Microsoft closed at $381.58 on July 23, near a one-year low, three days before it reported fiscal fourth-quarter earnings. It closed at $464.72 on July 31, up roughly 22 percent from that level. In between: a revenue beat that surprised even bullish analysts, an Azure growth number that reversed a share-loss narrative building since spring, and a Federal Reserve decision that split the rate-setting committee three ways on the same afternoon Microsoft reported.
Our July 26 preview identified three disclosures that would determine whether the quarter validated the bull case or the structural-decline thesis this publication has tracked since spring: Azure’s growth trajectory relative to Google Cloud, a dollar-denominated Copilot revenue figure distinct from seat counts, and fiscal 2027 capital-expenditure guidance. Two of those three arrived clearly. The third did not — and the one that didn’t is the one that matters most for judging whether Microsoft’s AI investment is actually converting into revenue rather than adoption metrics.
The Numbers That Beat
Microsoft reported revenue of $90.0 billion for the quarter, up 18 percent year-over-year against a consensus estimate that ranged roughly $81.4 billion to $87.6 billion depending on the source — a wide beat by any measure. Non-GAAP earnings per share came in at $4.74, against consensus estimates clustered between $4.21 and $4.33. GAAP diluted EPS was $4.81, up 32 percent year-over-year, with GAAP net income of $35.8 billion, up 31 percent — though a portion of that headline growth reportedly reflects a one-time investment gain rather than operating performance alone, a distinction worth holding onto when reading the raw percentage figures.
The number that mattered most to the market was Azure. Microsoft’s cloud infrastructure business grew 43 percent year-over-year, ahead of a consensus estimate near 40 percent and an acceleration from the prior quarter’s growth rate. In dollar terms, Azure crossed $100 billion in annualized revenue for the first time — the first hyperscaler cloud business to reach that specific threshold as a standalone disclosed figure at this growth rate. The broader Intelligent Cloud segment reported $39.31 billion in revenue, up 31.6 percent, and Microsoft Cloud overall — the combined commercial cloud figure spanning Azure, Microsoft 365, and Dynamics — came in at $59.3 billion, up 27 percent.
The stock’s reaction tracked the numbers closely. Shares rose roughly 8 percent in after-hours trading the night of July 29, reportedly extended those gains at the July 30 open by more than 14 percent at one point, and closed July 31 at $464.72 — a single-week move from $381.58 that represents one of the sharpest earnings-driven rallies Microsoft has produced in recent memory, off a base that had been sitting near its 52-week low days earlier.
What the Azure Number Answers, and What It Doesn’t
The AI-squeeze framing this narrative has applied to Microsoft since earlier in the year rested partly on a comparison that looked unfavorable in isolation: Azure growing near 40 percent against Google Cloud’s reported growth rate near 82 percent, more than double Microsoft’s pace. That comparison was never a clean one — a smaller revenue base grows faster in percentage terms almost mechanically — but it had accumulated enough repetition in market commentary to function as a genuine share-loss narrative heading into this report.
Azure’s acceleration to 43 percent, on a base that just crossed $100 billion annualized, complicates that narrative without fully retiring it. Microsoft’s own guidance for the next quarter points to roughly 45 percent constant-currency Azure growth — continued acceleration, not deceleration, which is the single most important trend line for anyone tracking whether Microsoft is losing the cloud-infrastructure race or simply running a larger, more mature business that happens to compound more slowly in percentage terms than a smaller competitor. Google Cloud’s growth rate for the same period was not directly addressed in Microsoft’s own disclosures, and a fair comparison requires waiting for Alphabet’s own report before drawing conclusions about relative share shifts. What can be said with the data in hand: Azure did not decelerate, which was one of the two outcomes the bear case needed, and the acceleration was large enough to move the stock roughly 20 percent in a week.
Copilot’s Missing Number
The Copilot monetization gap this publication documented earlier in the year is the thread that did not close with this report, and it is worth being precise about what did and did not happen.
Microsoft disclosed that Microsoft 365 Copilot has surpassed 30 million paid seats, up from roughly 20 million as of the prior quarterly disclosure in April — a genuinely large adoption jump, on the order of 50 percent sequential seat growth in three months. That is the kind of number that supports a bull thesis built on Copilot reaching enterprise scale.
