--- title: "MSFT (Trans): Pricing Model Shifting from Per-Seat to Usage-Based" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/40337927.md" description: "Below is Dolphin Research's Trans of $Microsoft(MSFT.US) FY26Q3 earnings call. For our earnings analysis, please see 'Post‑breakup pains: When will Microsoft regain its mojo?'I. Core results recap. 1) Shareholder returns: $10.2bn was returned to shareholders this quarter via dividends and buybacks. 2) Key metrics: revenue of $82.9bn (+18% YoY; +15% cc) and GPM of 68% (down YoY, mainly due to AI infra build‑out and increased AI product usage...)." datetime: "2026-04-30T16:54:10.000Z" locales: - [en](https://longbridge.com/en/topics/40337927.md) - [zh-CN](https://longbridge.com/zh-CN/topics/40337927.md) - [zh-HK](https://longbridge.com/zh-HK/topics/40337927.md) author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)" --- # MSFT (Trans): Pricing Model Shifting from Per-Seat to Usage-Based **Below is Dolphin Research’s transcript of** $Microsoft(MSFT.US) **FY26Q3 earnings call. For our earnings take, see** [**The Post-Breakup Pain: When Can Microsoft Regain Its Mojo?**](https://longbridge.com/zh-CN/topics/40336562)**.** **I. Key takeaways** 1\. **Shareholder returns**: $10.2bn was returned via dividends and share repurchases. This was for the quarter. 2\. **Key financials**: Revenue of $82.9bn (+18% YoY, +15% YoY cc). GPM at 68% (down YoY on AI infra spend and higher AI usage, partly offset by efficiency gains in Azure and M365 Commercial Cloud). OP ex +9% YoY (+8% cc), OP +20% YoY (+16% cc), OPM rose to 46%. EPS $4.27 (+21% YoY ex-OpenAI investment impact, +18% cc). 3\. **CapEx**: CapEx was $31.9bn, down QoQ. About two-thirds went to short-lived assets (mainly GPUs/CPUs), with the balance to long-lived assets monetizable over 15+ years. Finance leases were $4.7bn, and PP&E cash outlay was $30.9bn. 4\. **FY26Q4 guide**: Total revenue of $86.7–87.8bn (+13–15% YoY), with acceleration in Commercial partly offset by Consumer. \- **Segments**: Productivity & Business Processes $37.0–37.3bn (+12–13% YoY); Intelligent Cloud $37.95–38.25bn (+27–28% YoY), with Azure +39–40% YoY cc; More Personal Computing $11.75–12.25bn. **\- Azure:** Next quarter growth expected at +39–40% cc, a step-up vs. this quarter’s 39%. Management also expects a further uptick in H2 CY26. \- **Line items**: M365 Commercial Cloud +15–16% cc on an adjusted basis (lapping last year’s 2ppt in-period rev. recognition), reported +13–14%. M365 Commercial products low single-digit growth; M365 Consumer Cloud low-20s% (beginning to lap last year’s price increase). LinkedIn ~+10%; Dynamics 365 low double-digit; on-prem server down low single-digit. Windows OEM down high-teens (Win10 EOS -6ppt, channel inventory digestion -6ppt, DRAM price headwinds to PC market -6ppt). Windows OEM and Devices combined down mid-to-high teens; Search ads ex-TAC up high single-digit. Xbox content and services down low-teens; hardware revenue down YoY. \- **Costs**: COGS $29.4–29.6bn (+22–23% YoY), incl. ~$350mn one-time voluntary separation costs. Opex $19.3–19.4bn (~+7% YoY), incl. ~$550mn one-time separation costs, for a total of ~$900mn one-offs. Microsoft Cloud GPM ~64% (down YoY on AI investment and rising GitHub Copilot usage). Q4 adj. effective tax rate ~19%. Full-year FY26 OPM to expand by ~1ppt YoY. 5\. **CapEx and FY27 outlook**: FY26Q4 CapEx raised to $40bn+ (QoQ increase of ~$5bn from component price hikes; finance leases introduce in-period recognition swings). CY26 CapEx ~ $190bn (with ~ $25bn from component inflation). Supply tightness likely persists at least through 2026, while Azure growth modestly accelerates in H2 vs. H1 on a calendar basis. FY27: total headcount to decline YoY; opex growth mid-to-high single-digit; aim for double-digit growth in revenue and OP. **II. Earnings call details** **2.1 Management highlights** 1\. **Strategy** **a. AI ARR topped $37bn (+123% YoY).** b. The company is in the early phase of a platform shift where agents will become the dominant workload, expanding TAM and reshaping value creation. c. Two priorities: build world-class cloud and AI infra for the agentic computing era; and build high-value agentic systems in productivity, coding, and security. These two layers reinforce each other. 