微软的「泳池难题」：6250亿缓慢加注，375亿当季开闸，水位线是升是降？

Article by: DaiDai, MaiDian

You must have encountered a math problem like this when you were in elementary school: a swimming pool with a water pipe at the top desperately adding water, while the drain at the bottom is wide open letting water out. When will the pool be full or emptied?

Have you ever thought this problem was illogical? Interestingly, Microsoft’s latest Q2 FY2026 financial report presents a simple yet real capital game: on one side, $625 billion in RPO (Remaining Performance Obligations, locking future orders in a “safe”), and on the other side, $37.5 billion in CapEx (real cash pouring into data centers and GPUs).

However, from a calm perspective, while the $625 billion RPO shows terrifying potential for water inflow, it is limited by data center construction cycles and other “pipeline” constraints, causing delays in revenue recognition; meanwhile, the $37.5 billion CapEx for the current quarter is actual water being released—resulting in a very “counterintuitive” scene: double beats on performance, strong guidance, but the stock price after hours once dipped about 6%.

On the surface, Azure’s 39% growth, with RPO doubling to $625 billion year-over-year, looks very stable on the demand side; but what the market is truly worried about is that Microsoft is stacking capacity at an almost “arms race” pace—this quarter’s CapEx of $37.5 billion, up 66% YoY, and management emphasizing “demand exceeds supply.”

Ultimately, Wall Street’s reason for selling despite double beats is fundamentally due to anxiety about this “water level”—if the speed of inflow (revenue conversion) cannot short-term cover the outflow rate (depreciation and expenses), then no matter how big the pool, the valuation buoyancy will decrease.

1. Performance Overview: “Top Growth” vs “Top Investment” — Why is Wall Street Panicking?

A key premise is that Microsoft is completing its transformation from a “software giant” to an “AI infrastructure powerhouse.”

If we translate Nadella’s statement in the earnings call—“AI business scale has surpassed some of our largest franchise businesses”—into investor language, it’s straightforward: AI is no longer a cherry on top; it has become the new main engine of growth and expenditure.

Looking at core financial data, for the second quarter, revenue was $81.3 billion, up 17% YoY, beating analyst expectations of $80.31 billion; operating profit was $38.3 billion, up 21%, also exceeding expectations; net profit under GAAP was $38.5 billion, up 60%, and non-GAAP net profit was $30.9 billion, up 23% (with fixed exchange rates, YoY growth is 21%).

Earnings per share (EPS) under GAAP: $5.16, up 60%, beating the expected $3.92; non-GAAP EPS: $4.14, up 24% (or 21% at fixed FX). Capital expenditure (CapEx) hit $37.5 billion, up 66%, setting a record and surpassing analyst estimates of $36.2 billion.

Source: ECONOMY INSIGHTS

Surprisingly, Microsoft introduced a “new element” this time: Reuters mentioned that the company disclosed for the first time the core usage metrics of M365 Copilot. CEO Nadella said there are 15 million annual users of M365 Copilot, pushing the narrative that “AI is not just a concept, but a paid product.”

This disclosure, along with higher-than-expected infrastructure spending (including data centers) and weaker-than-expected gaming sales, caused Microsoft’s stock to drop about 6% after hours.

MSFT stock chart

This after-hours plunge seems more like the market posing a question: “Growth is strong, but at what cost?”

The market is now paying attention not just to revenue, but to “the price paid for revenue.” Management repeatedly emphasizes “it’s not that demand is lacking, but that supply is insufficient.” As a result, CapEx this quarter was directly ramped up to $37.5 billion, a 66% YoY increase, with hints of further increases next quarter.

A subtle point: Azure’s 39% growth isn’t bad, but in the “AI hype valuation,” it might only be considered “just enough.” Barron’s perspective is quite typical—beat is beat, Azure’s growth slowed slightly from 40% last quarter, and the company’s guidance for next quarter is cautious, which may cause traders who are already overly optimistic to start easing off. The market is scared by this money-burning pace, worried that before capacity is fully released, profit margins will be dragged down by depreciation, leading to a phase “severe mismatch” in ROI.

