in Agent 365 and governed by our security products — the same infra reported under different product lines.

Q: What underpins the Azure CC growth guide of 49%–50% next quarter?

A: Engineering efficiencies freeing capacity and token pricing, while some supply shifted out; we still need incremental efficiency to hit the guide. In Q1, engineering drove efficiency across the stack — not just GPUs, but how apps and Foundry use the platform — enabling more capacity to be shifted to Foundry; demand still exceeds supply, so freed capacity is prioritized for Foundry.

Token pricing in Q1 was better than expected, with high-quality models supporting solid margins. Efficiency gains span CPUs, storage and memory — Agent workloads are full-stack, and capacity unlocked from the existing fleet monetizes quickly.

These efficiency benefits carry into Q2, but some supply initially planned for Q2 moved into 2H, requiring further efficiency. Token pricing can also change intra-quarter, with lower predictability.

Q: Is Azure AI growth coming more from enterprises or AI-native customers? What share is enterprise workloads?

A: No specific mix disclosed; enterprise demand breadth is just beginning. Examples: Unilever built a fully Agent-driven, touchless system for order intake and trade dispute resolution. Rolls-Royce manages an engine maintenance supply chain with 30k parts on Foundry. Austria's RBI showcased 20+ end-to-end processes.

Customers are shifting from the next hot use case to whether they can build systems, retain ownership, control cost, and ensure reliability and security — similar to the enterprise database wave, a decade-long journey. Mgmt noted that relying on a handful of heavy compute buyers to carry growth for a few quarters is 'not a biz.' We serve large customers, but only at the right price; otherwise, we will patiently build breadth.

Q: With MAI targeting parity with frontier models in 3–6 months and the questioner citing OpenAI at 11% of this quarter's revenue, how will token mix and value evolve across frontier, MAI and enterprise-owned models?

A: Customers have three options — new frontier models, multi-model routing, and training their own models with proprietary traces — and will switch dynamically. First, adopt new models after evaluating results; last quarter, several OpenAI models drove solid demand.

Second, combine multiple frontier models with wide price gaps: in third-party coding Agent benchmarks, Fusion ranks No. 2 or No. 3, achieving frontier-level performance 45% faster and 75% cheaper, forcing Anthropic and OpenAI to compete task-by-task.

Third, like Unilever, accumulate traces with frontier models and then train proprietary models — in this case on MAI, though open-weight options also work. Choices are time-dependent: start on frontier, shift to open-weight or MAI, and potentially switch back when next-gen frontier models arrive; Foundry, Fabric and Copilot must avoid locking customers into a single option.

Q: With operating margin near 50%, how do you sustain that at high growth?

A: Primarily via revenue growth, then operating efficiency, while focusing OpEx on new products and security quality. Segment changes reflect higher integration and new delivery models. Top-line acceleration naturally lifts margins and profit dollars, and we want investors to also watch profit dollars.

Efficiency lowers cost, but the real value is stability, security and reliability — delivering the highest-quality capacity at the lowest unit cost. Significant efficiency gains this quarter allowed revenue growth without higher costs, supporting margins.

Q: As M365 core customers shift to usage-based, can it scale quickly like GitHub without heavy sales effort?

A: Faster in coding, slower outside coding with more sales effort, but early signs are positive. Developers are used to paying by usage, and Copilot Fusion helps customers consume purchased credits most efficiently.

M365 Copilot continually injects each model upgrade's value into the subscription, funded by IT as default, while enabling usage-based where needed. We want customers to understand both options, not overbuy either; thus we invested in Frontier-facing orgs and adjusted the M365 sales motion.

Q: If financing is needed, how would you choose between debt and equity?

A: No specific plan disclosed; we will use leases, bond markets and other channels when attractive. We already use lease financing; with our credit rating we can access debt markets, though we have not tapped them yet; other options exist. With positive FCF and confidence in returns per dollar of capex, we have ample flexibility.

Q: Any disclosure of Copilot seats this quarter?

A: No Copilot seat update this quarter; we discussed revenue and usage momentum instead.

<End of text>

Risk disclosure and statements:Dolphin Research disclaimer and general disclosure