2\. **AI infrastructure** a. Full-stack optimizations (DC design, silicon, system software, model architecture and tuning) are translating into operational gains. YTD, dock-to-live time for the largest regional GPU clusters shortened by nearly 20%. The Fairwater DC in Wisconsin went live six weeks early, pulling revenue forward; Copilot inference throughput on mainstream models improved by 40%. **b. Added 1GW of capacity this quarter, with the goal to double capacity over the next two years unchanged; announced new DC investments across four continents.** **c. First-party silicon**: Maia 200 AI accelerators are live in Iowa and Arizona DCs, delivering 30%+ more tokens per dollar vs. the latest fleet silicon. Cobalt server CPUs are deployed in nearly half of DC regions, running workloads for Databricks, Siemens, and Snowflake, and are scaling rapidly. Azure Boost and in-house networking, security, and virtualization chips support millions of servers. 3\. **Agent application platform (Foundry)** a. Foundry offers a broad model catalog (OpenAI, Anthropic, open source, etc.). Over 10k customers have used more than one model, and 5k have used open-source models. Customers using Anthropic and OpenAI models doubled QoQ. b. 300+ customers are on track to process over 1 trillion tokens on Foundry this year, with the pace accelerating ~30% QoQ. c. Bayer is building an internal agent platform on Foundry with over 20k MAUs. **d. First-party models**: launched MAI-Transcribe-1 (SOTA speech-to-text) and MAI-Image-2 (among top-tier image generators), already used in Bing and PowerPoint image creation. Transcribe-1 improved GPU efficiency by 67%, while Image-2 improved by up to 260%. MAI models are now offered via Foundry to commercial customers such as Shutterstock and WPP. e. Continued innovation on OpenAI IP: Copilot’s Work IQ multi-step retrieval and adaptive reasoning in Researcher reduce latency and improve accuracy. These features are designed to boost user value. 4\. **Copilot Studio and Agent 365** a. Nearly 90% of the Fortune 500 have built active agents using low/no-code tools. The consumption-based Copilot credit SKU nearly doubled QoQ. b. Agent 365 provides identity, governance, and security control planes for agents. Tens of thousands of companies are already managing tens of millions of agents on it. 5\. **M365 Copilot (knowledge work)** **a. Net adds of paid seats rose +250% YoY, the fastest since launch, taking total paid seats to over 20mn.** b. Customers with 50k+ seats quadrupled YoY; Accenture has over 740k seats (the largest to date). Bayer, Johnson & Johnson, Mercedes, and Roche each committed to 90k+ seats. c. Work IQ corpus surpassed 17EB (+35% YoY), adding billions of emails, docs, and chats daily, hundreds of millions of Teams meetings, and millions of SharePoint sites. d. Shipped 625+ Copilot updates over the past 12 months (+50%); Agent Mode is now default in Word, Excel, and PowerPoint. Cowork enables task delegation; chat uses multi-model intelligent routing by default; Agents can use Critique and Council to ensemble models for optimal responses. e. First-party agents’ MAUs grew 6x YTD. Queries per user on Copilot rose nearly 20% QoQ, and weekly engagement is now at Outlook levels. 6\. **Biz apps (Dynamics)** a. A new 'seat + usage' pattern is emerging. Nearly 60% of customer service clients have purchased usage-based credits. b. HSBC uses Dynamics 365 prebuilt agents for customer inquiries, reducing case resolution time by 30%+. c. LinkedIn Talent Solutions’ agentic products surpassed $450mn in ARR. 7\. **Coding (GitHub)** a. GitHub growth is at a record pace driven by agentic coding. GitHub Copilot is used by nearly 140k orgs, with enterprise subs up almost 3x YoY; most users run multiple models. b. GitHub Copilot CLI MAUs are almost doubling month-over-month. c. Announced a shift to usage-based pricing for GitHub Copilot effective Jun 1, aligning pricing with actual usage and costs. 8\. **Security** a. AI is compressing the 'vuln-to-exploit' window. Defender provides protection as patches are released for AI-discovered vulns, and multimodal AI harness scanning is close to productization. b. Security Copilot customers doubled YoY. The data security triage agent processed 2mn unique alerts in the quarter; Purview has audited 35bn Copilot interactions (+7x YoY). 