Microsoft has started collecting rent.

Microsoft released its earnings report last night, did you see it?

In a single quarter, revenue hit 90 billion U.S. dollars. Profits and cloud business all exceeded expectations. Let me start with the conclusion: Microsoft is transforming itself from a software company into a rent-collecting landlord. A landlord in the AI era.

This statement is supported by the figures listed one by one in the earnings report, look, the first piece of evidence: the speed of capital expenditure.

In fiscal year 2024, Microsoft's capital expenditure was 32 billion U.S. dollars. In fiscal year 2025, it rose to 64.6 billion. For this calendar year 2026, the company's guidance is 190 billion U.S. dollars.

About 25 billion U.S. dollars of that figure, according to CFO Amy Hood on the April 2026 earnings call, was driven up directly by price hikes of components such as memory.

Three years, six times. What does that mean?

Open any company's earnings report, if capital expenditure increases sixfold in three years, there are most likely only two scenarios: either the company has gone crazy, or it is betting on a far larger future than it has now.

That's not all. The second piece of evidence: the signed leases.

I checked a trace in the notes to the earnings report: as of the end of June 2025, Microsoft's lease commitments that are "signed under contract but not yet billed" amounted to 92.7 billion U.S. dollars.

By the end of December 2025, the figure reached 155.1 billion. By the end of March 2026, it hit 196.6 billion. In nine months, it more than doubled. The effective dates of these leases extend to fiscal year 2031.

These contracts are not on the balance sheet right now, they are pieces of paper that have been signed and are waiting to be converted into physical properties.

There is a third piece of evidence: the buildings that have already been built.

At the end of June 2025, the net fixed assets on Microsoft's balance sheet stood at 205 billion U.S. dollars; at the end of June 2026, the number reached 313.1 billion, up 53% year on year.

There is also a set of supporting figures on the same sheet: the total stock of servers, network equipment and software expanded from 132.8 billion to 190.9 billion in nine months. Accounts payable for equipment purchased but not yet paid rose from 6.9 billion to 22.6 billion U.S. dollars.

Spending money, signing contracts, building properties. The three actions point to the same direction.

Have you ever thought about this: why would a software company rewrite its asset structure into this form? The answer lies in the change of rules at the gambling table.

Microsoft's business logic is easy to understand, right? Everything revolves around the three words "asset-light"; to put it plainly, you develop a set of software, build various applications, copy tens of thousands of copies, the marginal cost is almost zero, and most of the money is spent on programmers.

Now its bet is the opposite: money is first converted into concrete, chips and electricity, and then converted back into revenue inch by inch. Data centers are its buildings, GPUs are the rooms in the buildings, and the parties that sign leases are not only companies all over the world, but also its own products.

How big is this bet? Let me do the math on two accounts:

The first one is the revenue structure. Personal computing business, including Windows, Surface and Xbox, which used to be Microsoft's main business, only accounted for 14% of total revenue in the fourth quarter of fiscal year 2026.

The second account is the scale: Azure's revenue in this fiscal year exceeded 100 billion U.S. dollars for the first time. The Microsoft that used to "do cloud business as a side project" is no longer the same company no matter you look at its assets or its revenue.

Satya Nadella has won a bet once.

When he took over in 2014, Microsoft had a market cap of 300 billion U.S. dollars and was regarded as a legacy of the PC era. He bet the whole company on cloud business, and in 2024, the market cap once reached 3 trillion U.S. dollars.

This time when betting on AI, the betting method is almost the same: first completely replace the entire asset structure, then wait for the revenue structure to catch up; last time, the market took five years to believe that those buildings could be rented out, this time, the buildings are six times more expensive, but the market only gave six months.

......

In these six months, the market has cast its vote once. The result is very clear.

If you don't believe it, look, within a year, Microsoft's stock price once peaked at 555 U.S. dollars. On the eve of this earnings report release, it only stayed around 390 U.S. dollars, down nearly 30% at most.

In April, it closed below the 200-week moving average for three consecutive weeks. What is the 200-week line? It is the psychological bottom line for long-term capital. The last time Microsoft fell below that line was back in 2013.

The falling price itself is not the source of fear, the bills are. Over the past six months, the market has issued six bills to Microsoft, each with a corresponding receipt.

The first one: there is no upper limit for spending.

