> ## Content Index
> Fetch the complete content index at: https://www.sotkn.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# What Does Microsoft Do? The Azure Conversion Model
- URL: https://www.sotkn.com/stack/hyperscalers/what-does-microsoft-do/
- Published: 2026-09-30T04:50:33.000Z
- Updated: 2026-09-30T04:50:33.000Z
- Description: Microsoft books commercial cloud, spends operating cash to stand up Azure halls, and gets paid as remaining performance obligations convert — demand still exceeds available capacity, so the hall is the stuck step.
- Author: Sean
- Tags: #layer-hyperscalers

*Microsoft books commercial cloud, spends operating cash to stand up Azure halls, and gets paid as remaining performance obligations convert — demand still exceeds available capacity, so the hall is the stuck step.*

`HYPERSCALERS` · `Data as at 29 Jul 2026` · `Next update: after Q1 FY27 results`  
**Affects:** Hyperscalers · Neoclouds · US

It is a [hyperscaler](https://www.sotkn.com/explained/hyperscaler-explained/) in the AI build chain: a large cloud operator whose capital spending pulls chips, packaging, halls, and power. The AI cash, and the stuck step, sit in Azure — converting a commercial remaining-performance-obligation book. For that book, see [remaining performance obligation, explained](https://www.sotkn.com/explained/remaining-performance-obligation-explained/). For the specialist-fleet split, see [neocloud vs hyperscaler](https://www.sotkn.com/explained/neocloud-vs-hyperscaler/). For a peer that prints megawatts and utilisation, see [What Does Oracle Do?](https://www.sotkn.com/stack/hyperscalers/what-does-oracle-do/).

**In short:** Microsoft books commercial cloud, funds Azure halls from operating cash, and invoices as remaining performance obligations convert. In the quarter ended 30 June 2026, Azure and other cloud services grew 43%, Microsoft Cloud was $59.3bn, commercial remaining performance obligations were $678bn, and CFO Amy Hood said customer demand continues to exceed available capacity. It added another gigawatt in the quarter. The model works if halls go live on a clock the book can convert — not if $678bn is treated as this year’s cash.

*Education only, not advice to buy or sell any security. Figures are from company filings and company releases dated as shown.*

## Numbers that matter

Azure and other cloud services

+43%

YoY · quarter ended 30 Jun 2026 · FY Azure surpassed $100bn, +41%

Growing

Commercial remaining performance obligation

$678bn

+84% YoY · 30 Jun 2026 · 2.3-year weighted duration

Booked

Demand vs available capacity

Exceeds supply

Hood, Q4 FY26 call · extra in-quarter capacity “quickly monetized”

Stuck

Capacity added this quarter

+1 GW

31 new datacentres this quarter · 88 this fiscal year · Nadella

Delivery

Capex vs operating cash flow

$41bn / $55.4bn

Cash PPE $35.8bn · FCF $19.6bn · \~two-thirds short-lived CPUs/GPUs

Front-loaded

Microsoft Cloud revenue

$59.3bn

+27% YoY · FY $214.4bn · cloud GM 65% in Q4

Watch mix

Sources: Microsoft FY26 Q4 results, 29 Jul 2026 (company release and metrics); capex, cash flow, remaining-performance-obligation duration, and “demand exceeds available capacity” from the FY26 Q4 earnings call the same day. Status chips are Second Order's read.

**Contents:** [What does Microsoft actually do?](#what-it-does) · [How does it make money?](#how-it-earns) · [What limits it?](#what-limits) · [Who else does this?](#who-else) · [What would change this view](#what-would-change) · [Practical takeaways](#practical-takeaways) · [FAQ](#faq)

---

## What does Microsoft actually do?

Three steps, in order.

Hyperscalers · Microsoft

**Book. Energise. Convert.** 

01 

### Book

Commercial remaining performance obligations swell. A signature is not a hall.

02 

### Energise

Megawatts, CPUs, GPUs. Demand still exceeds available capacity. This is the stuck step.

03 

### Convert

Azure and Microsoft Cloud print when the obligation is performed, not when it is only contracted.

Source: Second Order reading of Microsoft FY26 Q4\. Educational only. Not a buy list.

**1\. It books commercial cloud.** Microsoft sells a broad stack — Microsoft 365, Copilot, GitHub, Dynamics, and Azure. On 29 July 2026 commercial remaining performance obligation grew 84% to **$678bn** as at 30 June 2026\. That is contracted work not yet recognised as revenue. It is not cash in the bank. Hood said the book, including OpenAI, has a weighted average duration of **2.3 years**, that roughly **30%** will be recognised in the next twelve months, and that remaining performance obligations still rose **25%** excluding OpenAI. All sequential growth came from customers outside frontier model companies. See [remaining performance obligation, explained](https://www.sotkn.com/explained/remaining-performance-obligation-explained/).