9\. **Consumer** a. Windows, Xbox, Bing, and Edge focus on core UX. Windows improved performance on low-memory devices and the Update experience; last week’s Game Pass changes on Xbox were in response to gamer feedback. b. Windows has over 1.6bn MAU devices; Edge has gained share for 20 straight quarters; Bing MAUs topped 1bn for the first time. c. LinkedIn has 1.3bn members and remains a key B2B channel for sales and advertisers. d. Xbox MAUs and cloud gaming hours hit record highs. M365 Consumer subs are near 95mn, and default Agent Mode is lifting early satisfaction. **2.2 Q&A** **Q: How does strong demand translate into Commercial bookings? Longer term, who pays for AI?** A: **Bookings exhibit cyclicality tied to expirations and multi-year Azure deals.** More importantly, the model is shifting from seat-based to 'worker + agent', creating a dual track of license plus usage. This changes the shape of bookings: seats remain, but Azure-like meters add usage that may bypass bookings and be billed on consumption. In the near term, bookings may not capture this shift and should be viewed through a broader lens. **All 'per user' businesses (productivity, coding, security) will transition to 'per user + per usage'.** Coding is already at scale, and the GitHub Copilot pricing change reflects this. Funding ultimately comes from measurable output agents deliver—customer service, individual productivity, team productivity, and business processes are compressed by agents, driving cost-out or revenue lift. Copilot’s multiple modalities (chat, reasoning, Cowork, and Agent Mode across Word/Excel/PowerPoint) are embedded in specific task flows. When customers see compressed workflows, higher revenue, and lower cost, usage follows. Therefore, it will be less about seat coverage and more about 'intense users + intense usage'. **Q: CapEx guidance implies a sharp step-up in H2 on a calendar basis. How confident are you in overcoming physical supply constraints, and how do you balance 1P vs. 3P capacity?** A: We are confident in breaking through physical bottlenecks as **the industrial logic of the supply chain is now aligned.** Much of the work must be done within the end-to-end cycle to bring capacity online, including securing CPUs, GPUs, and storage to support strong demand signals. On allocation, **the FQ4 Azure cc guide of 39–40% reflects optimizing supply across varied Azure workloads.** Copilot usage moved to a new level this quarter across coding and productivity, and we are confident security will follow. Looking to **H1 FY27 (H2 CY26), we expect some acceleration as we keep improving efficiency and speed up getting DCs ready for revenue.** The 1P vs. Azure 3P balance will remain dynamic, but **we will do everything we can to accelerate compute coming online, which drives the H2 CapEx step-up.** **Q: Will AI structurally lift margins? What might investors be missing?** A: AI’s position in the lifecycle is comparable to early cloud, but with higher margins. The key is aligning the business model with application value and **capturing more value via consumption and usage-based pricing, which may be underappreciated today.** **We are raising margins through multiple levers: leveraging partner IP (frontier model IP from OpenAI is free to us long term), extracting infra-layer margin via first-party silicon, and efficiency work across hardware and software.** While we are accelerating capacity build-out, we are also driving ongoing efficiency gains. The most important point: **once you move to usage-based, you must deliver very high customer value so that usage translates into productive outcomes, enabling sustained TAM expansion and ROI.** **Q: How do you address the gap between CapEx growth and revenue growth? Can you break down short-term vs. long-term assets and monetization timing?** A: For Azure, given its scale and expected acceleration (from 39–40% higher), most spend is going to **short-lived assets** that tie directly to revenue. Roughly one-third goes to 15-year assets or lease-driven timing effects, which can make the data look noisy. This is reminiscent of the last cloud build-out. With such a broad TAM and supply undershoot, we have strong confidence in platform-side ROI. **What investors really care about is whether 'seat + consumption' will monetize at the app and services layer.