Azure Crosses $100 Billion and the AI Capex Narrative Flips

Three numbers frame Microsoft’s fiscal Q4 earnings report: 43% Azure growth, $100 billion in Azure annual revenue, and $678 billion in commercial backlog. The first shows the cloud business accelerating, not decelerating. The second is a milestone the company had never reached before. The third — a measure of already-contracted future revenue — is the figure that quietly answers the biggest question investors have been asking all year: is the AI buildout actually paying for itself? For the quarter ended June 30, Microsoft posted $90.0 billion in revenue, up 18% year over year and ahead of the $87.72 billion consensus. Operating income rose 18% to $40.6 billion. GAAP net income jumped 31% to $35.8 billion, or $4.81 per diluted share, while adjusted EPS landed at $4.74 — a 23% increase that blew past the $4.25 analyst estimate. For the full fiscal year: $331.8 billion in revenue, $133.7 billion in net income, and $17.95 in adjusted EPS, up from $13.64 a year earlier. None of that, alone, would have been enough to move the stock. In the prior four quarters, Microsoft beat bottom-line estimates three times and still watched its shares fall in each post-earnings session. The pattern was frustrating for even the most patient bulls. But this time was different: MSFT surged 16.96% to $456.76 the following day, erasing months of accumulated doubt in a single session. That doubt was never about revenue or earnings quality. It was about the shape of the story. For four consecutive quarters, the market heard Microsoft say “AI demand is strong, capex is rising, profits are fine” and responded with a shrug. The Q4 report changed the terms of the debate — because Azure stopped decelerating and started pulling away.

Azure and other cloud services grew 43% in constant currency in Q4, up from roughly 40% in Q3, ahead of the company’s own guidance of 39–40%, and ahead of the ~41% that analysts had been modeling after management guidance. Microsoft’s guidance for Q1 FY2027 points to approximately 45% constant-currency growth — an acceleration, not a stabilization. That is the single most important number in the report. A business of this size is not supposed to accelerate. The fact that it did, while carrying a fully loaded AI infrastructure buildout, reframes how the growth story should be read. | Metric | Q4 FY2026 | YoY Change | |---|---|---| | Azure & other cloud services growth | 43% | vs. ~40% in Q3 | | Azure full-year revenue | $100B+ | +41% | | Microsoft Cloud quarterly revenue | $59.3B | +27% | | Microsoft Cloud full-year revenue | $214B+ | +27% | | Intelligent Cloud segment revenue | $39.3B | +32% | Microsoft Cloud — the broader bundle including Azure, Office 365, LinkedIn, and Dynamics — generated $59.3 billion in the quarter, an annualized run rate of $237.2 billion. Intelligent Cloud, the segment that houses Azure, reached $39.3 billion in quarterly revenue, up 32%. CEO Satya Nadella tied the milestone to adoption in the earnings release: “Azure revenue surpassed $100 billion for the first time, and Microsoft 365 Copilot reached over 30 million paid seats, reflecting the confidence customers are placing in us to power their AI transformation.”

The AI growth engine: $37 billion run rate, 30 million Copilot seats

The AI numbers are no longer a side story. Microsoft’s annualized AI revenue run rate has crossed $37 billion, up from roughly $13 billion a year earlier — a 175% increase in a single fiscal year. At that scale, absolute additions still dwarf most companies’ total cloud revenue. Several metrics from Q4 illustrate how the AI layer is compounding: - Microsoft 365 Copilot surpassed 30 million paid seats. Quarterly net seat additions more than doubled from the prior quarter. - Customers with deployments exceeding 50,000 seats increased sevenfold year over year. - The share of enterprise customers rolling out Copilot to a majority of their information workers grew nearly 75% quarter over quarter. - Conversations per user nearly doubled from a year earlier. - GitHub Copilot reached 50 million users, with Copilot revenue accelerating more than 60% quarter over quarter. GitHub’s overall user base now sits at 225 million. - Azure AI Foundry hit 100,000 customers, with revenue more than doubling year over year. The infrastructure underneath is scaling just as aggressively. Microsoft added 31 data centers across five continents in Q4 alone, bringing the fiscal year total to 88. The company added roughly 1 GW of new data center capacity in the quarter and says it remains on track to double total compute capacity within two years. The time required to bring new GPU capacity online in major regions has been cut by nearly half over the past year. The demand-supply imbalance is still there — management said as much on the call. But the speed at which new capacity converts to revenue has changed. CFO Amy Hood described the situation plainly: “Demand continues to exceed available supply. You can see it in the pricing of even spot-market capacity.” Every new rack is going straight to work.

