Apple and Amazon Report Quarterly Results Thursday, a Day After Microsoft Beat Estimates and Meta's Guidance Fell Short

Two Reports, Two Reactions

Microsoft and Meta both grew sales by double digits Wednesday. One stock went up. One fell about 11% after hours^[3][14]. That split, more than either number alone, is the story heading into Thursday, when Apple and Amazon report their own results after the close^[8][9][10].

Microsoft said revenue rose 18% to $90.0 billion in the quarter ending in June, beating the roughly $87.6 billion Wall Street expected^[1][3]. Its Azure cloud business grew 43%, and Microsoft said customers have already signed contracts for $678 billion of cloud and software work it hasn't delivered yet, up 84% from a year earlier^[1]. Microsoft also left its spending plans unchanged. The stock rose.

Meta said revenue rose 28% to $60.80 billion, an even bigger jump than Microsoft's^[2]. But profit fell 14%, to $15.85 billion, and Meta guided next quarter's revenue to $61 billion to $64 billion, below the roughly $63.2 billion analysts wanted^[2][15]. Shares dropped about 11% after hours^[14].

Both companies grew fast. The market punished the one whose forecast came in soft. That's the pattern investors will be watching for again Thursday.

The Bill That Arrives Later

To understand why guidance matters more than the quarter that just happened, you need to understand capital expenditures, or "capex" — the money companies spend building things they'll use for years, in this case AI data centers and computer chips.

Four companies — Microsoft, Meta, Amazon, and Alphabet — are on pace to spend roughly $725 billion on capex in 2026. That's up from about $410 billion in 2025^[11]. Everyone agrees on that number. What they disagree on is what it means, and the disagreement runs through a piece of accounting most readers never think about: depreciation.

When a company builds a data center, the cost doesn't hit its profits all at once. Instead, accountants spread that cost out over the years the building or the chips inside it are useful, charging a slice of it against earnings every quarter. So money spent building in 2026 quietly lowers reported profits in 2027 and 2028, whether or not customers ever rent that capacity. Computer chips wear out and become obsolete fast, which shortens their useful life and makes that yearly charge even bigger.

That's why cash, not profit, is where the strain shows up first. Free cash flow is what's left after a company pays its operating costs and its capital spending — the money actually available for stock buybacks, dividends, or paying down debt. Meta's free cash flow fell to $784 million this quarter, down from $8.55 billion a year ago^[2][14]. Alphabet, which reported earlier in July, saw its free cash flow turn negative for the first time since it went public in 2004^[5][6]. Those are the numbers skeptics point to.

Same Contract, Read Two Ways

Company executives look at that $678 billion backlog Microsoft cited and see proof the spending is justified — real customers, real signed deals, not a guess^[1]. Meta CEO Mark Zuckerberg has made a version of this case directly to investors, arguing that building too little AI capacity would be the more expensive mistake than building too much^[4].

Their argument rests on a real constraint: compute capacity is a physical queue. Land, electricity hookups, transformers, and chips all take years to secure. A company that waits for demand to prove itself before committing to build risks losing that demand to a rival who broke ground first^[11]. From that seat, spending now isn't recklessness — it's securing a place in line.

Skeptical investors don't dispute that logic entirely. They just want to know when the payoff shows up, and they note that no company has yet published a figure showing how much revenue it earns per dollar of AI capex spent. Their touchstone is the dot-com-era fiber-optic buildout: demand for that capacity eventually arrived, but many of the companies that laid the cable didn't survive long enough to profit from it. Last week, the S&P 500 slipped just 0.6% — but the seven biggest tech stocks as a group fell 5.7%, a sign the selling is concentrated exactly where the AI spending is^[12].

Apple sits outside this fight almost entirely. It spends far less on AI hardware than its rivals and instead runs many AI features directly on the iPhone, while licensing outside technology where it needs more, including a deal to use Google's Gemini system inside Siri^[17]. That approach means Apple isn't racing to build data centers, but it also means it doesn't fully control the technology it's selling. Analysts expect Apple to report revenue near $108.8 billion to $110.1 billion Thursday^[9].

Amazon has taken the opposite bet, committing roughly $200 billion to 2026 capex and selling $25 billion in bonds to help pay for it^[16]. Analysts expect Amazon's revenue near $196.9 billion, with its cloud unit, AWS, around $40.5 billion^[10].

Behind the Concrete, a Supply Chain Placing Its Own Bets

There's a quieter group of stakeholders in all this: the chipmakers, construction crews, bondholders, and power utilities that actually build what the hyperscalers are paying for. From their vantage point, the "bubble" debate is a stock-market argument about who ultimately profits — not a question of whether the concrete is actually being poured. Chips, transformers, and turbines are booked years in advance. Amazon's willingness to raise $25 billion in bonds is, to this group, evidence that credit markets — whose whole job is pricing risk skeptically — are still willing to lend at scale against these projects^[16].

This group is also the most exposed if the mood shifts. If a hyperscaler quietly trims its spending plans next year, the first sign won't be a stock price — it'll be a canceled chip order or a stalled power project, long before it shows up in anyone's electricity bill.

How the Same Numbers Read Differently Around the World

Coverage of this split has broken along familiar lines. Market-friendly U.S. outlets, including CNBC's investing commentary and Forbes, framed the week as a fair test that strong companies can pass, treating Microsoft's backlog number as proof the AI spending is grounded in booked demand rather than speculation^[1][3][12]. Their framing locates the problem specifically in Meta's execution, not in the broader AI spending trade.

Fortune and Axios, writing from a more skeptical U.S. vantage, led instead with the scale of the spending and its cash cost, describing markets as being "in revolt" over AI spending — language that arguably overstates a week in which the S&P 500 moved less than 1%^[5][6][12]. The Fortune piece, notably, is actually a Bloomberg wire story republished under syndication, meaning its skeptical framing traces back to one financial newswire rather than two separate editorial voices.

Outlets outside the U.S., such as the Abu Dhabi-based The National, covered the same numbers with far less bubble language, treating the results as a signal about global cloud demand rather than a verdict on American tech excess — a framing that likely reflects audiences, including Gulf sovereign funds, with direct financial stakes in the infrastructure being built^[8].

What Thursday Still Has to Answer

Apple and Amazon hadn't reported by the time this was written^[8][9][10]. Their results will test the same question from opposite ends: whether a company that barely builds AI infrastructure and one that's betting hundreds of billions on it can both convince investors they've made the right call.

Expect each camp to reach for a different line in the same press releases — capex guidance for the skeptics, cloud growth and contract backlogs for the believers, and profit margins for anyone comparing Apple's approach of licensing AI against Amazon's approach of building it themselves. The one number nobody can yet supply, on either side, is how much revenue this spending is actually generating.