**2\. It energises halls and accelerators.** That is the operating job that binds. Nadella said Microsoft added **another gigawatt** of capacity in the quarter, opened **31** new datacentres (88 in the fiscal year), and remains on track to roughly **double overall capacity in two years**. Hood was blunt: **customer demand continues to exceed available capacity**. Extra in-quarter capacity was “quickly monetized.”

**3\. It converts the book into cloud revenue.** Azure and other cloud services grew **43%**. Intelligent Cloud revenue was **$39.306bn**, up 32%. Microsoft Cloud was **$59.3bn**, up 27%. For the full fiscal year Nadella said Azure surpassed **$100bn**, up 41%, and Microsoft Cloud surpassed **$214bn**, up 27%. Hood added that nearly **90%** of that full-year cloud print came from customers outside frontier model companies. The AI cash sits in Azure. Office and Copilot still earn; they are not the stuck step on this layer.

| Step     | What it means                                   | Where the risk sits                                   |
| -------- | ----------------------------------------------- | ----------------------------------------------------- |
| Book     | Commercial remaining performance obligations    | Mix (OpenAI vs the rest), duration, conversion timing |
| Energise | GW, CPUs, GPUs, customer-usable capacity        | Power, sites, chips, and the attach clock             |
| Convert  | Azure / Microsoft Cloud recognised as performed | How fast the $678bn book becomes this year’s line     |

---

## How does Microsoft make money?

It gets paid when software, seats, and — this cycle — delivered Azure are recognised as revenue. Three things decide how fast the Azure line can grow:

- **Delivery** — a booked megawatt earns nothing until the hall is live
- **Utilisation of what is live** — Hood said efficiency gains across the CPU and GPU fleet were quickly monetized because demand still exceeds supply
- **Mix and conversion** — remaining performance obligations only become revenue on a clock, and Azure is a larger share of the company than it was a year ago

On 29 July 2026 Microsoft printed **$90.007bn** of total revenue, up 18%. GAAP operating income was **$40.603bn**. Unlike [CoreWeave](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/), it is still profitable on a GAAP operating basis while it spends to stand up AI halls.

**Our arithmetic.** Microsoft Cloud was $59.3bn, or **65.9%** of the $90.007bn total (59.3 ÷ 90.007). Intelligent Cloud was $39.306bn, or **43.7%** of sales (39.306 ÷ 90.007). Operating income of $40.603bn is **45.1%** of sales, matching the company’s “operating margins … 45%.” Azure is not disclosed as a quarterly dollar line — only as a growth rate and the full-year “surpassed $100bn” print — so we do not invent one.

Cash tells a second story. Operating cash flow was **$55.4bn**, up 30%. Capital expenditures were **$41bn**, of which Hood said roughly **two-thirds** was short-lived assets, primarily CPUs and GPUs. Cash paid for property, plant and equipment was **$35.8bn**. Finance leases were **$5.6bn**, primarily large datacentre sites. Free cash flow was **$19.6bn**. **Our check:** 55.4 − 35.8 = **19.6**. Capex of $41bn is a wider definition than cash PPE; it includes those finance leases.

Who gets paid: Microsoft, for live Azure and for the software that rides on it. Money sticks at energised capacity. The remaining-performance-obligation print is a claim on future delivery.

Custom silicon is a cost claim. Nadella said Maia 200 delivers **30%** better performance per dollar than latest fleet hardware — a company claim, dated to the call.

---

## What limits it?

Delivery first. Power and sites second. Mix third.

**1\. Demand still exceeds available capacity.** Adding a gigawatt did not create spare seats. Hood said demand exceeds supply, that this has been true for a number of quarters, and that you can see it in spot-market pricing for assets. Efficiency gains were monetized in the same quarter because the fleet was short. See [speed-to-power](https://www.sotkn.com/explained/speed-to-power-explained-for-investors/) and the [US market door](https://www.sotkn.com/markets/us/).

**2\. Remaining performance obligations are not a shipping schedule.** Hood said roughly 30% of the $678bn book, including OpenAI, should be recognised in the next twelve months, up 37% year-over-year, and that the portion beyond twelve months rose 112%. **Our arithmetic:** 30% of $678bn is about **$203bn** in the next year (0.30 × 678 = 203.4). The other **$475bn** sits on a longer clock (678 − 203 = 475). $203bn over four quarters is about **$51bn** a quarter of commercial conversion — against $59.3bn of Microsoft Cloud this quarter, which already includes seats and software that are not Azure consumption. The $678bn print is not $678bn of FY27 Azure.

**3\. Sites do not all come online at once.** Nadella said the company is on track to roughly double overall capacity in two years. That is a multi-year energise clock, not a quarter. Crusoe’s 28 September 2026 note named a **900 MW** Abilene campus for Microsoft — a builder print, not a Microsoft IT-megawatt disclosure. See the [Wire](https://www.sotkn.com/wire/crusoe-google-armstrong-county-campus-wire/). Do not back-solve live gigawatts from that campus.