The capex question gets a real answer

Capital expenditure came in at $41 billion in Q4, up 70% year over year. Roughly two-thirds of that — about $27 billion — went into short-lived assets: CPUs and GPUs, the kind of equipment that gets consumed and replaced quickly. FY2026 total capex exceeded $145 billion. For FY2027, Microsoft guided to roughly $175 billion — down from the approximately $190 billion previously outlined, but explicitly not because of reduced physical investment. What changed is accounting, not ambition. CFO Amy Hood confirmed on the earnings call that “expectations for investment have not changed other than the accounting standard change.” Microsoft extended the depreciation period for data centers and buildings from 15 to 25 years, and reclassified some financing leases as operating leases. The $190 billion number became $175 billion on paper; the actual silicon hitting data center floors is a different story. This is the part of the narrative that had investors worried, and for good reason. Free cash flow in Q4 fell 23% year over year to $19.6 billion, even as operating cash flow rose 30% to $55.4 billion. The heavy mix of short-lived assets means depreciation is hitting the income statement faster than the longer-lived real estate. But there’s a counter-argument that gained a lot of traction after this report: the contracted demand is growing faster than the spending. Commercial remaining performance obligation (RPO) surged 84% year over year to $678 billion. Around 30% of that backlog is expected to convert to revenue within 12 months; obligations extending beyond that period grew 112% year over year. Even excluding OpenAI, commercial RPO still rose 25% — a critical nuance for investors worried that this is a two-company bubble. As Investing.com put it in a post-earnings analysis: “CapEx is buying capacity that’s already sold.” That one sentence captures why this quarter felt different. The market had been treating Microsoft’s AI investment as a leap of faith. The RPO data turns it into a math problem: the money is contracted, the capacity is being deployed, and the remaining question is simply whether margins hold up long enough for the cash flow to catch up.

The accounting controversy: 25-year depreciation, 8% “cut”

The depreciation change drew immediate pushback from financial media sharp enough to notice what happened. Chinese outlet Wallstreetcn was characteristically blunt: the market celebrated an 8% capex cut that doesn’t really exist. That critique has merit and deserves attention. AI infrastructure is a fast-moving asset class: GPU generations turn over faster than the buildings that house them. If a data center shell has a 25-year useful life but the chips inside are replaced every four to five years, the depreciation schedule may look good on paper while the actual hardware is cycling out far sooner. Critics argue this is accounting relief, not operating improvement. Still, Microsoft is not inventing the playbook. Google extended server and network equipment useful lives in 2023, and Meta made a similar adjustment in 2022. The difference this earnings cycle is that Google raised its 2026 capex outlook to $195–205 billion, up from $190 billion, while Microsoft’s number fell — with a note that says nothing physical changed. Both companies are spending heavily; the optics around that spending are now measurably different. Microsoft’s view is simple and consistent: extending the depreciation period had minimal impact on FY2027 operating profit, CFO Hood said. And the company’s operating margin held at 45% despite $41 billion in quarterly capex, which is a number that gets harder to dismiss as purely accounting-driven.

The competitive picture: AWS’s lead narrows

Azure’s $100 billion milestone carries a competitive dimension that extends well beyond any single company’s ledger. According to Synergy Research Group, AWS still leads global cloud infrastructure spending with roughly 28–30% market share, Azure holds about 20–25%, and Google Cloud is in the 10–13% range. The three hyperscalers collectively control about 68% of the market — a share that has been stable for two years. But the growth rates underneath that stable aggregate are anything but stable. Azure grew 43% in Q4, making it the fastest-growing of the three in absolute terms. Google Cloud, per Synergy Research, grew at roughly 63%, the fastest percentage rate. AWS’s growth, according to the same data, has slowed to the high-20s percentage range. The Register framed the shift concisely: “AWS is still the largest beast in this sector, taking 28 percent of the market, but its lead over Microsoft is now less impressive, with the Redmond giant making up another 20 percent.” At the current trajectory, Azure’s absolute annual revenue gap to AWS is closing faster than most observers expected — though the gap remains material. AWS still holds a commanding lead in raw scale and in the broader ecosystem that startups and cloud-native teams default to. What Azure has that AWS doesn’t is the Microsoft 365 distribution base and the enterprise sales motion to go with it.