In the earnings report released in January, Microsoft for the first time reported a single-quarter capital expenditure of 37.5 billion U.S. dollars, with 72.4 billion in the first half of the fiscal year, which was almost equal to the full-year figure of fiscal year 2025. When the market saw this number, it finally realized: this machine is swallowing money nonstop.

Right. Then let's look at the cash flow.

In the third quarter of fiscal year 2026, that is, the first three months of this year, after subtracting the cash spent on buying buildings and equipment from Microsoft's operating cash flow, the remaining free cash flow was only about 15.8 billion U.S. dollars.

At the same time, the gross margin of Microsoft's cloud business slipped from 68% to 66%, and the company's own guidance for the next quarter is 64%. 68%, 66%, 64%, it keeps going down all the way.

Then let's talk about Copilot.

In the same January earnings report, Microsoft for the first time disclosed the number of paid Copilot seats: 15 million. You may think 15 million is a large number, but hold on, it's not that large when you compare it to the 450 million commercial users.

By April, the number of seats exceeded 20 million; the paid conversion rate was 4.4% after calculation. What does that mean? For every 100 commercial users, only just over 4 of them paid for the service.

It's not over yet. Then three consecutive negative hits came one after another.

In February, The Wall Street Journal published an article saying that Copilot had functional issues and its market share was being taken away by Google Gemini. The stock price fell 3% that day.

In March, the same newspaper reported that the Copilot team was restructured, and the stock price fell another 2%. By June, investors even sued Microsoft in court, accusing it of misleading the market about Copilot's technical problems and actual adoption status.

Speaking of tenants, the largest one has also made moves.

In April, Microsoft and OpenAI re-signed their agreement; the revenue sharing arrangement was canceled, Microsoft lost the priority of being OpenAI's "preferred cloud", and OpenAI immediately signed a cloud contract with Amazon. The market's interpretation is very straightforward: the largest tenant is starting to look for other buildings for itself.

Your own tenants are leaving, while tenants of the competing buildings are accelerating their expansion.

Still in April, Google Cloud released its performance for the first quarter of the calendar year: growth of 63%. "Is Microsoft's cloud faster or Google's cloud faster" suddenly became a question that needed to be answered seriously.

Then, even Microsoft's credit began to be re-evaluated.

In July, Moody's calculated an industry-wide account: the six major cloud and AI platforms will spend about 785 billion U.S. dollars this year, and the figure will approach one trillion U.S. dollars in 2027. Its conclusion is just one sentence: "The final return on investment is unclear."

Wall Street began to measure Microsoft with a new ruler. Stifel cut its target price from 540 U.S. dollars to 392 U.S. dollars. The most widely spread sentence in that report is: the AI tail has started wagging the dog.

Alright. The six bills stacked together form one single question.

Erik Jackson, who has long been bullish on Microsoft, studied 84 past earnings calls. His conclusion is very clear: the pricing logic of this stock has changed from "cash machine" to "riding the AI tailwind".

The meaning of "cash machine" is that you can count the money with your eyes closed. The meaning of "riding the tailwind" is that the wind will stop one day.

It's true. The most torturous lesson came in January.

Azure grew 38% that quarter, exceeding the company's own guidance. That would be an excellent performance in any year, but the next day, the stock price fell 10%, and the market value evaporated by 357 billion U.S. dollars.

357 billion U.S. dollars, in one day, for only one reason: the whisper number spread privately on Wall Street was 39.4%. Meeting the official guidance is not enough, you have to beat the whisper number.

This is the predicament in the past six months: you can't prove where the upper limit of spending is, you can't prove that tenants won't leave, even proving strong growth is useless, there is always a higher number waiting for you to hit.

Then, what happened last night is obvious to all, the stock price rose about 8% in after-hours trading.

However, the Federal Reserve was holding a meeting the same night, and other large companies also released their earnings reports, so this 8% rise can't be all attributed to Microsoft; the 10% drop in January and this 8% rise are two sides of the same expectation mechanism.

......

What on earth made the same company, the same set of fears, be reweighed after this earnings report?

Everyone who runs a company knows it for sure: there must be evidence for growth, paid conversion, and all kinds of business boundaries.

Azure grew 43% this quarter, pay attention to this number: both the report caliber and the caliber excluding exchange rate impact are 43%, exactly the same.