**4\. Capex is the cash cost of the same bind.** Hood guided calendar-year 2026 capital expenditures to approximately **$175bn** after a useful-life change that shifts more future datacentre leases from finance to operating leases. The underlying plan was unchanged. Q1 FY27 capex is expected **over $50bn**. FY27 capex “will grow year-over-year.” The company still expects to remain free-cash-flow positive. Microsoft funds the bind from operating cash. The bind is still physical.

The honest uncertainty is conversion. Sold-out demand can still miss an Azure growth rate if halls slip, and a huge remaining-performance-obligation print can miss this year’s Azure. Those are different clocks. +1 GW does not settle both.

---

## Who else does this?

Roles, not a league table.

| Company                           | What it does                                                                                | What makes it different                                                                                                                              |
| --------------------------------- | ------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Microsoft**                     | Software plus Azure; converts commercial remaining performance obligations as halls go live | RPO $678bn; Azure +43%; demand exceeds supply; still FCF-positive on $19.6bn                                                                         |
| **Oracle**                        | Cloud infrastructure plus database; AI cloud capacity delivery                              | Prints MW and GPU utilisation (850 MW, 97.9% full). See [What Does Oracle Do?](https://www.sotkn.com/stack/hyperscalers/what-does-oracle-do/)        |
| **Amazon, Alphabet**              | Broad cloud; large AI capex prints                                                          | Set the demand dial; different capex definitions; see [hyperscaler, explained](https://www.sotkn.com/explained/hyperscaler-explained/)               |
| **Meta**                          | Hyperscaler-scale spend, ads as the main earn                                               | Buys capacity for itself; not a merchant cloud in the Microsoft sense                                                                                |
| **CoreWeave and other neoclouds** | Specialist AI compute, often debt-funded                                                    | One product; overflow when hyperscaler queues are long. See [What Does CoreWeave Do?](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/) |

What separates them is who can deliver accepted megawatts, who funds the hall, and whether demand still exceeds what is live after the add.

---

## What would change this view?

The read on the model weakens if:

- Azure growth slows while remaining performance obligations keep rising — the conversion gap would widen
- Hood’s “demand exceeds available capacity” language disappears in a dated print, which would mean the fleet is no longer the short clock
- Calendar 2026 capex is cut in a way that is not just the finance-to-operating lease reclass
- A named campus slips in a way that *does* hit the \~45% Q1 Azure guide
- Hyperscaler peers cut 2027 capex — the demand-dial risk on the [explained page](https://www.sotkn.com/explained/hyperscaler-explained/)
- First Vera Rubin / Helios attach is late because packaging or [HBM](https://www.sotkn.com/explained/hbm-explained/) is short — the stuck step would then sit in [semiconductors](https://www.sotkn.com/stack/semiconductors/)

It stays intact while demand exceeds live capacity after gigawatt adds, Azure keeps converting into a 40%+ growth line, and capex stays funded from operating cash.

---

## Practical takeaways

- Microsoft’s AI model on this layer is **Book → Energise → Convert**: commercial remaining performance obligations swell; megawatts have to go live; Azure prints after that
- Q4 FY26 (ended 30 Jun 2026): Azure +43%, Microsoft Cloud $59.3bn, remaining performance obligations **$678bn**, **+1 GW** added, demand still exceeds supply
- GAAP operating income $40.6bn (45%) and free cash flow $19.6bn — profitable, spending ahead of conversion
- Capex $41bn versus $55.4bn of operating cash flow; roughly two-thirds is short-lived CPUs and GPUs
- About **30%** of the $678bn book is the next-twelve-month envelope (\~$203bn)
- The stuck step is **capacity delivery**, not “did Microsoft book another contract”
- Next hard read: Q1 FY27 results (Azure guided \~45% constant currency; capex over $50bn)

---

## FAQ

**What does Microsoft do?**  
Microsoft sells software, Copilot, and cloud. The scarce job is delivering Azure capacity customers can use, then converting remaining performance obligations into recognised cloud revenue.

**What does Microsoft do in simple terms?**  
It books long commercial cloud contracts, switches on data-centre megawatts, and charges for Azure once the hall is actually running.

**Is Microsoft profitable?**  
Yes, on a reported GAAP operating basis. It posted $40.603bn of operating income on $90.007bn of revenue in the quarter ended 30 June 2026 (29 Jul 2026). Free cash flow was $19.6bn. Capex is large; it has not turned the cash engine negative.