What the analysts and market are saying

Wells Fargo described the $100 billion Azure milestone in exactly the terms bulls had been waiting to hear: “a milestone bulls have been pointing to as the moment the AI infrastructure investment starts to look like a real business rather than a capital consumption story.” Bernstein highlighted the street’s under-appreciation of the guidance: Azure accelerating to roughly 45% in the next quarter, combined with Copilot passing 30 million paid seats, suggests the AI commercialization engine has more room than consensus models reflect. Bank of America framed the report as the response to the central debate: “AI execution remains the central debate” going into the quarter, and Azure growth at or above the guided range was precisely what the stock needed to stabilize. eMarketer’s research team wrote simply that “Azure was central to the story” in a note published the day after the print. Dongwu Securities maintained a Buy rating on Microsoft and raised earnings forecasts following the beat; Caitong Securities kept an Overweight rating. The one beat that finally stuck — that’s the phrase several analysts landed on independently. In a way, that alone is a more meaningful signal than any valuation multiple: the market’s prior pattern of punishing Microsoft on earnings was a statement of skepticism, not a trading anomaly. Q4 was the first report to crack that pattern.

Risks that haven’t gone away

If there’s a lesson from the past four quarters, it’s that Microsoft’s problems don’t disappear just because one earnings report looks good. Several caveats deserve a place in the conversation. Copilot retention is still a black box. Microsoft has not disclosed renewal rates, usage intensity, or seat churn for Microsoft 365 Copilot. Third-party estimates vary widely, with some enterprise adoption analyses putting first-year active usage between 30% and 55% of purchased seats, and weekly active usage as low as 20–30% in certain environments. One independent survey found that 30% of seats were unused or underused within the first 90 days. If the engagement reality sits at the low end, the 30 million paid seats number becomes a renewal risk rather than a new revenue floor. OpenAI concentration remains real. Based on management’s disclosure that excluding OpenAI, commercial RPO grew 25%, analysts estimate OpenAI-related commitments account for roughly $218 billion of the total — about a third of the $678 billion backlog. For a company Microsoft’s size, that is an unusually high customer concentration. Microsoft is working to diversify supply by offering over 11,000 models on Azure, and the number of customers building applications on multiple AI providers has grown fivefold since the start of the year. Still, the single-customer dependency has not structurally changed inside the RPO numbers. Legal and regulatory headwinds are building. The EU AI Act’s substantive obligations take effect in August 2026, and EU regulators have preliminarily flagged both Microsoft and AWS as potential Digital Markets Act gatekeepers. A derivative lawsuit filed in late June alleges Microsoft’s board breached fiduciary duties related to AI investments and copyright compliance. A separate securities class action covering May 2025 through January 2026 remains pending. Capacity constraints still cap the valuation story. Management’s own guidance assumes the supply picture will keep improving. If data center construction or GPU deliveries slip, the 45% Azure growth guide could come under pressure from the supply side — not the demand side.

The other story nobody is talking about enough

Windows was the business that built Microsoft. It is no longer the business that defines it. The More Personal Computing segment — Windows, Surface, Xbox, and search — generated $12.854 billion in Q4 revenue, down 4% year over year. Xbox hardware revenue fell 32%. Windows OEM revenue declined in the mid-single digits. Nadella has publicly acknowledged Windows is the “lagging” part of the portfolio. Meanwhile, Azure alone is now worth more than five times the personal computing division in annual revenue. Not long ago, a statement like that would have been unthinkable. At $100 billion, Azure’s annual revenue is roughly equivalent to what Microsoft’s entire PC ecosystem generated at the height of the Windows era — now it’s a single segment, growing at 41% a year, inside a company that also has a $214 billion cloud business running on top of it. The organizational shift matters as much as the revenue shift. Microsoft has been consolidating resources around AI infrastructure, cutting roughly 13,000 positions across non-core teams — including 4,800 announced in early July — while funneling compensation budgets into compute, specialized engineering talent, and its $600-million-plus Frontier deployment team. The company is pruning the old tree so the new one can grow faster.

The bottom line

Azure crossing $100 billion in annual revenue is not just a number. It is the missing page of the AI narrative that Microsoft’s bulls have been trying to write for two years: the infrastructure investment is now producing revenue at a scale that most of the Fortune 500 cannot reach, with a growth rate that industry leaders are not matching. The market’s reaction — a nearly 17% jump — suggests the long period of “show me proof” is giving way to a more generous reading of progress. But the stakes for FY2027 are even higher. Microsoft will spend more than $50 billion in a single quarter, and investors have been told the capex keeps climbing while free cash flow remains positive — a delicate balance that depends on continued Azure acceleration and disciplined cost conversion. The demand question is no longer open-ended; the RPO backlog is real and diversified beyond OpenAI. What remains to be proven is whether the conversion happens with the efficiency that supports the margin story. With $100 billion of Azure revenue as a base, and a 45% growth guide for the coming quarter, the burden of proof has quietly shifted. The question now isn’t whether Microsoft can sell AI. It’s how much of that sold demand eventually shows up as free cash flow.