** M365 Commercial Cloud accelerated QoQ and the FQ4 guide remains strong, marking the start of that monetization; GitHub follows the same logic, with usage-driven consumption accelerating revenue. Also consider RPO of $600bn+ yet to be recognized, which excludes the accelerating Copilot seat growth, and we are very confident in that figure. Next, we must land capacity as fast as possible and convert it to revenue. In 2026, the most exciting AI-era apps are still Word/Excel plugins and the coding CLI, underscoring our structural edge in **knowledge work, coding, and security**—all very large TAMs. We need CapEx to bring capacity online in time to match usage growth. Model capability improves exponentially—for example, Excel’s Agent Mode may see little use until it gets good, then rapidly take over lots of work—so we must be ready for such step-changes. **Q: Reflect on Copilot’s tech and commercial milestones over the past three months. What worked and what didn’t, and how does that inform E7 and Copilot Cowork?** A: Think in terms of **product modalities**. **Chat**: reasoning grounded in work context (Work IQ). **Agents**: research analysts or customer-built custom agents. **Edit Mode**: polish directly within Word/Excel after chat insights. **Cowork**: a new mode where tasks are **delegated** rather than done via interactive prompts. Copilot engagement is already at Outlook-like levels, so it is not merely useful—it is used daily at scale. Why is it useful? Because it combines **multi-model capability** with **real-time context** (dynamic data from Teams, SharePoint, etc.). Our goal is to decouple the underlying architecture from specific models so customers can orchestrate multiple models together, and the emerging 'seat + usage' model is proving out. **Q: What is the financial impact of the changes to the OpenAI agreement? Does this imply faster diversification at the model layer?** **A:** We are pleased with the OpenAI partnership and focused on sustaining a win-win. First, we have free IP rights to frontier models through 2032 and will fully utilize these IP rights. Second, OpenAI is a large customer of ours for both AI accelerators and general compute. Third, we own equity in OAI. The framework adapts to evolving customer expectations for model diversity, and we are happy with where it is. Two modeling points: revenue sharing continues through 2030; for IP usage, we no longer owe OAI revenue sharing. **Q: E7 remains primarily seat-based with usage add-ons, and customers want predictability as AI costs climb. How do you keep predictability while growing usage-based mix, and what does the mix look like in 3–5 years?** A: Predictability in budgeting and procurement is important. **A seat is essentially a right to consume—it bundles a base consumption entitlement into the seat.** For users, it is a convenient way to buy, effectively purchasing a consumption pack assigned to a specific seat or agent. **Overages move to pure usage billing, and longer-term usage commitments earn discounts,** which is the path forward. From the customer’s perspective, **the ultimate yardstick is whether tokens produce commensurate value.** As we discuss IT budgets, **they will be reshaped by tying spend to business outcomes.** Those outcomes will increasingly flow into IT budgets and may be reallocated from other P&L lines (e.g., **OpEx**). **Risk disclosure and disclaimer:**[**Dolphin Research Disclaimer and General Disclosures**](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md) - [DXYZ.US](https://longbridge.com/en/quote/DXYZ.US.md) - [002245.CN](https://longbridge.com/en/quote/002245.CN.md) - [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md) - [MSFL.US](https://longbridge.com/en/quote/MSFL.US.md) - [MSFO.US](https://longbridge.com/en/quote/MSFO.US.md) - [MSFD.US](https://longbridge.com/en/quote/MSFD.US.md) - [MSFU.US](https://longbridge.com/en/quote/MSFU.US.md) - [MSFX.US](https://longbridge.com/en/quote/MSFX.US.md) - [MSFY.US](https://longbridge.com/en/quote/MSFY.US.md)