The company's own guidance was 39% to 40%, and the most optimistic expectation on Wall Street from BNP Paribas was 41%. The actual figure is two percentage points higher than the most optimistic forecast.

Remember what happened in January? 38% beat the guidance, but was 1.4 percentage points lower than the whisper number, causing 357 billion U.S. dollars of market value to evaporate in one day. This time, the guidance, the consensus expectation, and the whisper number were all broken through together.

The structure of growth has also changed.

The Intelligent Cloud division recorded revenue of 39.3 billion U.S. dollars this quarter, exceeding the Productivity division where Office belongs for the first time, and the latter's revenue was 37.8 billion. Microsoft's largest business has officially shifted from "selling software" to "selling cloud services".

Then let's talk about paid conversion.

To judge whether a building is well rented, you have to look at the electricity meter. The number of paid Copilot seats is exactly the electricity meter of Microsoft's AI building.

In January, the number was 15 million. By July, it exceeded 30 million, with nearly 10 million new seats added in a single quarter. What was the sell-side forecast for new seats in this quarter? About 6 million, and the electricity meter also started to collect money directly.

That's not all. Starting from June 1, GitHub Copilot changed from seat-based subscription to usage-based billing, you pay for how much you use. AI has changed from a "bundled free option" to a "metered billing business".

Then let's look at the business boundary.

This quarter, Microsoft's commercial remaining performance obligations, that is, contracts that have been signed but not yet recognized as revenue, reached 678 billion U.S. dollars, up 84% year on year.

You may want to ask: is there any moisture in this number?

There is a split caliber. Last quarter, Microsoft disclosed for the first time that after excluding the part related to OpenAI, the growth rate of RPO was 26%. After removing the largest single tenant, the vacant rooms in the building are still being filled.

The narrative that "OpenAI is leaving" also needs an additional receipt.

In October 2025, OpenAI signed a new agreement with Microsoft, promising to purchase an additional 250 billion U.S. dollars of services on Azure. Leaving means it found other buildings, but signing the new agreement means the lease of this building has become even heavier.

Both things are true at the same time, and Microsoft's own AI investment map no longer only includes OpenAI. This quarter, it recognized 3.2 billion U.S. dollars of income from its investment in Anthropic.

Last but not least, cash flow. This quarter, operating cash flow reached 55.4 billion U.S. dollars, up 30% year on year. After subtracting the 35.8 billion U.S. dollars of cash spent on buying buildings and equipment, the remaining free cash flow was 19.6 billion U.S. dollars. It is thinner than a year ago, but it is positive in a year when capital expenditure doubled.

Why do we need to look at these three indicators together?

In the same earnings season, Alphabet also delivered results that exceeded expectations, but its stock price still plummeted. What the market punished in this round is "spending money until the cash flow is broken". Accelerated growth, positive cash flow, and visible contract visibility, if any of the three is missing, the story can't be told smoothly.

Just right, Microsoft put all three of these on the table at the same time this quarter.

......

If you ask me, Microsoft got all the good things this quarter. Then what about the bad things? There are bad things, and they are coming in the next four quarters.

The "mortgage" will keep rising. On the earnings call, the management set the tone for capital expenditure in the new fiscal year: maintain a high level, with the first fiscal quarter's single-quarter capital expenditure exceeding 50 billion U.S. dollars, including the impact of lease reclassification, subject to the official meeting minutes.

In the just-past fiscal year, the actual cash spent on buying equipment and building properties totaled 115.9 billion U.S. dollars. In the first quarter of the new fiscal year, the single-quarter spending will exceed 50 billion U.S. dollars. The spending benchmark is still moving upward.

Wall Street has also reacted. Deutsche Bank raised its capital expenditure estimate for the new fiscal year from 215 billion U.S. dollars to 238 billion U.S. dollars. UBS sees the figure reaching 261 billion U.S. dollars.

There is an even colder sentence in Deutsche Bank's report: At this spending pace, the free cash flow in the new fiscal year may be close to break-even, and this string will be stretched the tightest next year.

The "mortgage" is not only on the balance sheet, do you still remember that stack of "signed papers waiting to be converted into buildings"? By the end of March, the figure had reached 196.6 billion U.S. dollars. Note, this is only the number for March, the full account for the quarter ending June 30 will not be available until the annual report is released.

Also, in the neighboring camp, Google Cloud recorded a 63% growth rate in April. On July 22, Google's