What Microsoft did not disclose, again, is a dollar-denominated Copilot revenue figure. Copilot’s financial contribution remains folded into segment-level reporting — Microsoft 365 Commercial, Intelligent Cloud, and the broader Microsoft Cloud aggregate — rather than broken out as its own line. This is the third consecutive quarter this publication has tracked in which Copilot’s seat-count growth has outpaced the market’s ability to verify what that growth is actually worth in revenue terms. Thirty million paid seats at even a conservative average price point would represent a meaningful, quantifiable revenue stream; Microsoft’s decision not to disclose that number, even as it discloses adoption metrics prominently, is itself informative. Companies generally lead with the disclosure that makes their story strongest. A company that leads with seat count rather than revenue, quarter after quarter, is making an implicit statement about which number it would rather investors focus on.
None of this means Copilot revenue is small, or that the adoption numbers are meaningless — 30 million paid enterprise seats is a real distribution achievement regardless of the associated dollar figure. It means the specific question this narrative has been asking since spring — is Copilot’s adoption converting into revenue at a rate that justifies the AI infrastructure spend funding it — remains open after this report, in a quarter when nearly every other major question got a clear answer.
The Capex Number That Moved and What Moved It
Fiscal 2027 capital-expenditure guidance, the third item on the July 26 checklist, arrived — but not as a hard figure, and the way it arrived deserves more scrutiny than the headline treatment it received in most coverage.
CFO Amy Hood said capital expenditure will “grow year over year” into fiscal 2027, without providing a specific dollar target. She cited a calendar-2026 capex figure near $175 billion — notably below the roughly $190 billion some analyst models had been circulating before the report — and disclosed more than $50 billion in capital spending for the first quarter of fiscal 2027 alone, an annualized pace that, if sustained, would land well above the $175 billion calendar-year figure.
The gap between the lowered $175 billion figure and the $190 billion the market had been modeling is explained, according to the earnings disclosures, primarily by an accounting change: Microsoft extended the useful life it assigns to data-center assets from 15 years to 25 years, and reclassified some data-center leases from finance leases to operating leases. Both changes affect how spending is categorized and depreciated on the balance sheet without necessarily reflecting a reduction in actual cash spent building AI infrastructure. This is a meaningful distinction for anyone using the headline capex number as a proxy for AI investment intensity. A company can report a lower capex figure through a depreciation-schedule and lease-classification change while its actual infrastructure buildout continues on the same trajectory — the $50 billion-plus first-quarter fiscal 2027 figure, which is not subject to the same reclassification ambiguity, is arguably the more reliable signal of near-term spending pace, and it points toward continued aggressive investment rather than restraint.
The widely circulated $255 billion to $260 billion figure for full fiscal-year 2027 capex that appeared in pre-earnings analyst commentary was a market estimate, not a number Microsoft itself provided. Attributing that figure to Microsoft’s own guidance would be inaccurate; what the company actually said was directional — spending grows — without the specificity investors were hoping for going into the report. On the specific disclosure this narrative flagged as decisive, the report delivered a qualitative answer wrapped in a genuinely complicated accounting change, rather than the clean number the market was positioned for.
The Fed Split the Same Afternoon
Microsoft’s earnings landed on a trading day that also carried a Federal Open Market Committee decision, and the two events are worth examining together rather than in isolation, because the market absorbed both within hours of each other.
The Fed held its target rate at 3.50 to 3.75 percent for a fifth consecutive meeting — but the vote was 9-3, not unanimous, with three regional Fed presidents — Beth Hammack of Cleveland, Neel Kashkari of Minneapolis, and Lorie Logan of Dallas — dissenting in favor of a 25-basis-point hike. A three-way hawkish dissent of this kind is unusual; coverage described it as the most contentious split in years. Chair Kevin Warsh’s press conference matched the tenor of the vote: asked about the division, Warsh said “I asked for a good family fight, and I got one,” and elsewhere in the session stated plainly that “there is no soft inflation target — there’s only a target, and it’s 2 percent,” language that markets read as leaving the door open to a September hike depending on incoming data.
Bond markets took the hawkish signal seriously even as equity markets celebrated Microsoft’s numbers: the 30-year Treasury yield rose from roughly 5.1 percent to 5.21 percent during Warsh’s remarks, its highest level since 2007. That is not a small move for a single afternoon, and it reflects a market pricing meaningfully higher odds of further tightening than existed a week earlier — precisely the environment in which growth stocks with large capital-expenditure programs are supposed to struggle, given how sensitive long-duration cash flows are to the discount rate.