**How is Microsoft different from a neocloud?**  
A neocloud sells AI compute as the business. Microsoft sells a broad cloud and still earns from software and seats; Azure is the line that is growing fastest and the step that is stuck. See [neocloud vs hyperscaler](https://www.sotkn.com/explained/neocloud-vs-hyperscaler/) and [What Does CoreWeave Do?](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/).

**What is commercial remaining performance obligation?**  
$678bn of contracted commercial performance not yet recognised as revenue as at 30 June 2026\. Hood said roughly 30% should convert over the next twelve months, and that the book still grew 25% excluding OpenAI. It is a backlog clock, not this quarter’s cash.

**Should I buy Microsoft shares?**  
We do not make buy or sell calls, and nothing here is a recommendation. What we can tell you is what the company does, how it gets paid, and which figures would change the picture. The decision is yours.

---

## Keep reading on the map

- **The layer:** [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/) — capex pull and the delivery bind
- **The term:** [Hyperscaler, explained](https://www.sotkn.com/explained/hyperscaler-explained/)
- **The mechanism:** [Remaining performance obligation, explained](https://www.sotkn.com/explained/remaining-performance-obligation-explained/)
- **The split:** [Neocloud vs hyperscaler](https://www.sotkn.com/explained/neocloud-vs-hyperscaler/)
- **The peer:** [What Does Oracle Do?](https://www.sotkn.com/stack/hyperscalers/what-does-oracle-do/) — MW delivered, fleet still full
- **The Wire:** [Crusoe is building Google’s Armstrong County campus](https://www.sotkn.com/wire/crusoe-google-armstrong-county-campus-wire/) — same builder named a 900 MW Microsoft Abilene hall
- **Where it happens:** the [US](https://www.sotkn.com/markets/us/) — [speed-to-power](https://www.sotkn.com/markets/us/us-speed-to-power-september-2026/) is the long clock under the halls

## Members' Analysis: if packaging catches up, does Microsoft's bind stay on power?

A $678bn book into a fleet where demand still exceeds capacity is a delivery bind in a dated print. Members: [Where does the stuck step move?](https://www.sotkn.com/analysis/where-does-the-stuck-step-move/). This page is the free Stack read of the Microsoft model.

## Get the Wire

Free daily brief on what changed in the AI chip map. [Subscribe free →](https://www.sotkn.com/#daily-wire)

## Sean

Writes Second Order by TKN — a plain-English map of AI infrastructure and semiconductors for non-specialist investors. Focus: who gets paid, where money sticks, and which step is stuck. Not investment advice.

[Author page →](https://www.sotkn.com/author/sean/) · [Editorial standards →](https://www.sotkn.com/editorial-standards/)

## How we check this

Figures come from company filings and company releases, dated below. Capacity and timing from the earnings call are the company’s spoken print. Where we work something out, the arithmetic is shown in full.

**Last reviewed:** 30 Sep 2026 · **Next review:** after Q1 FY27 results. Spot an error? Tell us and we will correct it and note the change here. [Our editorial standards →](https://www.sotkn.com/editorial-standards/)

## Sources

1. Microsoft, [Microsoft Cloud and AI strength fuels fourth quarter results](https://news.microsoft.com/source/2026/07/29/microsoft-cloud-and-ai-strength-fuels-fourth-quarter-results-4/?ref=sotkn.com) (company release), 29 Jul 2026 — total revenue, operating income, Microsoft Cloud $59.3bn, commercial remaining performance obligation $678bn, Azure +43%, Intelligent Cloud $39.3bn, FY Azure surpassed $100bn
2. Microsoft Investor Relations, [FY26 Q4 press release](https://www.microsoft.com/en-us/investor/earnings/fy-2026-q4/press-release-webcast?ref=sotkn.com) / [metrics](https://www.microsoft.com/en-us/investor/earnings/fy-2026-q4/metrics?ref=sotkn.com), 29 Jul 2026 — Microsoft Cloud $214.4bn FY, cloud gross margin 65% in Q4 / 66% FY
3. Microsoft, [FY26 Q4 earnings conference call](https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4?ref=sotkn.com), 29 Jul 2026 — +1 GW, demand exceeds capacity, capex $41bn, cash PPE $35.8bn, OCF $55.4bn, FCF $19.6bn, RPO duration 2.3 years and \~30% next twelve months, CY2026 capex \~$175bn, Q1 FY27 capex over $50bn, Azure \~45% CC
4. Second Order, [Crusoe is building Google’s Armstrong County campus](https://www.sotkn.com/wire/crusoe-google-armstrong-county-campus-wire/) — Wire dated 29 Sep 2026; Crusoe named 900 MW Abilene for Microsoft
5. Second Order, [Oracle delivered 850 MW of AI capacity — the fleet still ran 97.9% full](https://www.sotkn.com/wire/oracle-850mw-gpu-fleet-still-full-wire/) — peer Wire dated 24 Sep 2026

All external links `rel="nofollow noopener"`.