Data center expansion as a physical bottleneck constraining Azure’s growth

2. B-side Dominance, C-side Mediocrity — Every Step of Growth on “Expensive Ground”

Microsoft’s current structure is highly divided: B-side is extremely strong, C-side is mediocre. This isn’t accidental but a necessary resource tilt in the AI era—B-side opportunities are greater, but at a higher cost.

1. $625 billion RPO: Is it an “Insurance Vault” for performance, or a “Stress Test” for delivery?

Business RPO doubling to $625 billion signals that enterprise customers are voting with contracts: AI and cloud budgets are no longer “pilot projects,” but entering a longer-term binding cycle. On a business level, Microsoft’s model is shifting—from one-time software revenue to a more continuous, consumption-based income similar to “power, water, electricity”—the more load and deeper calls, the stronger the revenue stickiness.

But the significance of these “locked-in” orders is shifting from “growth certainty” to a different perspective: RPO indicates future revenue visibility, not current cash flow certainty. Longer contracts mean longer fulfillment chains: supply of compute power, delivery pace, customer onboarding speed, and backend costs (depreciation/energy/chips) will determine whether it’s a “moat” or a “burden.”

In other words, RPO as an anchor can prove demand, but it cannot replace a more pragmatic question: with so many orders, can Microsoft deliver capacity on time, turn contracts into confirmed revenue, and keep profit margins from being eroded?

2. Three “constraints” behind growth

(1) Capacity as the ceiling: demand exceeds supply is good news, but “delivery delays” are a hard cost

Satya Nadella, Microsoft’s Chairman and CEO, said: “We are still in the early stages of AI diffusion, but Microsoft has already built an AI business that exceeds some of our largest traditional businesses. We are continuously advancing frontier innovations across the entire AI tech stack, creating new value for customers and partners.”

Yet, management admits there are delivery delays—these “unable to keep up” short-term issues sound like bragging, but long-term they mean two things: first, revenue recognition is delayed; second, competitors have a window to fill the gap. Data center build cycles + GPU supply are typical physical constraints—no matter how strong demand is, it must pass the “power on, install, go live” stage.

This explains why the market links “strong demand” with “stock price decline”: when capacity becomes a bottleneck, growth logic shifts from “demand side” to “supply side,” and filling supply often requires higher capital expenditure and heavier depreciation burdens.

Nadella, CEO of Microsoft Image source: jason redmond/AFP/Getty Images

(2) Lack of hedge on C-side: B-side leads, C-side struggles as a buffer

This quarter’s structural divergence is sharper: B-side driven by Azure/AI, while C-side remains flat—Xbox content revenue -5%, Windows OEM +1%. This means that if B-side’s revenue realization slows (even just delays), the company can’t rely on C-side to “support,” and market sensitivity to a single main line will increase significantly.

Plainly: as Microsoft becomes more like an “AI cloud infrastructure company,” valuation volatility will resemble infrastructure stocks—focusing on utilization, delivery, capital expenditure return cycles, rather than just one quarter’s beat.

(3) Efficiency audit begins: it’s not that burning money is impossible, but the market wants to see “echoes”

This quarter’s CapEx +66%, but Azure’s growth is only slightly above guidance (mentioned as 2 points higher). This is the core of the “counterintuitive” after-hours move: the market isn’t denying growth, but is re-pricing the speed of ROI realization.

The implicit issues in the market are very real:

• How much more “375 billion” must be spent to sustain nearly 40% growth?
• Is this spending leading to “capacity release → revenue acceleration → profit recovery,” or “depreciation hits first → profits compress → delayed realization”?
• Under this framework, Microsoft’s valuation narrative is changing: spending for a few quarters is still seen as “racing ahead,” but if high-intensity investment extends over more quarters, it becomes harder to justify as a “light asset software company,” leaning more toward “heavy capital, long-recovery infrastructure” valuation—growth remains valuable, but discount rates will be higher.