Microsoft rallied anyway. That divergence — a hawkish Fed decision landing the same afternoon as a stock-specific catalyst large enough to override the macro signal — is itself a data point worth registering against the framing this publication has applied to other assets in 2026. Bitcoin’s price action through July consistently showed macro and rate expectations dominating crypto-specific news; Microsoft’s earnings week showed the opposite pattern for a company whose fundamentals delivered a genuine surprise. The lesson is not that company-specific news always beats macro — plenty of counterexamples exist — but that a large enough fundamental surprise can, for at least one trading session, override a meaningfully hawkish rate signal. Whether that pattern holds through August, as the bond market continues digesting a 2007-level 30-year yield and a Fed that has now shown three sitting presidents want to hike, is a separate and open question this narrative will continue tracking.
Why a Third Quarter Without a Number Is Different From a First
It is worth being precise about why the absence of a Copilot revenue figure carries more weight now than it did two quarters ago, because the significance of an unreported number compounds with repetition in a way that is easy to understate.
The first quarter Microsoft declined to break out Copilot revenue, the omission was unremarkable — plenty of product lines take years to reach the scale at which standalone disclosure becomes standard practice, and a company can reasonably argue that a new product category isn’t yet material enough to warrant its own line item. The second quarter without disclosure, arriving alongside a seat-count jump from roughly 11.5 million to 20 million, started to look like a pattern rather than an early-stage reporting choice — the adoption numbers were growing fast enough that investors reasonably began asking when the revenue figure would follow. This third quarter, with seats now above 30 million and adoption having roughly tripled since the pattern was first flagged, the absence of a dollar figure has shifted from a reporting-maturity question to a disclosure-choice question. Companies that want to demonstrate a product’s financial contribution generally find a way to do so once the adoption numbers are large enough to make the disclosure flattering. Microsoft’s continued silence on this specific figure, even as its adoption story gets stronger every quarter, is the kind of pattern that should make analysts more skeptical of the implied revenue-per-seat economics, not less.
There is a legitimate business reason Microsoft might prefer not to disclose Copilot revenue even if the number were strong: doing so would invite scrutiny of the gap between that revenue and the capital being deployed to support it, a comparison the company may not want investors making explicitly, quarter over quarter, in a single disclosed ratio. Whether that explanation or a less flattering one accounts for the continued omission is not resolvable from the outside. What is resolvable is that the pattern itself — three consecutive quarters of accelerating adoption without an accompanying revenue figure — is now well-established enough to treat as likely to continue rather than as an anomaly due to correct itself the following quarter.
How the Analyst Community Moved
Price target activity after the report was uniformly upward, though the range across firms remained wide enough to indicate genuine disagreement about how much of the rally is durable.
Morgan Stanley reiterated its Overweight rating with an unchanged $600 target — notable in that the firm’s pre-earnings thesis required exactly the kind of Azure acceleration the report delivered, and the firm evidently saw no reason to raise the number further on the strength of a single quarter. Wedbush raised its target to $625 with an Outperform rating. Citi raised its target to $600 from $575. Evercore ISI moved to $528, and Wolfe Research to $550 — both meaningfully below the Morgan Stanley and Wedbush figures, suggesting those firms remain more conservative about how much of the current AI-infrastructure investment converts to durable earnings growth even after a strong quarter.
That spread — roughly $528 to $625 across firms that all saw the same beat — is itself informative. A quarter this clearly positive should, in principle, compress the range of reasonable valuations across analysts who are all modeling the same disclosed numbers. That it didn’t fully compress suggests the unresolved questions — Copilot’s actual revenue contribution chief among them — are doing real work in analysts’ models, holding some price targets below where the headline beat alone might justify.
The Accounting Change Deserves a Closer Look
The extension of data-center asset useful life from 15 years to 25 years, and the reclassification of some data-center leases from finance to operating treatment, is the kind of disclosure that tends to get a sentence in coverage and then disappear from the conversation. It deserves more scrutiny than that, because it changes how investors should read every capex figure Microsoft reports going forward, not just this quarter’s.
Extending an asset’s useful life reduces the annual depreciation expense recognized against it, which increases reported earnings without any change in the underlying cash spent to build the asset. This is a standard, permitted accounting judgment — useful-life estimates are inherently uncertain, and companies routinely revise them as they gain operating experience with a given asset class — but the timing of this particular revision, arriving in the same quarter the market was bracing for a potentially alarming capex number, is worth noting even without alleging anything improper about the change itself. A 25-year useful life for a data center built primarily around today’s AI accelerator hardware is also a genuinely debatable assumption on its own terms: GPU-centric data-center infrastructure has historically been replaced or substantially upgraded on cycles far shorter than 25 years, given how quickly chip generations have turned over during the current AI buildout. If the actual economic life of the hardware inside these facilities is shorter than the 25-year accounting assumption, the useful-life extension understates the true ongoing cost of maintaining Microsoft’s AI infrastructure at a competitive standard, even though it is fully compliant with how such judgments are made and disclosed.
The lease reclassification has a similar effect through a different mechanism. Finance leases are capitalized on the balance sheet and their payments show up partly as capex and partly as interest expense; operating leases are typically expensed differently and can reduce the capex figure a company reports even when the underlying cash commitment is unchanged. Both changes point in the same direction: a lower headline capex number this quarter, achieved through presentation choices rather than a change in the pace of physical infrastructure being built. Investors reading Microsoft’s calendar-2026 capex figure as a clean signal of spending discipline should weigh it against the $50 billion-plus first-quarter fiscal 2027 figure, which reflects actual cash outlay in a more straightforward way and points toward continued, not moderating, investment intensity.
What Didn’t Change
The end of Microsoft’s exclusive arrangement with OpenAI, which removed a structural competitive moat earlier this year, was not directly addressed in this earnings report in any disclosure this publication could identify — and its absence from the conversation this quarter is itself notable. A strong Azure number and a large Copilot seat count are compatible with a world in which the OpenAI exclusivity’s end has not yet meaningfully affected Microsoft’s competitive position; they are also compatible with a world in which it is too early to see that effect in a single quarter’s aggregate numbers. Nothing in this report resolves which is true, and the question remains exactly where it was before earnings: a structural change whose competitive consequences will show up gradually, if at all, rather than in any single disclosure.
Xbox and the Arkane Lyon restructuring similarly went unaddressed on the earnings call in any way this publication found reported. The roughly 3,200 gaming-division layoffs and the divestiture of four studios — Double Fine and Compulsion Games spun out independently, Ninja Theory and Undead Labs sold to undisclosed buyers — remain a smaller, separate thread from the AI-infrastructure story that dominated this quarter’s numbers, and Arkane Lyon’s fate remains unresolved, still subject to the French Works Council consultation process required before Microsoft can finalize a sale, closure, or continuation. Given gaming’s minor weight in Microsoft’s overall revenue mix, its absence from the earnings narrative this quarter is unsurprising, but it is a reminder that Microsoft’s broader corporate story in 2026 has two tracks running simultaneously — aggressive AI-infrastructure expansion and contraction in businesses that don’t fit that narrative — and only one of those tracks made news this week.
Reading the Quarter Against the Structural-Decline Thesis
The structural-decline framing this narrative has applied to Microsoft since spring rested on a specific claim: that the company was caught between capital-intensive AI infrastructure commitments and a monetization timeline that kept extending, with Azure share loss and unquantified Copilot revenue as the two clearest symptoms. This quarter directly addressed the Azure symptom and left the Copilot symptom exactly where it was.
That is not a clean vindication of either the bull case or the bear case, and the market’s roughly 20 percent one-week rally should be read with that ambiguity in mind rather than as a verdict. Azure’s acceleration to 43 percent growth, on a $100 billion annualized base, with next-quarter guidance pointing higher still, is a genuinely strong result that undercuts the share-loss narrative more than it confirms it. The capex figure, once the accounting reclassification is accounted for, shows a company still spending aggressively — the $50 billion-plus in first-quarter fiscal 2027 capital expenditure is not the number of a company pulling back, whatever the headline calendar-year figure implies. And Copilot’s 30 million paid seats, without an accompanying revenue figure for a third consecutive quarter, leaves the single most important open question in this narrative exactly as open as it was in June.
The next checkpoint is Microsoft’s fiscal first-quarter fiscal 2027 report, expected in late October, which will be the first opportunity to see whether the $50 billion-plus quarterly capex pace sustains, whether Azure’s guided 45 percent growth materializes, and whether three consecutive quarters without a Copilot revenue disclosure becomes four. Until then, this report stands as genuine evidence against the sharpest version of the structural-decline thesis, without fully resolving the question that thesis was built around.