3. Removing OpenAI’s “$7.6 billion paper wealth,” four “life-and-death” indicators

1. Excluding the $7.6 billion net profit noise: “accounting revaluation” or “operational realization”?

This quarter’s GAAP net profit surged, but the most obvious variable isn’t core business but OpenAI investments’ net gains. Microsoft’s own report states plainly—“Net gains from OpenAI investments increased net profit by about $7.6 billion, EPS by $1.02.”

The issue is, this looks more like an “accounting revaluation” rather than “better product sales.” Media explanations are straightforward: last year’s reorganization/re-capitalization of OpenAI triggered a one-time GAAP accounting gain (GeekWire reported about $10 billion, after tax about $7.6 billion). It inflates GAAP profit figures but does not equate to operational cash flow improvements of the same magnitude.

2. Cutting through the noise: four “verification lines” to reshape market expectations

(1) Azure growth: is it “supply constrained” or demand starting to cool?

Azure’s +39% YoY (FX +38%) this quarter is a typical “numbers are not bad, but expectations were higher.”

The key point: the market is giving Microsoft the benefit of the doubt with “supply constrained”—meaning sales are happening, just delivery is slow. Once you stop hearing management repeatedly emphasize “supply constraints / capacity bottlenecks,” a growth slowdown is more likely interpreted as “demand cooling.”

(2) $37.5 billion arms race: is it “buying out the future” or “overdrawing the profit statement”?

CapEx hit $37.5 billion (+66% YoY), with about 2/3 invested in chips/hardware, even exceeding Visible Alpha’s consensus of $34.31 billion.

This is the core of the “counterintuitive” after-hours decline: the market isn’t denying growth but asking—“You’re growing fast, but at what cost?”

Management said CapEx might slightly decrease next quarter, but more importantly, does capacity release lead to jumps in revenue/orders? Reuters also mentioned CFO hinted that memory chip costs will pressure cloud margins—meaning the market will treat “cost curves” as equally important variables. Burning money isn’t scary; what’s scary is burning for a long time with weak echoes.

(3) RPO reservoir: how tough is the $625 billion?

Business RPO reaching $625 billion, +110% YoY, is one of Microsoft’s most solid bottom-line indicators: it proves enterprise clients are willing to sign longer contracts, binding budgets to Microsoft.

Microsoft expects the converted value of future customer commitments to grow more than double YoY, mainly due to a new $250 billion deal with OpenAI. The company states its cloud backlog has doubled YoY to $625 billion, surpassing Oracle’s $523 billion announced in December. But about 45% of the remaining performance obligations are from OpenAI alone, highlighting Microsoft’s high dependence on this startup. OpenAI previously committed to a total AI investment of about $1.4 trillion but has not disclosed specific financing plans.

This time, the market’s interpretation of RPO is more “detailed”—since roughly 45% of the RPO increase is believed driven by OpenAI. This shifts the narrative from “strong lock-in” to another question: how much of your certainty comes from a single mega-client or partner?

(4) OpenAI monetization: don’t listen to valuation stories, focus on “cash in” signals

The $7.6 billion this quarter makes many think “OpenAI contributes huge profits.” But what really matters is whether it can continue to generate scalable commercial revenue and how its costs/share of profits impact Microsoft’s margins—that determines whether it’s a “divine teammate” or a “money pump.”

Valuation revaluation is a story; cash flow realization is the real deal.

Final Words

Returning to that simplest, cruelest math problem: when both the inflow and outflow are turned up, whether the water level rises depends on which side slows down first.

From this report, Microsoft has clearly proven one thing—there’s no problem with inflow, as the $625 billion RPO indicates enterprise clients have already locked in future AI budgets, and demand certainty is being front-loaded.

But on the other side, the drain remains wide open. The $37.5 billion quarterly CapEx won’t disappear immediately; it will continue to test the profit statement through depreciation, energy, and delivery costs. That’s why, after double beats, the market remains hesitant.

Microsoft still holds the most advantageous position: as long as in the next few quarters, inflow outpaces outflow, the change in water level will become self-evident, needing no further explanation.